@inproceedings{zeng-etal-2026-self,
title = "Self-Corrected Speech Reasoning: Diagnosing and Improving Condition Updating via {PACER}",
author = "Zeng, Zijie and
Shi, Liwen and
Yunji and
Wang, Bowen and
Li, Nanjie and
Guo, Xiaoyong and
Chen, Zeyu and
Sun, Xinzheng and
Xiong, Xuyang and
Wu, Haoyang and
Zhang, Sheng",
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.1313/",
pages = "29882--29911",
ISBN = "979-8-89176-459-0",
abstract = "Speech large language models (Speech LLMs) are increasingly expected to reason in everyday spoken interactions, where users often revise task conditions through self-correction. Unlike ordinary disfluencies, self-corrections alter the premise structure of a problem: obsolete conditions must be discarded, while final corrected conditions must guide reasoning. We formulate this setting as self-corrected speech reasoning and study it as a condition-update problem. We construct SCUR-Bench (Spoken Condition-Update Reasoning Benchmark), a controlled Chinese benchmark of 13,295 instances organized by reasoning difficulty and self-correction difficulty, and introduce a three-path diagnostic protocol to analyze speech-side recovery, condition use, and reasoning. Our analysis reveals a recurrent failure pattern in current Speech LLMs: corrected conditions may be perceived, transcribed, or even mentioned, yet still fail to serve as active premises in reasoning. We term this final-condition adoption failure. A controlled English parallel evaluation exhibits the same recovery{--}adoption gap, providing preliminary evidence that this failure is not unique to Chinese. To mitigate it, we propose PACER, a perception-aware condition-enhanced reinforcement framework based on GRPO. On Qwen2.5-Omni-7B, PACER reduces end-to-end error and condition-related errors by 39.8{\%} and 67.5{\%}, respectively, relative to the untrained baseline."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zeng-etal-2026-self">
<titleInfo>
<title>Self-Corrected Speech Reasoning: Diagnosing and Improving Condition Updating via PACER</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zijie</namePart>
<namePart type="family">Zeng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liwen</namePart>
<namePart type="family">Shi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name>
<namePart>Yunji</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bowen</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nanjie</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaoyong</namePart>
<namePart type="family">Guo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zeyu</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xinzheng</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuyang</namePart>
<namePart type="family">Xiong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haoyang</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sheng</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>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>Speech large language models (Speech LLMs) are increasingly expected to reason in everyday spoken interactions, where users often revise task conditions through self-correction. Unlike ordinary disfluencies, self-corrections alter the premise structure of a problem: obsolete conditions must be discarded, while final corrected conditions must guide reasoning. We formulate this setting as self-corrected speech reasoning and study it as a condition-update problem. We construct SCUR-Bench (Spoken Condition-Update Reasoning Benchmark), a controlled Chinese benchmark of 13,295 instances organized by reasoning difficulty and self-correction difficulty, and introduce a three-path diagnostic protocol to analyze speech-side recovery, condition use, and reasoning. Our analysis reveals a recurrent failure pattern in current Speech LLMs: corrected conditions may be perceived, transcribed, or even mentioned, yet still fail to serve as active premises in reasoning. We term this final-condition adoption failure. A controlled English parallel evaluation exhibits the same recovery–adoption gap, providing preliminary evidence that this failure is not unique to Chinese. To mitigate it, we propose PACER, a perception-aware condition-enhanced reinforcement framework based on GRPO. On Qwen2.5-Omni-7B, PACER reduces end-to-end error and condition-related errors by 39.8% and 67.5%, respectively, relative to the untrained baseline.</abstract>
<identifier type="citekey">zeng-etal-2026-self</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1313/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>29882</start>
<end>29911</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Self-Corrected Speech Reasoning: Diagnosing and Improving Condition Updating via PACER
%A Zeng, Zijie
%A Shi, Liwen
%A Wang, Bowen
%A Li, Nanjie
%A Guo, Xiaoyong
%A Chen, Zeyu
%A Sun, Xinzheng
%A Xiong, Xuyang
%A Wu, Haoyang
%A Zhang, Sheng
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%A Yunji
%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 zeng-etal-2026-self
%X Speech large language models (Speech LLMs) are increasingly expected to reason in everyday spoken interactions, where users often revise task conditions through self-correction. Unlike ordinary disfluencies, self-corrections alter the premise structure of a problem: obsolete conditions must be discarded, while final corrected conditions must guide reasoning. We formulate this setting as self-corrected speech reasoning and study it as a condition-update problem. We construct SCUR-Bench (Spoken Condition-Update Reasoning Benchmark), a controlled Chinese benchmark of 13,295 instances organized by reasoning difficulty and self-correction difficulty, and introduce a three-path diagnostic protocol to analyze speech-side recovery, condition use, and reasoning. Our analysis reveals a recurrent failure pattern in current Speech LLMs: corrected conditions may be perceived, transcribed, or even mentioned, yet still fail to serve as active premises in reasoning. We term this final-condition adoption failure. A controlled English parallel evaluation exhibits the same recovery–adoption gap, providing preliminary evidence that this failure is not unique to Chinese. To mitigate it, we propose PACER, a perception-aware condition-enhanced reinforcement framework based on GRPO. On Qwen2.5-Omni-7B, PACER reduces end-to-end error and condition-related errors by 39.8% and 67.5%, respectively, relative to the untrained baseline.
%U https://aclanthology.org/2026.emnlp-main.1313/
%P 29882-29911
Markdown (Informal)
[Self-Corrected Speech Reasoning: Diagnosing and Improving Condition Updating via PACER](https://aclanthology.org/2026.emnlp-main.1313/) (Zeng et al., EMNLP 2026)
ACL
- Zijie Zeng, Liwen Shi, Yunji, Bowen Wang, Nanjie Li, Xiaoyong Guo, Zeyu Chen, Xinzheng Sun, Xuyang Xiong, Haoyang Wu, and Sheng Zhang. 2026. Self-Corrected Speech Reasoning: Diagnosing and Improving Condition Updating via PACER. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 29882–29911, Budapest, Hungary. Association for Computational Linguistics.