@inproceedings{li-etal-2026-stanceagent,
title = "{S}tance{A}gent-{R}1: Agentic Reinforcement Learning for Zero-Shot Conversational Stance Detection",
author = "Li, Xing and
Ning, Jinzhong 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.1129/",
doi = "10.18653/2026.findings-emnlp.1129",
pages = "23895--23919",
ISBN = "979-8-89176-462-0",
abstract = "Conversational stance detection requires identifying the stance of a current utterance toward a target within its dialogue context. Existing methods often rely on sequential encoding, fixed knowledge-integration pipelines, or uniform reasoning strategies, which limits their ability to capture cross-turn dependencies and adapt to instance-specific needs. We propose StanceAgent-R1, an agentic reinforcement learning framework that adaptively routes each instance to evidence-path selection and/or knowledge elicitation. The model is first warm-started with rejection-sampled expert trajectories and then optimized with a progressive trajectory-faithful reward that encourages correct, well-formed, and route-consistent reasoning. Experiments on ZS-CSD, MT-CSD, and C-MTCSD show that StanceAgent-R1 achieves state-of-the-art performance, with further analyses confirming the effectiveness of its agentic design and reward strategy."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-stanceagent">
<titleInfo>
<title>StanceAgent-R1: Agentic Reinforcement Learning for Zero-Shot Conversational Stance Detection</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xing</namePart>
<namePart type="family">Li</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">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>Conversational stance detection requires identifying the stance of a current utterance toward a target within its dialogue context. Existing methods often rely on sequential encoding, fixed knowledge-integration pipelines, or uniform reasoning strategies, which limits their ability to capture cross-turn dependencies and adapt to instance-specific needs. We propose StanceAgent-R1, an agentic reinforcement learning framework that adaptively routes each instance to evidence-path selection and/or knowledge elicitation. The model is first warm-started with rejection-sampled expert trajectories and then optimized with a progressive trajectory-faithful reward that encourages correct, well-formed, and route-consistent reasoning. Experiments on ZS-CSD, MT-CSD, and C-MTCSD show that StanceAgent-R1 achieves state-of-the-art performance, with further analyses confirming the effectiveness of its agentic design and reward strategy.</abstract>
<identifier type="citekey">li-etal-2026-stanceagent</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1129</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1129/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>23895</start>
<end>23919</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T StanceAgent-R1: Agentic Reinforcement Learning for Zero-Shot Conversational Stance Detection
%A Li, Xing
%A Ning, Jinzhong
%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 li-etal-2026-stanceagent
%X Conversational stance detection requires identifying the stance of a current utterance toward a target within its dialogue context. Existing methods often rely on sequential encoding, fixed knowledge-integration pipelines, or uniform reasoning strategies, which limits their ability to capture cross-turn dependencies and adapt to instance-specific needs. We propose StanceAgent-R1, an agentic reinforcement learning framework that adaptively routes each instance to evidence-path selection and/or knowledge elicitation. The model is first warm-started with rejection-sampled expert trajectories and then optimized with a progressive trajectory-faithful reward that encourages correct, well-formed, and route-consistent reasoning. Experiments on ZS-CSD, MT-CSD, and C-MTCSD show that StanceAgent-R1 achieves state-of-the-art performance, with further analyses confirming the effectiveness of its agentic design and reward strategy.
%R 10.18653/2026.findings-emnlp.1129
%U https://aclanthology.org/2026.findings-emnlp.1129/
%U https://doi.org/10.18653/2026.findings-emnlp.1129
%P 23895-23919
Markdown (Informal)
[StanceAgent-R1: Agentic Reinforcement Learning for Zero-Shot Conversational Stance Detection](https://aclanthology.org/2026.findings-emnlp.1129/) (Li et al., Findings 2026)
ACL