@inproceedings{shi-etal-2026-lsh,
title = "{LSH}-{RL}: Long-Short Horizon Reinforcement Learning for Persona-Level Role-Playing Agents",
author = "Shi, Wenlong and
Lian, Jianxun and
Qin, Haiming and
Zhang, Jiwei and
Zhang, Wei and
Zhou, Mingyang and
Liao, Hao",
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.1619/",
pages = "34453--34470",
ISBN = "979-8-89176-462-0",
abstract = "Role-playing agents (RPAs) must sustain persona fidelity, natural interaction, and scene progression across multi-turn trajectories. However, supervised fine-tuning, preference optimization, and coarse episode-level RL can over-reinforce safe stylized responses that appear in-character but are repetitive, low-information, or weakly grounded in the evolving scene. We propose LSH-RL, a Long-Short Horizon Reinforcement Learning framework for persona-level RPAs. LSH-RL trains a shared policy through dynamic multi-role self-play, combining short-horizon utterance/action rewards with long-horizon character-trajectory rewards. A gated trajectory-to-turn reward fusion mechanism prevents locally degenerate turns from receiving undeserved long-horizon credit, while a dialogue-action curriculum supports staged optimization. Experiments on PersonaArena and CharacterBox show that LSH-RL improves over SFT and DPO in both dialogue-only and full interaction settings. Further analyses show reduced repetition and stronger human preferences, suggesting a path toward more coherent and interaction-rich persona-level RPAs. The source code is available at https://github.com/Shiwenlong-source/LSH-RL."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shi-etal-2026-lsh">
<titleInfo>
<title>LSH-RL: Long-Short Horizon Reinforcement Learning for Persona-Level Role-Playing Agents</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wenlong</namePart>
<namePart type="family">Shi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jianxun</namePart>
<namePart type="family">Lian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haiming</namePart>
<namePart type="family">Qin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiwei</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wei</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mingyang</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hao</namePart>
<namePart type="family">Liao</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>Role-playing agents (RPAs) must sustain persona fidelity, natural interaction, and scene progression across multi-turn trajectories. However, supervised fine-tuning, preference optimization, and coarse episode-level RL can over-reinforce safe stylized responses that appear in-character but are repetitive, low-information, or weakly grounded in the evolving scene. We propose LSH-RL, a Long-Short Horizon Reinforcement Learning framework for persona-level RPAs. LSH-RL trains a shared policy through dynamic multi-role self-play, combining short-horizon utterance/action rewards with long-horizon character-trajectory rewards. A gated trajectory-to-turn reward fusion mechanism prevents locally degenerate turns from receiving undeserved long-horizon credit, while a dialogue-action curriculum supports staged optimization. Experiments on PersonaArena and CharacterBox show that LSH-RL improves over SFT and DPO in both dialogue-only and full interaction settings. Further analyses show reduced repetition and stronger human preferences, suggesting a path toward more coherent and interaction-rich persona-level RPAs. The source code is available at https://github.com/Shiwenlong-source/LSH-RL.</abstract>
<identifier type="citekey">shi-etal-2026-lsh</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1619/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>34453</start>
<end>34470</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T LSH-RL: Long-Short Horizon Reinforcement Learning for Persona-Level Role-Playing Agents
%A Shi, Wenlong
%A Lian, Jianxun
%A Qin, Haiming
%A Zhang, Jiwei
%A Zhang, Wei
%A Zhou, Mingyang
%A Liao, Hao
%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 shi-etal-2026-lsh
%X Role-playing agents (RPAs) must sustain persona fidelity, natural interaction, and scene progression across multi-turn trajectories. However, supervised fine-tuning, preference optimization, and coarse episode-level RL can over-reinforce safe stylized responses that appear in-character but are repetitive, low-information, or weakly grounded in the evolving scene. We propose LSH-RL, a Long-Short Horizon Reinforcement Learning framework for persona-level RPAs. LSH-RL trains a shared policy through dynamic multi-role self-play, combining short-horizon utterance/action rewards with long-horizon character-trajectory rewards. A gated trajectory-to-turn reward fusion mechanism prevents locally degenerate turns from receiving undeserved long-horizon credit, while a dialogue-action curriculum supports staged optimization. Experiments on PersonaArena and CharacterBox show that LSH-RL improves over SFT and DPO in both dialogue-only and full interaction settings. Further analyses show reduced repetition and stronger human preferences, suggesting a path toward more coherent and interaction-rich persona-level RPAs. The source code is available at https://github.com/Shiwenlong-source/LSH-RL.
%U https://aclanthology.org/2026.findings-emnlp.1619/
%P 34453-34470
Markdown (Informal)
[LSH-RL: Long-Short Horizon Reinforcement Learning for Persona-Level Role-Playing Agents](https://aclanthology.org/2026.findings-emnlp.1619/) (Shi et al., Findings 2026)
ACL