@inproceedings{li-etal-2026-mulferl,
title = "{M}ul{F}e{RL}: Enhancing Reinforcement Learning with Verbal Feedback in a Multi-turn Loop",
author = "Li, Xuancheng and
Li, Haitao and
Zhou, Yujia and
Liu, Yiqun and
Zhang, Min and
Ai, Qingyao",
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.2116/",
pages = "48119--48153",
ISBN = "979-8-89176-459-0",
abstract = "Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning across domains, but outcome-only scalar rewards are often sparse and uninformative, especially for failed samples, where they indicate only failure without explaining why the reasoning breaks down. In this paper, we leverage richer verbal feedback to guide RLVR on failed samples and convert feedback-induced progress into trainable learning signals. We propose \textbf{MulFeRL} (\textbf{Mul}ti-turn \textbf{Fe}edback-guided \textbf{R}einforcement \textbf{L}earning), a multi-turn, event-triggered RLVR framework that combines \textit{progress induction} for feedback-guided regeneration of failed samples, \textit{progress credit assignment} for learning from verifier-confirmed progress, and \textit{structured feedback injection} for integrating feedback into the model{'}s reasoning process. Trained on sampled OpenR1-Math, MulFeRL outperforms supervised, self-distillation-based, and RLVR baselines in-domain and shows strong out-of-domain generalization. [We release our code at {\ensuremath{<}}https://anonymous.4open.science/r/MulFeRL{\ensuremath{>}}.]"
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-mulferl">
<titleInfo>
<title>MulFeRL: Enhancing Reinforcement Learning with Verbal Feedback in a Multi-turn Loop</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xuancheng</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haitao</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yujia</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yiqun</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Min</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qingyao</namePart>
<namePart type="family">Ai</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>Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning across domains, but outcome-only scalar rewards are often sparse and uninformative, especially for failed samples, where they indicate only failure without explaining why the reasoning breaks down. In this paper, we leverage richer verbal feedback to guide RLVR on failed samples and convert feedback-induced progress into trainable learning signals. We propose MulFeRL (Multi-turn Feedback-guided Reinforcement Learning), a multi-turn, event-triggered RLVR framework that combines progress induction for feedback-guided regeneration of failed samples, progress credit assignment for learning from verifier-confirmed progress, and structured feedback injection for integrating feedback into the model’s reasoning process. Trained on sampled OpenR1-Math, MulFeRL outperforms supervised, self-distillation-based, and RLVR baselines in-domain and shows strong out-of-domain generalization. [We release our code at \ensuremath<https://anonymous.4open.science/r/MulFeRL\ensuremath>.]</abstract>
<identifier type="citekey">li-etal-2026-mulferl</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2116/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>48119</start>
<end>48153</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T MulFeRL: Enhancing Reinforcement Learning with Verbal Feedback in a Multi-turn Loop
%A Li, Xuancheng
%A Li, Haitao
%A Zhou, Yujia
%A Liu, Yiqun
%A Zhang, Min
%A Ai, Qingyao
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%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 li-etal-2026-mulferl
%X Reinforcement Learning with Verifiable Rewards (RLVR) is widely used to improve reasoning across domains, but outcome-only scalar rewards are often sparse and uninformative, especially for failed samples, where they indicate only failure without explaining why the reasoning breaks down. In this paper, we leverage richer verbal feedback to guide RLVR on failed samples and convert feedback-induced progress into trainable learning signals. We propose MulFeRL (Multi-turn Feedback-guided Reinforcement Learning), a multi-turn, event-triggered RLVR framework that combines progress induction for feedback-guided regeneration of failed samples, progress credit assignment for learning from verifier-confirmed progress, and structured feedback injection for integrating feedback into the model’s reasoning process. Trained on sampled OpenR1-Math, MulFeRL outperforms supervised, self-distillation-based, and RLVR baselines in-domain and shows strong out-of-domain generalization. [We release our code at \ensuremath<https://anonymous.4open.science/r/MulFeRL\ensuremath>.]
%U https://aclanthology.org/2026.emnlp-main.2116/
%P 48119-48153
Markdown (Informal)
[MulFeRL: Enhancing Reinforcement Learning with Verbal Feedback in a Multi-turn Loop](https://aclanthology.org/2026.emnlp-main.2116/) (Li et al., EMNLP 2026)
ACL