@inproceedings{wang-etal-2026-right,
title = "Right Answers for the Right Reasons: Coherence-Enhanced Policy Optimization for Large Language Models",
author = "Wang, Xiangqi and
Huang, Yue and
Zhou, Yujun and
Luo, Xiaonan and
Guo, Kehan and
Ma, Yuchen and
Zhang, Xiangliang",
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.293/",
pages = "6729--6748",
ISBN = "979-8-89176-459-0",
abstract = "Large language models (LLMs) trained with reinforcement objectives often achieve superficially correct answers via shortcut strategies, pairing correct outputs with spurious or unfaithful reasoning and degrading under small input perturbations. We introduce Coherence-Enhanced Policy Optimization (CE-PO), a model agnostic reward fusion framework that augments policy optimization with a differentiable Jacobian coherence signal along the generation pathway from prompt ($X$) to rationale ($Y$) to answer ($Z$). CE-PO estimates internal gradient influence through Jacobian-based sensitivities, uses a reshuffling-based residualization procedure to reduce shortcut-driven reward hacking, and fuses the resulting coherence score with task-accuracy via an $L_p$ combiner that controls the strength of coherence regularization. The unified reward integrates with PPO/GRPO without architectural changes. Across multiple reasoning benchmarks, CE-PO improves accuracy over baselines by 5.49 on average (up to 9.58), while increasing the model{'}s faithful reasoning capability (12.14 {\%} on average across five backbones)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-right">
<titleInfo>
<title>Right Answers for the Right Reasons: Coherence-Enhanced Policy Optimization for Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xiangqi</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yujun</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaonan</namePart>
<namePart type="family">Luo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kehan</namePart>
<namePart type="family">Guo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuchen</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiangliang</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>Large language models (LLMs) trained with reinforcement objectives often achieve superficially correct answers via shortcut strategies, pairing correct outputs with spurious or unfaithful reasoning and degrading under small input perturbations. We introduce Coherence-Enhanced Policy Optimization (CE-PO), a model agnostic reward fusion framework that augments policy optimization with a differentiable Jacobian coherence signal along the generation pathway from prompt (X) to rationale (Y) to answer (Z). CE-PO estimates internal gradient influence through Jacobian-based sensitivities, uses a reshuffling-based residualization procedure to reduce shortcut-driven reward hacking, and fuses the resulting coherence score with task-accuracy via an L_p combiner that controls the strength of coherence regularization. The unified reward integrates with PPO/GRPO without architectural changes. Across multiple reasoning benchmarks, CE-PO improves accuracy over baselines by 5.49 on average (up to 9.58), while increasing the model’s faithful reasoning capability (12.14 % on average across five backbones).</abstract>
<identifier type="citekey">wang-etal-2026-right</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.293/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>6729</start>
<end>6748</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Right Answers for the Right Reasons: Coherence-Enhanced Policy Optimization for Large Language Models
%A Wang, Xiangqi
%A Huang, Yue
%A Zhou, Yujun
%A Luo, Xiaonan
%A Guo, Kehan
%A Ma, Yuchen
%A Zhang, Xiangliang
%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 wang-etal-2026-right
%X Large language models (LLMs) trained with reinforcement objectives often achieve superficially correct answers via shortcut strategies, pairing correct outputs with spurious or unfaithful reasoning and degrading under small input perturbations. We introduce Coherence-Enhanced Policy Optimization (CE-PO), a model agnostic reward fusion framework that augments policy optimization with a differentiable Jacobian coherence signal along the generation pathway from prompt (X) to rationale (Y) to answer (Z). CE-PO estimates internal gradient influence through Jacobian-based sensitivities, uses a reshuffling-based residualization procedure to reduce shortcut-driven reward hacking, and fuses the resulting coherence score with task-accuracy via an L_p combiner that controls the strength of coherence regularization. The unified reward integrates with PPO/GRPO without architectural changes. Across multiple reasoning benchmarks, CE-PO improves accuracy over baselines by 5.49 on average (up to 9.58), while increasing the model’s faithful reasoning capability (12.14 % on average across five backbones).
%U https://aclanthology.org/2026.emnlp-main.293/
%P 6729-6748
Markdown (Informal)
[Right Answers for the Right Reasons: Coherence-Enhanced Policy Optimization for Large Language Models](https://aclanthology.org/2026.emnlp-main.293/) (Wang et al., EMNLP 2026)
ACL