@inproceedings{zhou-etal-2026-prism,
title = "{PRISM}: Probabilistic Reward Model with Inherent Structural Modeling",
author = "Zhou, Yuhang and
Cao, Yixin and
Ni, Yuchen and
Dou, Shihan and
Chen, Xutian and
Zhang, Ge and
Liu, Xiang and
Ye, Guangnan",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.acl-long.563/",
pages = "12345--12362",
ISBN = "979-8-89176-390-6",
abstract = "Standard evaluators, such as reward models, compress diverse human judgments into a single scalar, conflating valid Subjective Preference with Cognitive Uncertainty. This structural mismatch often leads to brittle alignment and reward hacking. To address this, we propose PRISM which reinterprets reward evaluation as a conditional distribution parameterized by a Mixture of Gaussians. PRISM structurally disentangles these factors: distinct Gaussian experts emerge to capture conflicting preference dimensions, while their variance estimates quantify uncertainty, acting as a dynamic reliability gate during optimization. We introduce a two-stage training strategy to learn these disentangled representations from scalable pairwise comparisons without requiring massive fine-grained annotations. Empirical results show that PRISM significantly outperforms scalar baselines in both accuracy and generalization. Furthermore, in downstream Reinforcement Learning, PRISM effectively mitigates reward hacking, yielding policies that are more robust and resilient to distribution shifts."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhou-etal-2026-prism">
<titleInfo>
<title>PRISM: Probabilistic Reward Model with Inherent Structural Modeling</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yuhang</namePart>
<namePart type="family">Zhou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yixin</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuchen</namePart>
<namePart type="family">Ni</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shihan</namePart>
<namePart type="family">Dou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xutian</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ge</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiang</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guangnan</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-390-6</identifier>
</relatedItem>
<abstract>Standard evaluators, such as reward models, compress diverse human judgments into a single scalar, conflating valid Subjective Preference with Cognitive Uncertainty. This structural mismatch often leads to brittle alignment and reward hacking. To address this, we propose PRISM which reinterprets reward evaluation as a conditional distribution parameterized by a Mixture of Gaussians. PRISM structurally disentangles these factors: distinct Gaussian experts emerge to capture conflicting preference dimensions, while their variance estimates quantify uncertainty, acting as a dynamic reliability gate during optimization. We introduce a two-stage training strategy to learn these disentangled representations from scalable pairwise comparisons without requiring massive fine-grained annotations. Empirical results show that PRISM significantly outperforms scalar baselines in both accuracy and generalization. Furthermore, in downstream Reinforcement Learning, PRISM effectively mitigates reward hacking, yielding policies that are more robust and resilient to distribution shifts.</abstract>
<identifier type="citekey">zhou-etal-2026-prism</identifier>
<location>
<url>https://aclanthology.org/2026.acl-long.563/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>12345</start>
<end>12362</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PRISM: Probabilistic Reward Model with Inherent Structural Modeling
%A Zhou, Yuhang
%A Cao, Yixin
%A Ni, Yuchen
%A Dou, Shihan
%A Chen, Xutian
%A Zhang, Ge
%A Liu, Xiang
%A Ye, Guangnan
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-390-6
%F zhou-etal-2026-prism
%X Standard evaluators, such as reward models, compress diverse human judgments into a single scalar, conflating valid Subjective Preference with Cognitive Uncertainty. This structural mismatch often leads to brittle alignment and reward hacking. To address this, we propose PRISM which reinterprets reward evaluation as a conditional distribution parameterized by a Mixture of Gaussians. PRISM structurally disentangles these factors: distinct Gaussian experts emerge to capture conflicting preference dimensions, while their variance estimates quantify uncertainty, acting as a dynamic reliability gate during optimization. We introduce a two-stage training strategy to learn these disentangled representations from scalable pairwise comparisons without requiring massive fine-grained annotations. Empirical results show that PRISM significantly outperforms scalar baselines in both accuracy and generalization. Furthermore, in downstream Reinforcement Learning, PRISM effectively mitigates reward hacking, yielding policies that are more robust and resilient to distribution shifts.
%U https://aclanthology.org/2026.acl-long.563/
%P 12345-12362
Markdown (Informal)
[PRISM: Probabilistic Reward Model with Inherent Structural Modeling](https://aclanthology.org/2026.acl-long.563/) (Zhou et al., ACL 2026)
ACL
- Yuhang Zhou, Yixin Cao, Yuchen Ni, Shihan Dou, Xutian Chen, Ge Zhang, Xiang Liu, and Guangnan Ye. 2026. PRISM: Probabilistic Reward Model with Inherent Structural Modeling. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 12345–12362, San Diego, California, United States. Association for Computational Linguistics.