<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lv-etal-2026-mitigating">
    <titleInfo>
        <title>Mitigating False Credit Propagation: Probabilistic Graphical Reward Aggregation for Rubric-Based Reinforcement Learning</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Can</namePart>
        <namePart type="family">Lv</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Mingju</namePart>
        <namePart type="family">Chen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Heng</namePart>
        <namePart type="family">Chang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Shiji</namePart>
        <namePart type="family">Zhou</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>Rubric-based rewards are increasingly used for open-ended language model post-training, but criterion-level scores are commonly aggregated as independent utilities. This flat scalarization ignores semantic dependencies among criteria, allowing downstream credit or penalties to be counted even when the conditions that license them are absent. We call this structural failure False Credit Propagation (FCP).\par We propose GEAR (Graphical Event Aggregation for Rubric rewards), a dependency-aware aggregation framework for rubric rewards. GEAR represents rubric criteria as latent events in a typed dependency graph, uses monotone retention factors to infer dependency-adjusted criterion marginals, and computes a normalized expected-utility reward.\par This preserves compatibility with standard rubric-based reinforcement learning while adding only linear-time aggregation over criteria and edges. Experiments on HealthBench, WritingBench, and PLawBench with two policy backbones show that GEAR consistently improves over flat aggregation and hard gating, both in controlled aggregation-only settings and as a plug-in module for existing rubric-guided RL pipelines.\par FCP diagnostics further show that GEAR substantially reduces invalid downstream utility while avoiding the over-suppression of deterministic gating.</abstract>
    <identifier type="citekey">lv-etal-2026-mitigating</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1933/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>43980</start>
            <end>44003</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Mitigating False Credit Propagation: Probabilistic Graphical Reward Aggregation for Rubric-Based Reinforcement Learning
%A Lv, Can
%A Chen, Mingju
%A Chang, Heng
%A Zhou, Shiji
%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 lv-etal-2026-mitigating
%X Rubric-based rewards are increasingly used for open-ended language model post-training, but criterion-level scores are commonly aggregated as independent utilities. This flat scalarization ignores semantic dependencies among criteria, allowing downstream credit or penalties to be counted even when the conditions that license them are absent. We call this structural failure False Credit Propagation (FCP).\par We propose GEAR (Graphical Event Aggregation for Rubric rewards), a dependency-aware aggregation framework for rubric rewards. GEAR represents rubric criteria as latent events in a typed dependency graph, uses monotone retention factors to infer dependency-adjusted criterion marginals, and computes a normalized expected-utility reward.\par This preserves compatibility with standard rubric-based reinforcement learning while adding only linear-time aggregation over criteria and edges. Experiments on HealthBench, WritingBench, and PLawBench with two policy backbones show that GEAR consistently improves over flat aggregation and hard gating, both in controlled aggregation-only settings and as a plug-in module for existing rubric-guided RL pipelines.\par FCP diagnostics further show that GEAR substantially reduces invalid downstream utility while avoiding the over-suppression of deterministic gating.
%U https://aclanthology.org/2026.emnlp-main.1933/
%P 43980-44003

Markdown (Informal)

[Mitigating False Credit Propagation: Probabilistic Graphical Reward Aggregation for Rubric-Based Reinforcement Learning](https://aclanthology.org/2026.emnlp-main.1933/) (Lv et al., EMNLP 2026)

• Mitigating False Credit Propagation: Probabilistic Graphical Reward Aggregation for Rubric-Based Reinforcement Learning (Lv et al., EMNLP 2026)

ACL

• Can Lv, Mingju Chen, Heng Chang, and Shiji Zhou. 2026. Mitigating False Credit Propagation: Probabilistic Graphical Reward Aggregation for Rubric-Based Reinforcement Learning. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 43980–44003, Budapest, Hungary. Association for Computational Linguistics.