@inproceedings{jia-etal-2026-prioritizing,
title = "Prioritizing the Best: Incentivizing Reliable Multimodal Reasoning by Rewarding Beyond Answer Correctness",
author = "Jia, Mengzhao and
Zhang, Zhihan and
Jiang, Meng",
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.1377/",
pages = "29279--29295",
ISBN = "979-8-89176-462-0",
abstract = "Reinforcement Learning with Verifiable Rewards (RLVR) improves multimodal reasoning in answer-verifiable settings by rewarding final answers that can be checked. Yet answer-correct trajectories may still rely on incomplete derivations, weak evidence, or statements that contradict their conclusions. This gap between answer correctness and reasoning validity, which we call reasoning-answer inconsistency, motivates trajectory supervision in multimodal RLVR. We compare two main approaches: reward models (RMs) and generative rewards (GRs). RMs are efficient and help early in training, but their gains weaken as the policy distribution shifts; GRs improve performance, but they may yield unstable rewards and incur high pointwise judging cost. We therefore propose Groupwise Ranking Reward, which ranks verifier-passed trajectories for the same prompt in one pass and redistributes reward accordingly. Groupwise comparison better separates stronger and weaker correct trajectories with lower judge overhead than pointwise GRs. Experiments show that RLVR aggravates reasoning-answer inconsistency, while trajectory supervision alleviates it. Groupwise Ranking Reward performs best overall, improving reliability-conditioned accuracy from 47.4{\%} to 54.7{\%} over RLVR."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jia-etal-2026-prioritizing">
<titleInfo>
<title>Prioritizing the Best: Incentivizing Reliable Multimodal Reasoning by Rewarding Beyond Answer Correctness</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mengzhao</namePart>
<namePart type="family">Jia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhihan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Meng</namePart>
<namePart type="family">Jiang</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>Reinforcement Learning with Verifiable Rewards (RLVR) improves multimodal reasoning in answer-verifiable settings by rewarding final answers that can be checked. Yet answer-correct trajectories may still rely on incomplete derivations, weak evidence, or statements that contradict their conclusions. This gap between answer correctness and reasoning validity, which we call reasoning-answer inconsistency, motivates trajectory supervision in multimodal RLVR. We compare two main approaches: reward models (RMs) and generative rewards (GRs). RMs are efficient and help early in training, but their gains weaken as the policy distribution shifts; GRs improve performance, but they may yield unstable rewards and incur high pointwise judging cost. We therefore propose Groupwise Ranking Reward, which ranks verifier-passed trajectories for the same prompt in one pass and redistributes reward accordingly. Groupwise comparison better separates stronger and weaker correct trajectories with lower judge overhead than pointwise GRs. Experiments show that RLVR aggravates reasoning-answer inconsistency, while trajectory supervision alleviates it. Groupwise Ranking Reward performs best overall, improving reliability-conditioned accuracy from 47.4% to 54.7% over RLVR.</abstract>
<identifier type="citekey">jia-etal-2026-prioritizing</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1377/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>29279</start>
<end>29295</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Prioritizing the Best: Incentivizing Reliable Multimodal Reasoning by Rewarding Beyond Answer Correctness
%A Jia, Mengzhao
%A Zhang, Zhihan
%A Jiang, Meng
%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 jia-etal-2026-prioritizing
%X Reinforcement Learning with Verifiable Rewards (RLVR) improves multimodal reasoning in answer-verifiable settings by rewarding final answers that can be checked. Yet answer-correct trajectories may still rely on incomplete derivations, weak evidence, or statements that contradict their conclusions. This gap between answer correctness and reasoning validity, which we call reasoning-answer inconsistency, motivates trajectory supervision in multimodal RLVR. We compare two main approaches: reward models (RMs) and generative rewards (GRs). RMs are efficient and help early in training, but their gains weaken as the policy distribution shifts; GRs improve performance, but they may yield unstable rewards and incur high pointwise judging cost. We therefore propose Groupwise Ranking Reward, which ranks verifier-passed trajectories for the same prompt in one pass and redistributes reward accordingly. Groupwise comparison better separates stronger and weaker correct trajectories with lower judge overhead than pointwise GRs. Experiments show that RLVR aggravates reasoning-answer inconsistency, while trajectory supervision alleviates it. Groupwise Ranking Reward performs best overall, improving reliability-conditioned accuracy from 47.4% to 54.7% over RLVR.
%U https://aclanthology.org/2026.findings-emnlp.1377/
%P 29279-29295
Markdown (Informal)
[Prioritizing the Best: Incentivizing Reliable Multimodal Reasoning by Rewarding Beyond Answer Correctness](https://aclanthology.org/2026.findings-emnlp.1377/) (Jia et al., Findings 2026)
ACL