@inproceedings{zhang-etal-2026-sovtr,
title = "{S}o{VTR}: Set-of-Vision-Text Rewards for Attribute-Grounded Emotion Recognition in {VLLM}s",
author = "Zhang, Qixuan and
Wang, Zhifeng and
Zhang, Donghao and
Zheng, Jiyang and
Niu, Wenjia and
Sankaranarayana, Ramesh and
Caldwell, Sabrina and
Gedeon, Tom and
Liu, Yang and
Qin, Zhenyue",
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.1807/",
pages = "38611--38629",
ISBN = "979-8-89176-462-0",
abstract = "Vision Large Language Models (VLLMs) show promise for multi-modal understanding; yet, their application to emotion recognition remains limited by insufficient spatial and contextual awareness and by supervised fine-tuning that uniformly penalizes all reasoning tokens{---}suppressing valid alternatives. We present Set-of-Vision-Text Rewards (SoVTR), a unified, reward-based framework that guides VLLMs with interpretable and verifiable visual attributes, focusing optimization on the emotion-discriminative reasoning steps that truly determine correctness. SoVTR decomposes supervision into three rewards: (i) Face Reward capturing action units, gaze, and micro-expressions; (ii) Body Reward encoding pose and gesture cues; and (iii) Context Reward modeling person{--}object relations and scene activities. By combining BERTScore attribute matching with GRPO policy optimization, SoVTR produces attribute-grounded explanation traces while preserving non-critical reasoning diversity. Across DFEW, FERV39k, and ExpW, SoVTR consistently improves both recognition accuracy and grounding quality, delivering substantial gains over strong VLLM baselines."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2026-sovtr">
<titleInfo>
<title>SoVTR: Set-of-Vision-Text Rewards for Attribute-Grounded Emotion Recognition in VLLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Qixuan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhifeng</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Donghao</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiyang</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenjia</namePart>
<namePart type="family">Niu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ramesh</namePart>
<namePart type="family">Sankaranarayana</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sabrina</namePart>
<namePart type="family">Caldwell</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tom</namePart>
<namePart type="family">Gedeon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhenyue</namePart>
<namePart type="family">Qin</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>Vision Large Language Models (VLLMs) show promise for multi-modal understanding; yet, their application to emotion recognition remains limited by insufficient spatial and contextual awareness and by supervised fine-tuning that uniformly penalizes all reasoning tokens—suppressing valid alternatives. We present Set-of-Vision-Text Rewards (SoVTR), a unified, reward-based framework that guides VLLMs with interpretable and verifiable visual attributes, focusing optimization on the emotion-discriminative reasoning steps that truly determine correctness. SoVTR decomposes supervision into three rewards: (i) Face Reward capturing action units, gaze, and micro-expressions; (ii) Body Reward encoding pose and gesture cues; and (iii) Context Reward modeling person–object relations and scene activities. By combining BERTScore attribute matching with GRPO policy optimization, SoVTR produces attribute-grounded explanation traces while preserving non-critical reasoning diversity. Across DFEW, FERV39k, and ExpW, SoVTR consistently improves both recognition accuracy and grounding quality, delivering substantial gains over strong VLLM baselines.</abstract>
<identifier type="citekey">zhang-etal-2026-sovtr</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1807/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>38611</start>
<end>38629</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SoVTR: Set-of-Vision-Text Rewards for Attribute-Grounded Emotion Recognition in VLLMs
%A Zhang, Qixuan
%A Wang, Zhifeng
%A Zhang, Donghao
%A Zheng, Jiyang
%A Niu, Wenjia
%A Sankaranarayana, Ramesh
%A Caldwell, Sabrina
%A Gedeon, Tom
%A Liu, Yang
%A Qin, Zhenyue
%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 zhang-etal-2026-sovtr
%X Vision Large Language Models (VLLMs) show promise for multi-modal understanding; yet, their application to emotion recognition remains limited by insufficient spatial and contextual awareness and by supervised fine-tuning that uniformly penalizes all reasoning tokens—suppressing valid alternatives. We present Set-of-Vision-Text Rewards (SoVTR), a unified, reward-based framework that guides VLLMs with interpretable and verifiable visual attributes, focusing optimization on the emotion-discriminative reasoning steps that truly determine correctness. SoVTR decomposes supervision into three rewards: (i) Face Reward capturing action units, gaze, and micro-expressions; (ii) Body Reward encoding pose and gesture cues; and (iii) Context Reward modeling person–object relations and scene activities. By combining BERTScore attribute matching with GRPO policy optimization, SoVTR produces attribute-grounded explanation traces while preserving non-critical reasoning diversity. Across DFEW, FERV39k, and ExpW, SoVTR consistently improves both recognition accuracy and grounding quality, delivering substantial gains over strong VLLM baselines.
%U https://aclanthology.org/2026.findings-emnlp.1807/
%P 38611-38629
Markdown (Informal)
[SoVTR: Set-of-Vision-Text Rewards for Attribute-Grounded Emotion Recognition in VLLMs](https://aclanthology.org/2026.findings-emnlp.1807/) (Zhang et al., Findings 2026)
ACL
- Qixuan Zhang, Zhifeng Wang, Donghao Zhang, Jiyang Zheng, Wenjia Niu, Ramesh Sankaranarayana, Sabrina Caldwell, Tom Gedeon, Yang Liu, and Zhenyue Qin. 2026. SoVTR: Set-of-Vision-Text Rewards for Attribute-Grounded Emotion Recognition in VLLMs. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 38611–38629, Budapest, Hungary. Association for Computational Linguistics.