@inproceedings{zhang-etal-2025-improve,
title = "Improve Vision Language Model Chain-of-thought Reasoning",
author = "Zhang, Ruohong and
Zhang, Bowen and
Li, Yanghao and
Zhang, Haotian and
Sun, Zhiqing and
Gan, Zhe and
Yang, Yinfei and
Pang, Ruoming and
Yang, Yiming",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.82/",
doi = "10.18653/v1/2025.acl-long.82",
pages = "1631--1662",
ISBN = "979-8-89176-251-0",
abstract = "Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. However, current training recipes often relying on datasets dominated by short annotations with minimal rationales. In this work, we show that training VLM on short answers leads to poor generalization on reasoning tasks that require more detailed explanations. To address this limitation, we propose a two-stage post-training strategy that extends the usage of short answer data for enhanced CoT reasoning. First, we augment short answers with CoT reasoning generated by GPT-4o, enhancing the VLM{'}s CoT capabilities through fine-tuning. Second, we leverage short answers as outcome rewards for reinforcement learning. Specifically, short answers are used as correctness indicators to construct positive (correct) and negative (incorrect) pairs from model-generated reasoning chains. These pairs are then used to calibrate the model{'}s reasoning via Direct Preference Optimization. Our experiments show significant improvements in CoT reasoning on benchmark datasets, along with enhanced generalization to direct answer prediction. This work provides a critical data resource for VLM CoT training and demonstrates the effectiveness of outcome rewards for multimodal models post-training."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhang-etal-2025-improve">
<titleInfo>
<title>Improve Vision Language Model Chain-of-thought Reasoning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Ruohong</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bowen</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanghao</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haotian</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhiqing</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhe</namePart>
<namePart type="family">Gan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yinfei</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruoming</namePart>
<namePart type="family">Pang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yiming</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. However, current training recipes often relying on datasets dominated by short annotations with minimal rationales. In this work, we show that training VLM on short answers leads to poor generalization on reasoning tasks that require more detailed explanations. To address this limitation, we propose a two-stage post-training strategy that extends the usage of short answer data for enhanced CoT reasoning. First, we augment short answers with CoT reasoning generated by GPT-4o, enhancing the VLM’s CoT capabilities through fine-tuning. Second, we leverage short answers as outcome rewards for reinforcement learning. Specifically, short answers are used as correctness indicators to construct positive (correct) and negative (incorrect) pairs from model-generated reasoning chains. These pairs are then used to calibrate the model’s reasoning via Direct Preference Optimization. Our experiments show significant improvements in CoT reasoning on benchmark datasets, along with enhanced generalization to direct answer prediction. This work provides a critical data resource for VLM CoT training and demonstrates the effectiveness of outcome rewards for multimodal models post-training.</abstract>
<identifier type="citekey">zhang-etal-2025-improve</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.82</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.82/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>1631</start>
<end>1662</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Improve Vision Language Model Chain-of-thought Reasoning
%A Zhang, Ruohong
%A Zhang, Bowen
%A Li, Yanghao
%A Zhang, Haotian
%A Sun, Zhiqing
%A Gan, Zhe
%A Yang, Yinfei
%A Pang, Ruoming
%A Yang, Yiming
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F zhang-etal-2025-improve
%X Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. However, current training recipes often relying on datasets dominated by short annotations with minimal rationales. In this work, we show that training VLM on short answers leads to poor generalization on reasoning tasks that require more detailed explanations. To address this limitation, we propose a two-stage post-training strategy that extends the usage of short answer data for enhanced CoT reasoning. First, we augment short answers with CoT reasoning generated by GPT-4o, enhancing the VLM’s CoT capabilities through fine-tuning. Second, we leverage short answers as outcome rewards for reinforcement learning. Specifically, short answers are used as correctness indicators to construct positive (correct) and negative (incorrect) pairs from model-generated reasoning chains. These pairs are then used to calibrate the model’s reasoning via Direct Preference Optimization. Our experiments show significant improvements in CoT reasoning on benchmark datasets, along with enhanced generalization to direct answer prediction. This work provides a critical data resource for VLM CoT training and demonstrates the effectiveness of outcome rewards for multimodal models post-training.
%R 10.18653/v1/2025.acl-long.82
%U https://aclanthology.org/2025.acl-long.82/
%U https://doi.org/10.18653/v1/2025.acl-long.82
%P 1631-1662
Markdown (Informal)
[Improve Vision Language Model Chain-of-thought Reasoning](https://aclanthology.org/2025.acl-long.82/) (Zhang et al., ACL 2025)
ACL
- Ruohong Zhang, Bowen Zhang, Yanghao Li, Haotian Zhang, Zhiqing Sun, Zhe Gan, Yinfei Yang, Ruoming Pang, and Yiming Yang. 2025. Improve Vision Language Model Chain-of-thought Reasoning. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1631–1662, Vienna, Austria. Association for Computational Linguistics.