@inproceedings{imrattanatrai-etal-2026-generate,
title = "Generate, Verify, and Refine: Event Timelines for Robust Video-Grounded Dialogue Systems",
author = "Imrattanatrai, Wiradee and
Asada, Masaki and
Hasegawa, Kimihiro and
Holm, Susan and
Fukuda, Ken and
Mitamura, Teruko",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 2: Short Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-short.5/",
pages = "49--68",
ISBN = "979-8-89176-440-8",
abstract = "Video-grounded dialogue (VGD) requires models to generate coherent responses by tracking and reasoning over temporally evolving events in video content. However, existing approaches offer limited ability to capture temporal dependencies, particularly in complex scenarios. In this paper, we present a framework that integrates structured event timelines with verification-aware reasoning to improve temporal consistency in VGD. We represent video content as structured states and temporally ordered actions, enabling models to both infer event timelines from video segments and leverage them to guide response generation. In addition, to address the inherent noise in predicted structure timelines, we introduce a Generate-Verify-Refine (GVR) mechanism that explicitly evaluates and revises unreliable event predictions, thereby reducing the propagation of upstream errors. Experiments on the VDAct benchmark show that our approach consistently improves dialogue quality across models, achieving gains of up to +13.0 LAVE and accuracy over unstructured grounding. Human evaluation further demonstrates improvements of 20.7-31.9 points in response correctness. Additional analysis shows that while better event predictions correlate with stronger dialogue performance, naive reliance on structured outputs can degrade results, highlighting the need for verification-aware reasoning. Our implementation of GVR on event timelines for VGD is available at https://github.com/aistairc/GVR-VGD."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="imrattanatrai-etal-2026-generate">
<titleInfo>
<title>Generate, Verify, and Refine: Event Timelines for Robust Video-Grounded Dialogue Systems</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wiradee</namePart>
<namePart type="family">Imrattanatrai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Masaki</namePart>
<namePart type="family">Asada</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kimihiro</namePart>
<namePart type="family">Hasegawa</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Susan</namePart>
<namePart type="family">Holm</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ken</namePart>
<namePart type="family">Fukuda</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Teruko</namePart>
<namePart type="family">Mitamura</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 2: Short Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-440-8</identifier>
</relatedItem>
<abstract>Video-grounded dialogue (VGD) requires models to generate coherent responses by tracking and reasoning over temporally evolving events in video content. However, existing approaches offer limited ability to capture temporal dependencies, particularly in complex scenarios. In this paper, we present a framework that integrates structured event timelines with verification-aware reasoning to improve temporal consistency in VGD. We represent video content as structured states and temporally ordered actions, enabling models to both infer event timelines from video segments and leverage them to guide response generation. In addition, to address the inherent noise in predicted structure timelines, we introduce a Generate-Verify-Refine (GVR) mechanism that explicitly evaluates and revises unreliable event predictions, thereby reducing the propagation of upstream errors. Experiments on the VDAct benchmark show that our approach consistently improves dialogue quality across models, achieving gains of up to +13.0 LAVE and accuracy over unstructured grounding. Human evaluation further demonstrates improvements of 20.7-31.9 points in response correctness. Additional analysis shows that while better event predictions correlate with stronger dialogue performance, naive reliance on structured outputs can degrade results, highlighting the need for verification-aware reasoning. Our implementation of GVR on event timelines for VGD is available at https://github.com/aistairc/GVR-VGD.</abstract>
<identifier type="citekey">imrattanatrai-etal-2026-generate</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-short.5/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>49</start>
<end>68</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Generate, Verify, and Refine: Event Timelines for Robust Video-Grounded Dialogue Systems
%A Imrattanatrai, Wiradee
%A Asada, Masaki
%A Hasegawa, Kimihiro
%A Holm, Susan
%A Fukuda, Ken
%A Mitamura, Teruko
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 2: Short Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-440-8
%F imrattanatrai-etal-2026-generate
%X Video-grounded dialogue (VGD) requires models to generate coherent responses by tracking and reasoning over temporally evolving events in video content. However, existing approaches offer limited ability to capture temporal dependencies, particularly in complex scenarios. In this paper, we present a framework that integrates structured event timelines with verification-aware reasoning to improve temporal consistency in VGD. We represent video content as structured states and temporally ordered actions, enabling models to both infer event timelines from video segments and leverage them to guide response generation. In addition, to address the inherent noise in predicted structure timelines, we introduce a Generate-Verify-Refine (GVR) mechanism that explicitly evaluates and revises unreliable event predictions, thereby reducing the propagation of upstream errors. Experiments on the VDAct benchmark show that our approach consistently improves dialogue quality across models, achieving gains of up to +13.0 LAVE and accuracy over unstructured grounding. Human evaluation further demonstrates improvements of 20.7-31.9 points in response correctness. Additional analysis shows that while better event predictions correlate with stronger dialogue performance, naive reliance on structured outputs can degrade results, highlighting the need for verification-aware reasoning. Our implementation of GVR on event timelines for VGD is available at https://github.com/aistairc/GVR-VGD.
%U https://aclanthology.org/2026.aacl-short.5/
%P 49-68
Markdown (Informal)
[Generate, Verify, and Refine: Event Timelines for Robust Video-Grounded Dialogue Systems](https://aclanthology.org/2026.aacl-short.5/) (Imrattanatrai et al., AACL-IJCNLP 2026)
ACL
- Wiradee Imrattanatrai, Masaki Asada, Kimihiro Hasegawa, Susan Holm, Ken Fukuda, and Teruko Mitamura. 2026. Generate, Verify, and Refine: Event Timelines for Robust Video-Grounded Dialogue Systems. In Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 2: Short Papers), pages 49–68, Hengqin, China. Association for Computational Linguistics.