@inproceedings{dong-etal-2026-omnihalluc,
title = "{O}mni{H}alluc-{L}: Counterfactual Benchmarking and Modality-Perturbation Calibration for Long-Form Omni Hallucination",
author = "Dong, Zixuan and
Tang, Jiafu and
Lei, Zhide and
Cao, Zhe and
Zhang, Zijie and
Wang, Yanghai and
Li, Shihao and
Wang, Xiaodong and
Peng, Baoyun and
Liu, Jiaheng",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.882/",
pages = "20361--20377",
ISBN = "979-8-89176-459-0",
abstract = "Long-video Omni assistants often fail not by inventing content, but by misbinding real evidence: they hear the right utterance and see the right event, yet attach it to the wrong speaker, moment, or modality. These almost-true errors evade standard video QA because local evidence remains valid, so item-level scoring can reward both a supported claim and its near-counterfactual. We introduce a counterfactual event-binding protocol that constructs paired supported/counterfactual claims from the same audio-visual event evidence and evaluates them by strict-pair accuracy. We instantiate it as OmniHalluc-L, to our knowledge the first benchmark for long-video Omni hallucination, with 3600 single-claim QA items from 638 long-form videos averaging 24.16 minutes and covering 256.87 hours. Under this protocol, open-weight Omni models remain weak at pair-level binding: Qwen2.5-Omni-7B reaches 32.06{\%} and Qwen3-Omni-Instruct reaches 41.55{\%}, versus 76.54{\%} for a closed-source reference. To narrow this gap, we propose MPRC (Modality-Perturbation Reliability Calibration), a frozen-backbone calibrator that combines native confidence with response shifts from selected audio-negative probes. MPRC raises these two models to 36.22{\%} and 51.09{\%}, respectively, and improves target-adapted MCQ accuracy on OmniVideoBench and WorldSense. The code is available at https://github.com/ZexDong/omnihallucL-mprc."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dong-etal-2026-omnihalluc">
<titleInfo>
<title>OmniHalluc-L: Counterfactual Benchmarking and Modality-Perturbation Calibration for Long-Form Omni Hallucination</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zixuan</namePart>
<namePart type="family">Dong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiafu</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhide</namePart>
<namePart type="family">Lei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhe</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zijie</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanghai</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shihao</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaodong</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Baoyun</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiaheng</namePart>
<namePart type="family">Liu</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>Long-video Omni assistants often fail not by inventing content, but by misbinding real evidence: they hear the right utterance and see the right event, yet attach it to the wrong speaker, moment, or modality. These almost-true errors evade standard video QA because local evidence remains valid, so item-level scoring can reward both a supported claim and its near-counterfactual. We introduce a counterfactual event-binding protocol that constructs paired supported/counterfactual claims from the same audio-visual event evidence and evaluates them by strict-pair accuracy. We instantiate it as OmniHalluc-L, to our knowledge the first benchmark for long-video Omni hallucination, with 3600 single-claim QA items from 638 long-form videos averaging 24.16 minutes and covering 256.87 hours. Under this protocol, open-weight Omni models remain weak at pair-level binding: Qwen2.5-Omni-7B reaches 32.06% and Qwen3-Omni-Instruct reaches 41.55%, versus 76.54% for a closed-source reference. To narrow this gap, we propose MPRC (Modality-Perturbation Reliability Calibration), a frozen-backbone calibrator that combines native confidence with response shifts from selected audio-negative probes. MPRC raises these two models to 36.22% and 51.09%, respectively, and improves target-adapted MCQ accuracy on OmniVideoBench and WorldSense. The code is available at https://github.com/ZexDong/omnihallucL-mprc.</abstract>
<identifier type="citekey">dong-etal-2026-omnihalluc</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.882/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>20361</start>
<end>20377</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T OmniHalluc-L: Counterfactual Benchmarking and Modality-Perturbation Calibration for Long-Form Omni Hallucination
%A Dong, Zixuan
%A Tang, Jiafu
%A Lei, Zhide
%A Cao, Zhe
%A Zhang, Zijie
%A Wang, Yanghai
%A Li, Shihao
%A Wang, Xiaodong
%A Peng, Baoyun
%A Liu, Jiaheng
%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 dong-etal-2026-omnihalluc
%X Long-video Omni assistants often fail not by inventing content, but by misbinding real evidence: they hear the right utterance and see the right event, yet attach it to the wrong speaker, moment, or modality. These almost-true errors evade standard video QA because local evidence remains valid, so item-level scoring can reward both a supported claim and its near-counterfactual. We introduce a counterfactual event-binding protocol that constructs paired supported/counterfactual claims from the same audio-visual event evidence and evaluates them by strict-pair accuracy. We instantiate it as OmniHalluc-L, to our knowledge the first benchmark for long-video Omni hallucination, with 3600 single-claim QA items from 638 long-form videos averaging 24.16 minutes and covering 256.87 hours. Under this protocol, open-weight Omni models remain weak at pair-level binding: Qwen2.5-Omni-7B reaches 32.06% and Qwen3-Omni-Instruct reaches 41.55%, versus 76.54% for a closed-source reference. To narrow this gap, we propose MPRC (Modality-Perturbation Reliability Calibration), a frozen-backbone calibrator that combines native confidence with response shifts from selected audio-negative probes. MPRC raises these two models to 36.22% and 51.09%, respectively, and improves target-adapted MCQ accuracy on OmniVideoBench and WorldSense. The code is available at https://github.com/ZexDong/omnihallucL-mprc.
%U https://aclanthology.org/2026.emnlp-main.882/
%P 20361-20377
Markdown (Informal)
[OmniHalluc-L: Counterfactual Benchmarking and Modality-Perturbation Calibration for Long-Form Omni Hallucination](https://aclanthology.org/2026.emnlp-main.882/) (Dong et al., EMNLP 2026)
ACL
- Zixuan Dong, Jiafu Tang, Zhide Lei, Zhe Cao, Zijie Zhang, Yanghai Wang, Shihao Li, Xiaodong Wang, Baoyun Peng, and Jiaheng Liu. 2026. OmniHalluc-L: Counterfactual Benchmarking and Modality-Perturbation Calibration for Long-Form Omni Hallucination. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 20361–20377, Budapest, Hungary. Association for Computational Linguistics.