@inproceedings{sanli-etal-2026-physvidbench,
title = "{P}hys{V}id{B}ench: Language-Grounded Evaluation of Physical Commonsense in Text-to-Video Models",
author = "Sanli, Enes and
Tezcan, Baris Sarper and
Erdem, Erkut and
Erdem, Aykut",
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.2126/",
doi = "10.18653/2026.findings-emnlp.2126",
pages = "45311--45359",
ISBN = "979-8-89176-462-0",
abstract = "Text-to-video (T2V) models now produce striking visuals, yet they routinely violate everyday physics; objects float, tools are misused, and causal sequences break down. Existing benchmarks mostly probe isolated physical laws and rely on vision-language models to score videos directly, which entangles perception and reasoning in one judgment and correlates poorly with humans. We introduce PhysVidBench, a benchmark of 383 base prompts, expanded to 766 prompts with enriched variants and 4,123 manually reviewed QA items, together with a language-grounded evaluation framework that takes a different route: instead of asking a VLM ``Is this video physically correct?'', we caption the video, then ask a language model to answer prompt-derived yes/no physics questions using only the captions. This split between seeing and reasoning makes each score traceable to the captions that justify it and aligns more closely with human judgment than direct VLM scoring (Pearson r=0.45{--}0.69, compared with 0.21{--}0.47 for the strongest direct VLM evaluator). To test the framework, we carefully curate a set of human-validated prompts spanning seven physical dimensions, with a focus on tool use and affordances {--} areas largely absent from prior physics-focused benchmarks. Across 12 open and proprietary T2V systems, the best model reaches only 36.2{\%} average accuracy, and no model consistently handles everyday physical reasoning. The same pipeline can also guide iterative error-guided prompt refinement, improving CogVideoX-2B from 21.6 to 32.7 and CogVideoX-5B from 17.8 to 29.7 without retraining."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sanli-etal-2026-physvidbench">
<titleInfo>
<title>PhysVidBench: Language-Grounded Evaluation of Physical Commonsense in Text-to-Video Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Enes</namePart>
<namePart type="family">Sanli</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Baris</namePart>
<namePart type="given">Sarper</namePart>
<namePart type="family">Tezcan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Erkut</namePart>
<namePart type="family">Erdem</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aykut</namePart>
<namePart type="family">Erdem</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>Text-to-video (T2V) models now produce striking visuals, yet they routinely violate everyday physics; objects float, tools are misused, and causal sequences break down. Existing benchmarks mostly probe isolated physical laws and rely on vision-language models to score videos directly, which entangles perception and reasoning in one judgment and correlates poorly with humans. We introduce PhysVidBench, a benchmark of 383 base prompts, expanded to 766 prompts with enriched variants and 4,123 manually reviewed QA items, together with a language-grounded evaluation framework that takes a different route: instead of asking a VLM “Is this video physically correct?”, we caption the video, then ask a language model to answer prompt-derived yes/no physics questions using only the captions. This split between seeing and reasoning makes each score traceable to the captions that justify it and aligns more closely with human judgment than direct VLM scoring (Pearson r=0.45–0.69, compared with 0.21–0.47 for the strongest direct VLM evaluator). To test the framework, we carefully curate a set of human-validated prompts spanning seven physical dimensions, with a focus on tool use and affordances – areas largely absent from prior physics-focused benchmarks. Across 12 open and proprietary T2V systems, the best model reaches only 36.2% average accuracy, and no model consistently handles everyday physical reasoning. The same pipeline can also guide iterative error-guided prompt refinement, improving CogVideoX-2B from 21.6 to 32.7 and CogVideoX-5B from 17.8 to 29.7 without retraining.</abstract>
<identifier type="citekey">sanli-etal-2026-physvidbench</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.2126</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.2126/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>45311</start>
<end>45359</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T PhysVidBench: Language-Grounded Evaluation of Physical Commonsense in Text-to-Video Models
%A Sanli, Enes
%A Tezcan, Baris Sarper
%A Erdem, Erkut
%A Erdem, Aykut
%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 sanli-etal-2026-physvidbench
%X Text-to-video (T2V) models now produce striking visuals, yet they routinely violate everyday physics; objects float, tools are misused, and causal sequences break down. Existing benchmarks mostly probe isolated physical laws and rely on vision-language models to score videos directly, which entangles perception and reasoning in one judgment and correlates poorly with humans. We introduce PhysVidBench, a benchmark of 383 base prompts, expanded to 766 prompts with enriched variants and 4,123 manually reviewed QA items, together with a language-grounded evaluation framework that takes a different route: instead of asking a VLM “Is this video physically correct?”, we caption the video, then ask a language model to answer prompt-derived yes/no physics questions using only the captions. This split between seeing and reasoning makes each score traceable to the captions that justify it and aligns more closely with human judgment than direct VLM scoring (Pearson r=0.45–0.69, compared with 0.21–0.47 for the strongest direct VLM evaluator). To test the framework, we carefully curate a set of human-validated prompts spanning seven physical dimensions, with a focus on tool use and affordances – areas largely absent from prior physics-focused benchmarks. Across 12 open and proprietary T2V systems, the best model reaches only 36.2% average accuracy, and no model consistently handles everyday physical reasoning. The same pipeline can also guide iterative error-guided prompt refinement, improving CogVideoX-2B from 21.6 to 32.7 and CogVideoX-5B from 17.8 to 29.7 without retraining.
%R 10.18653/2026.findings-emnlp.2126
%U https://aclanthology.org/2026.findings-emnlp.2126/
%U https://doi.org/10.18653/2026.findings-emnlp.2126
%P 45311-45359
Markdown (Informal)
[PhysVidBench: Language-Grounded Evaluation of Physical Commonsense in Text-to-Video Models](https://aclanthology.org/2026.findings-emnlp.2126/) (Sanli et al., Findings 2026)
ACL