@inproceedings{lei-etal-2026-diagnosing,
title = "Diagnosing and Selecting Synthetic Motion Tasks for Video {RLVR}",
author = "Lei, Yishu and
Liu, Aofan and
Zhu, Danxiang and
Ye, Canming and
Zhang, Dan and
Deng, Songhe and
Luo, Xianlong and
Feng, Shikun",
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.1176/",
pages = "24858--24871",
ISBN = "979-8-89176-462-0",
abstract = "Synthetic video data with verifiable rewards offers a scalable source of supervision for temporal motion understanding in video-language models, yet it remains unclear which synthetic tasks transfer reliably to real-video benchmarks.We investigate this question through a diagnostic framework that treats each synthetic task as a controlled intervention.Starting from a shared Qwen3-VL-2B checkpoint with frozen vision encoder, we independently train nine programmatic motion tasks with RLVR and evaluate each resulting model across six temporal-motion dimensions, such as direction, rotation, and velocity, yielding an interference matrix of cross-dimensional transfer. The matrix reveals substantial variation in transfer that surface-level label alignment fails to explain: a frame-to-frame velocity task yields a large gain on its target dimension, whereas a nominally aligned trajectory-shape task degrades the dimension it is intended to improve. We find that transfer is better explained by computational alignment, namely whether the training task induces the local, order-sensitive temporal computation that the downstream capability requires.Counterfactual tasks that share semantic targets but differ in computational demands support this account, and the patterns hold on InternVL3-2B. We further introduce a training-free answer-invariance test under frame shuffling that flags shortcut-prone tasks prior to RLVR training. Using the interference matrix to filter tasks yields a task mixture that consistently outperforms uniform mixing on TOMATO, with positive transfer trends on MVBench, TempCompass, and TemporalBench.Our framework thus reframes synthetic video generation as an empirically grounded task-selection problem."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lei-etal-2026-diagnosing">
<titleInfo>
<title>Diagnosing and Selecting Synthetic Motion Tasks for Video RLVR</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yishu</namePart>
<namePart type="family">Lei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aofan</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Danxiang</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Canming</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dan</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Songhe</namePart>
<namePart type="family">Deng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xianlong</namePart>
<namePart type="family">Luo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shikun</namePart>
<namePart type="family">Feng</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>Synthetic video data with verifiable rewards offers a scalable source of supervision for temporal motion understanding in video-language models, yet it remains unclear which synthetic tasks transfer reliably to real-video benchmarks.We investigate this question through a diagnostic framework that treats each synthetic task as a controlled intervention.Starting from a shared Qwen3-VL-2B checkpoint with frozen vision encoder, we independently train nine programmatic motion tasks with RLVR and evaluate each resulting model across six temporal-motion dimensions, such as direction, rotation, and velocity, yielding an interference matrix of cross-dimensional transfer. The matrix reveals substantial variation in transfer that surface-level label alignment fails to explain: a frame-to-frame velocity task yields a large gain on its target dimension, whereas a nominally aligned trajectory-shape task degrades the dimension it is intended to improve. We find that transfer is better explained by computational alignment, namely whether the training task induces the local, order-sensitive temporal computation that the downstream capability requires.Counterfactual tasks that share semantic targets but differ in computational demands support this account, and the patterns hold on InternVL3-2B. We further introduce a training-free answer-invariance test under frame shuffling that flags shortcut-prone tasks prior to RLVR training. Using the interference matrix to filter tasks yields a task mixture that consistently outperforms uniform mixing on TOMATO, with positive transfer trends on MVBench, TempCompass, and TemporalBench.Our framework thus reframes synthetic video generation as an empirically grounded task-selection problem.</abstract>
<identifier type="citekey">lei-etal-2026-diagnosing</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1176/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>24858</start>
<end>24871</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Diagnosing and Selecting Synthetic Motion Tasks for Video RLVR
%A Lei, Yishu
%A Liu, Aofan
%A Zhu, Danxiang
%A Ye, Canming
%A Zhang, Dan
%A Deng, Songhe
%A Luo, Xianlong
%A Feng, Shikun
%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 lei-etal-2026-diagnosing
%X Synthetic video data with verifiable rewards offers a scalable source of supervision for temporal motion understanding in video-language models, yet it remains unclear which synthetic tasks transfer reliably to real-video benchmarks.We investigate this question through a diagnostic framework that treats each synthetic task as a controlled intervention.Starting from a shared Qwen3-VL-2B checkpoint with frozen vision encoder, we independently train nine programmatic motion tasks with RLVR and evaluate each resulting model across six temporal-motion dimensions, such as direction, rotation, and velocity, yielding an interference matrix of cross-dimensional transfer. The matrix reveals substantial variation in transfer that surface-level label alignment fails to explain: a frame-to-frame velocity task yields a large gain on its target dimension, whereas a nominally aligned trajectory-shape task degrades the dimension it is intended to improve. We find that transfer is better explained by computational alignment, namely whether the training task induces the local, order-sensitive temporal computation that the downstream capability requires.Counterfactual tasks that share semantic targets but differ in computational demands support this account, and the patterns hold on InternVL3-2B. We further introduce a training-free answer-invariance test under frame shuffling that flags shortcut-prone tasks prior to RLVR training. Using the interference matrix to filter tasks yields a task mixture that consistently outperforms uniform mixing on TOMATO, with positive transfer trends on MVBench, TempCompass, and TemporalBench.Our framework thus reframes synthetic video generation as an empirically grounded task-selection problem.
%U https://aclanthology.org/2026.findings-emnlp.1176/
%P 24858-24871
Markdown (Informal)
[Diagnosing and Selecting Synthetic Motion Tasks for Video RLVR](https://aclanthology.org/2026.findings-emnlp.1176/) (Lei et al., Findings 2026)
ACL
- Yishu Lei, Aofan Liu, Danxiang Zhu, Canming Ye, Dan Zhang, Songhe Deng, Xianlong Luo, and Shikun Feng. 2026. Diagnosing and Selecting Synthetic Motion Tasks for Video RLVR. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 24858–24871, Budapest, Hungary. Association for Computational Linguistics.