@inproceedings{liu-lee-2026-vision,
title = "Can Vision-Language Models Solve the Shell Game?",
author = "Liu, Tiedong and
Lee, Wee Sun",
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.1987/",
pages = "45198--45221",
ISBN = "979-8-89176-459-0",
abstract = "Visual entity tracking is an innate cognitive ability in humans, yet it remains a critical bottleneck for Vision-Language Models (VLMs). This deficit is often obscured by visual shortcuts in existing video benchmarks. We introduce VET-Bench, a synthetic diagnostic testbed featuring visually identical objects that necessitate tracking exclusively through spatiotemporal continuity. Our experiments reveal that current state-of-the-art VLMs perform at or near chance level on VET-Bench. We provide a theoretical analysis drawing connections to the state-tracking problem, suggesting that fixed-depth transformer-based VLMs are fundamentally limited in tracking indistinguishable objects without intermediate computation due to expressivity constraints. To address this, we propose Spatiotemporal Grounded Chain-of-Thought (SGCoT): generating object trajectories as explicit intermediate states. Leveraging Molmo2{'}s object-tracking capability, we elicit SGCoT reasoning by fine-tuning on synthesized text-only data. Our method achieves state-of-the-art accuracy exceeding 90{\%} on VET-Bench, demonstrating that VLMs can reliably solve the video shell-game task end-to-end without external tools. Our code and data are available at https://vetbench.github.io."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liu-lee-2026-vision">
<titleInfo>
<title>Can Vision-Language Models Solve the Shell Game?</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tiedong</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wee</namePart>
<namePart type="given">Sun</namePart>
<namePart type="family">Lee</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>Visual entity tracking is an innate cognitive ability in humans, yet it remains a critical bottleneck for Vision-Language Models (VLMs). This deficit is often obscured by visual shortcuts in existing video benchmarks. We introduce VET-Bench, a synthetic diagnostic testbed featuring visually identical objects that necessitate tracking exclusively through spatiotemporal continuity. Our experiments reveal that current state-of-the-art VLMs perform at or near chance level on VET-Bench. We provide a theoretical analysis drawing connections to the state-tracking problem, suggesting that fixed-depth transformer-based VLMs are fundamentally limited in tracking indistinguishable objects without intermediate computation due to expressivity constraints. To address this, we propose Spatiotemporal Grounded Chain-of-Thought (SGCoT): generating object trajectories as explicit intermediate states. Leveraging Molmo2’s object-tracking capability, we elicit SGCoT reasoning by fine-tuning on synthesized text-only data. Our method achieves state-of-the-art accuracy exceeding 90% on VET-Bench, demonstrating that VLMs can reliably solve the video shell-game task end-to-end without external tools. Our code and data are available at https://vetbench.github.io.</abstract>
<identifier type="citekey">liu-lee-2026-vision</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1987/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>45198</start>
<end>45221</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Can Vision-Language Models Solve the Shell Game?
%A Liu, Tiedong
%A Lee, Wee Sun
%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 liu-lee-2026-vision
%X Visual entity tracking is an innate cognitive ability in humans, yet it remains a critical bottleneck for Vision-Language Models (VLMs). This deficit is often obscured by visual shortcuts in existing video benchmarks. We introduce VET-Bench, a synthetic diagnostic testbed featuring visually identical objects that necessitate tracking exclusively through spatiotemporal continuity. Our experiments reveal that current state-of-the-art VLMs perform at or near chance level on VET-Bench. We provide a theoretical analysis drawing connections to the state-tracking problem, suggesting that fixed-depth transformer-based VLMs are fundamentally limited in tracking indistinguishable objects without intermediate computation due to expressivity constraints. To address this, we propose Spatiotemporal Grounded Chain-of-Thought (SGCoT): generating object trajectories as explicit intermediate states. Leveraging Molmo2’s object-tracking capability, we elicit SGCoT reasoning by fine-tuning on synthesized text-only data. Our method achieves state-of-the-art accuracy exceeding 90% on VET-Bench, demonstrating that VLMs can reliably solve the video shell-game task end-to-end without external tools. Our code and data are available at https://vetbench.github.io.
%U https://aclanthology.org/2026.emnlp-main.1987/
%P 45198-45221
Markdown (Informal)
[Can Vision-Language Models Solve the Shell Game?](https://aclanthology.org/2026.emnlp-main.1987/) (Liu & Lee, EMNLP 2026)
ACL
- Tiedong Liu and Wee Sun Lee. 2026. Can Vision-Language Models Solve the Shell Game?. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 45198–45221, Budapest, Hungary. Association for Computational Linguistics.