@inproceedings{sarangi-salam-2026-small,
title = "Can Small {LLM}s Learn a Robust Theory of Mind via {RLVR}? Investigating Generalization through the False-Belief Task",
author = "Sarangi, Sneheel and
Salam, Hanan",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-acl.2061/",
doi = "10.18653/v1/2026.findings-acl.2061",
pages = "41433--41448",
ISBN = "979-8-89176-395-1",
abstract = "Recent advancements in large language models (LLMs) have demonstrated emergent capabilities in complex reasoning, largely spurred by rule-based Reinforcement Learning (RL) techniques applied during post-training. This has raised the question of whether similar methods can instill more nuanced, human-like social intelligence, such as a Theory of Mind (ToM), in LLMs. This paper investigates whether small- scale LLMs can acquire a robust and generalizable ToM capability through RL with verifiable rewards (RLVR). We conduct a systematic evaluation by training models on various combinations of prominent ToM benchmarks (HiToM, ExploreToM, FANToM) and testing for generalization on held-out benchmarks (e.g., Open- ToM). Our findings indicate that small LLMs struggle to develop a generic ToM capability. While performance on in-distribution tasks improves, this capability fails to transfer to unseen ToM tasks with different characteristics. Even observed out-of-distribution (OOD) performance improvements occur unpredictably across the training run, and don{'}t generalize across other OOD benchmarks. Furthermore, we conduct analysis to show that the learned behavior is likely a form of narrow overfitting rather than the acquisition of a true, abstract ToM capability."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sarangi-salam-2026-small">
<titleInfo>
<title>Can Small LLMs Learn a Robust Theory of Mind via RLVR? Investigating Generalization through the False-Belief Task</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sneheel</namePart>
<namePart type="family">Sarangi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hanan</namePart>
<namePart type="family">Salam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-395-1</identifier>
</relatedItem>
<abstract>Recent advancements in large language models (LLMs) have demonstrated emergent capabilities in complex reasoning, largely spurred by rule-based Reinforcement Learning (RL) techniques applied during post-training. This has raised the question of whether similar methods can instill more nuanced, human-like social intelligence, such as a Theory of Mind (ToM), in LLMs. This paper investigates whether small- scale LLMs can acquire a robust and generalizable ToM capability through RL with verifiable rewards (RLVR). We conduct a systematic evaluation by training models on various combinations of prominent ToM benchmarks (HiToM, ExploreToM, FANToM) and testing for generalization on held-out benchmarks (e.g., Open- ToM). Our findings indicate that small LLMs struggle to develop a generic ToM capability. While performance on in-distribution tasks improves, this capability fails to transfer to unseen ToM tasks with different characteristics. Even observed out-of-distribution (OOD) performance improvements occur unpredictably across the training run, and don’t generalize across other OOD benchmarks. Furthermore, we conduct analysis to show that the learned behavior is likely a form of narrow overfitting rather than the acquisition of a true, abstract ToM capability.</abstract>
<identifier type="citekey">sarangi-salam-2026-small</identifier>
<identifier type="doi">10.18653/v1/2026.findings-acl.2061</identifier>
<location>
<url>https://aclanthology.org/2026.findings-acl.2061/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>41433</start>
<end>41448</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Can Small LLMs Learn a Robust Theory of Mind via RLVR? Investigating Generalization through the False-Belief Task
%A Sarangi, Sneheel
%A Salam, Hanan
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Findings of the Association for Computational Linguistics: ACL 2026
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-395-1
%F sarangi-salam-2026-small
%X Recent advancements in large language models (LLMs) have demonstrated emergent capabilities in complex reasoning, largely spurred by rule-based Reinforcement Learning (RL) techniques applied during post-training. This has raised the question of whether similar methods can instill more nuanced, human-like social intelligence, such as a Theory of Mind (ToM), in LLMs. This paper investigates whether small- scale LLMs can acquire a robust and generalizable ToM capability through RL with verifiable rewards (RLVR). We conduct a systematic evaluation by training models on various combinations of prominent ToM benchmarks (HiToM, ExploreToM, FANToM) and testing for generalization on held-out benchmarks (e.g., Open- ToM). Our findings indicate that small LLMs struggle to develop a generic ToM capability. While performance on in-distribution tasks improves, this capability fails to transfer to unseen ToM tasks with different characteristics. Even observed out-of-distribution (OOD) performance improvements occur unpredictably across the training run, and don’t generalize across other OOD benchmarks. Furthermore, we conduct analysis to show that the learned behavior is likely a form of narrow overfitting rather than the acquisition of a true, abstract ToM capability.
%R 10.18653/v1/2026.findings-acl.2061
%U https://aclanthology.org/2026.findings-acl.2061/
%U https://doi.org/10.18653/v1/2026.findings-acl.2061
%P 41433-41448
Markdown (Informal)
[Can Small LLMs Learn a Robust Theory of Mind via RLVR? Investigating Generalization through the False-Belief Task](https://aclanthology.org/2026.findings-acl.2061/) (Sarangi & Salam, Findings 2026)
ACL