@inproceedings{kouwenhoven-etal-2026-traces,
title = "Traces of Social Competence in Large Language Models",
author = "Kouwenhoven, Tom and
van der Meer, Michiel T. and
van Duijn, Max J.",
editor = "Bonial, Claire and
Berzak, Yevgeni",
booktitle = "Proceedings of the 30th Conference on Computational Natural Language Learning",
month = jul,
year = "2026",
address = "San Diego, California, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.conll-main.45/",
pages = "742--759",
ISBN = "979-8-89176-410-1",
abstract = "The False Belief Test (FBT) has been the main method for assessing Theory of Mind (ToM) and related socio-cognitive competencies. ForLarge Language Models (LLMs), the reliability and explanatory potential of this test have remained limited due to issues like data contamination, insufficient model details, and inconsistent controls. We address these issues by testing 17 open-weight models on a balanced set of 192 FBT variants (Trott et al., 2023) using Bayesian Logistic regression to identify how model size and post-training affect socio-cognitive competence. We find that scaling model size benefits performance, but not strictly. A cross-over effect reveals that explicating propositional attitudes (X *thinks*) fundamentally alters response patterns. Instruction tuning partially mitigates this effect, but further reasoning-oriented fine-tuning amplifies it. In a case study analysing social reasoning ability throughout OLMo 2 training, we show that this cross-over effect emerges during pre-training, suggesting that models acquire stereotypical response patterns tied to mental-state vocabulary that can outweigh other scenario semantics. Finally, vector steering allows us to isolate a *think* vector as the causal driver of observed FBT behaviour."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kouwenhoven-etal-2026-traces">
<titleInfo>
<title>Traces of Social Competence in Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tom</namePart>
<namePart type="family">Kouwenhoven</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Michiel</namePart>
<namePart type="given">T</namePart>
<namePart type="family">van der Meer</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Max</namePart>
<namePart type="given">J</namePart>
<namePart type="family">van Duijn</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>Proceedings of the 30th Conference on Computational Natural Language Learning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Claire</namePart>
<namePart type="family">Bonial</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yevgeni</namePart>
<namePart type="family">Berzak</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, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-410-1</identifier>
</relatedItem>
<abstract>The False Belief Test (FBT) has been the main method for assessing Theory of Mind (ToM) and related socio-cognitive competencies. ForLarge Language Models (LLMs), the reliability and explanatory potential of this test have remained limited due to issues like data contamination, insufficient model details, and inconsistent controls. We address these issues by testing 17 open-weight models on a balanced set of 192 FBT variants (Trott et al., 2023) using Bayesian Logistic regression to identify how model size and post-training affect socio-cognitive competence. We find that scaling model size benefits performance, but not strictly. A cross-over effect reveals that explicating propositional attitudes (X *thinks*) fundamentally alters response patterns. Instruction tuning partially mitigates this effect, but further reasoning-oriented fine-tuning amplifies it. In a case study analysing social reasoning ability throughout OLMo 2 training, we show that this cross-over effect emerges during pre-training, suggesting that models acquire stereotypical response patterns tied to mental-state vocabulary that can outweigh other scenario semantics. Finally, vector steering allows us to isolate a *think* vector as the causal driver of observed FBT behaviour.</abstract>
<identifier type="citekey">kouwenhoven-etal-2026-traces</identifier>
<location>
<url>https://aclanthology.org/2026.conll-main.45/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>742</start>
<end>759</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Traces of Social Competence in Large Language Models
%A Kouwenhoven, Tom
%A van der Meer, Michiel T.
%A van Duijn, Max J.
%Y Bonial, Claire
%Y Berzak, Yevgeni
%S Proceedings of the 30th Conference on Computational Natural Language Learning
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, USA
%@ 979-8-89176-410-1
%F kouwenhoven-etal-2026-traces
%X The False Belief Test (FBT) has been the main method for assessing Theory of Mind (ToM) and related socio-cognitive competencies. ForLarge Language Models (LLMs), the reliability and explanatory potential of this test have remained limited due to issues like data contamination, insufficient model details, and inconsistent controls. We address these issues by testing 17 open-weight models on a balanced set of 192 FBT variants (Trott et al., 2023) using Bayesian Logistic regression to identify how model size and post-training affect socio-cognitive competence. We find that scaling model size benefits performance, but not strictly. A cross-over effect reveals that explicating propositional attitudes (X *thinks*) fundamentally alters response patterns. Instruction tuning partially mitigates this effect, but further reasoning-oriented fine-tuning amplifies it. In a case study analysing social reasoning ability throughout OLMo 2 training, we show that this cross-over effect emerges during pre-training, suggesting that models acquire stereotypical response patterns tied to mental-state vocabulary that can outweigh other scenario semantics. Finally, vector steering allows us to isolate a *think* vector as the causal driver of observed FBT behaviour.
%U https://aclanthology.org/2026.conll-main.45/
%P 742-759
Markdown (Informal)
[Traces of Social Competence in Large Language Models](https://aclanthology.org/2026.conll-main.45/) (Kouwenhoven et al., CoNLL 2026)
ACL
- Tom Kouwenhoven, Michiel T. van der Meer, and Max J. van Duijn. 2026. Traces of Social Competence in Large Language Models. In Proceedings of the 30th Conference on Computational Natural Language Learning, pages 742–759, San Diego, California, USA. Association for Computational Linguistics.