PICTURE: Enhancing Theory-of-Mind in Large Language Models by Revealing, Not Hiding, Characters' Lack of Knowledge
2026-08-03 • Computation and Language
Computation and LanguageArtificial Intelligence
AI summaryⓘ
The authors study how large language models (LLMs) can better simulate Theory of Mind (ToM), which means understanding what others know or don't know. Instead of hiding events unknown to a character (event hiding), which limits performance, they let the model explain from a character's perspective without hiding information. They found that making the character's lack of knowledge explicit helps LLMs avoid talking about things the character doesn't know. Their new method called PICTURE improves performance on false-belief tasks, which test understanding others' perspectives.
Theory of MindPerspective-takingEvent hidingLarge Language ModelsChain-of-ThoughtFalse-belief tasksPrompting methodsNatural Language Processing
Authors
Eojin Jeon, SangKeun Lee
Abstract
Simulating human-like Theory of Mind (ToM) has been a longstanding problem in natural language processing (NLP). To address this, existing works introduce a reasoning step of event hiding (a.k.a. perspective-taking), where events unknown to a character are removed before question answering. However, resorting to event hiding for ToM reasoning presents a performance degradation issue due to the strict output format constraints involved in event hiding. To mitigate this issue, we propose generating perspective-taking outputs as free-form explanations without event hiding, but this poses a notable yet underexplored challenge: LLMs need to inhibit responses to events unknown to characters, because the absence of event hiding exposes LLMs to these events throughout reasoning. To address this challenge, we hypothesize and empirically verify that LLMs can achieve such inhibition if a character's lack of knowledge about events is made explicit during reasoning. Based on this finding, we introduce PICTURE, a new prompting method that enables LLMs to generate a character's lack of knowledge within free-form Chain-of-Thought (CoT). Experimental results show that PICTURE outperforms existing prompting methods by an average of 7.3% on false-belief tasks.