EnigmaToM: Improve LLMs’ Theory-of-Mind Reasoning Capabilities with Neural Knowledge Base of Entity States
Published in arXiv preprint, 2025
This paper introduces EnigmaToM, a novel approach to improve large language models’ theory-of-mind reasoning capabilities through a neural knowledge base of entity states. The method enhances LLMs’ ability to understand and reason about mental states and beliefs.
The work addresses the challenge of developing more sophisticated reasoning capabilities in language models, particularly in understanding complex social and cognitive scenarios.
Recommended citation: H Xu, S Qi, J Li, Y Zhou, J Du, C Catmur, Y He. (2025). "EnigmaToM: Improve LLMs Theory-of-Mind Reasoning Capabilities with Neural Knowledge Base of Entity States." arXiv preprint arXiv:2503.03340.
Download Paper