Constrained Reasoning Chains for Enhancing Theory-of-Mind in Large Language Models

Zizheng Lin, Chunkit Chan, Yangqiu Song, Xin Liu

Abstract: Theory-of-Mind (ToM) possessed by Large Language Models (LLMs) has been shown to be limited. Most existing methods for improving ToM in LLMs adopt zero-shot prompting, and they have poor performance in complex ToM reasoning tasks. We propose a zero-shot prompting method named Constrained Chain-of-ToM (CCoToM) that leverages domain knowledge and the causal relations between ToM dimensions to address the challenge. Specifically, CCoToM guides LLMs to construct explicit reasoning chains by first prompting LLMs to infer related ToM dimensions. Afterward, CCoToM prompts LLMs to infer the queried ToM dimension based on the related ToM dimensions and corresponding causal relations. Additionally, CCoToM adaptively imposes constraints on prompts to introduce inductive biases and improve consistency between ToM dimensions. Extensive experiments show that CCoToM consistently outperforms previous SOTA methods by large margins across all LLMs. We have made our code publicly available. (https://github.com/HKUST-KnowComp/Constrained-Chain-of-ToM)