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

TL;DR

Constrained Chain-of-ToM guides LLMs to construct explicit reasoning chains by first prompting LLMs to infer related ToM dimensions and prompts LLMs to infer the queried ToM dimension based on the generated related ToM dimensions and corresponding causal relations.

Abstract

Theory-of-Mind (ToM) ability 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 face challenges including poor performance in complex ToM reasoning tasks and an inability to handle non-narrative contexts. 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 these limitations. Specifically, CCoToM guides LLMs to construct explicit reasoning chains by first prompting LLMs to infer related ToM dimensions (e.g., belief). Afterward, CCoToM prompts LLMs to infer the queried ToM dimension based on the generated related ToM dimensions and corresponding causal relations. Additionally, CCoToM adaptively imposes constraints on prompts to introduce inductive biases and improve consistency between ToM dimensions. Besides narratives, CCoToM can also handle non-narrative contexts like conversations. Extensive experiments show that CCoToM consistently outperforms previous state-of-the-art methods by large margins across all LLMs and datasets used. We also conduct in-depth analyses to gain deeper insights into CCoToM. We have made our code publicly available.

Figures (17)

• Figure 1: An example testing the ToM in LLMs. The example is from the Forward Belief task in the BigToM dataset bigtom. Our zero-shot prompting method called CCoToM improves LLMs' ToM reasoning capability.
• Figure 2: Overview of the model of causal relations between ToM dimensions.
• Figure 3: Overview of CCoToM for different types of ToM reasoning tasks. Definitions of the shown task types are introduced in Section \ref{['sec:method:overview']}. "Constraint set’’ refers to the collection of CCoToM's constraints.
• Figure 4: CCoToM's prompt template for prompting the underlying LLM to infer the belief based on the inferred desire and action for the BB task. The constraints shown in the figure are described in Section \ref{['sec:method:constraints']}. The "{agent}’’ would be replaced by the name of the corresponding agent in the given question (e.g., Hiro in the example of Figure \ref{['fig:tom_example']}). We explain how CCoToM identifies the name of the agent in Section \ref{['sec:method:chain-of-tom']}. The "{desire response}" and "{action response}" are the inferred desire and action respectively.
• Figure 5: Results of analyzing constraints.