Mind the Perspective: Let's Reason Recursively for Theory of Mind

2026-06-10 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cognitive AI · Depth: Expert, quick

Summary

RecToM is an inference-time framework designed to enhance Large Language Models' (LLMs) Theory of Mind (ToM) reasoning capabilities, specifically addressing the challenge of inferring agents' beliefs from partial observations. Unlike existing prompting methods that lack explicit nested belief modeling, RecToM employs recursive perspective construction. This framework builds each character's perspective from the preceding one in a specified chain, effectively simplifying higher-order belief questions into actual-world questions within the final constructed perspective. A KD45 analysis confirms RecToM's perspective construction establishes a robust belief modality. Evaluated on ToM benchmarks like Hi-ToM, Big-ToM, and FanToM, RecToM consistently surpasses recent advanced approaches across various LLM backbones, achieving leading performance. Notably, it attained 100% accuracy on the higher-order Hi-ToM benchmark when paired with GPT-5.4 and Qwen3.5.

Key takeaway

For NLP Engineers and AI Scientists focused on enhancing LLM Theory of Mind capabilities, RecToM provides a highly effective inference-time framework. You should consider integrating its recursive perspective construction to explicitly model nested beliefs, especially for higher-order reasoning challenges. This approach has demonstrated 100% accuracy on complex benchmarks like Hi-ToM with advanced models, suggesting a significant improvement over existing prompting methods for robust belief inference.

Key insights

RecToM improves LLM Theory of Mind by recursively constructing nested perspectives.

Principles

• ToM requires modeling nested beliefs.
• Recursive perspective construction simplifies higher-order ToM.
• KD45 analysis validates belief modality.

Method

RecToM constructs character perspectives recursively from preceding ones along a question-specified chain, reducing higher-order belief questions to actual-world questions within the final perspective.

In practice

• Apply recursive perspective construction for complex ToM tasks.
• Use RecToM with LLMs for improved belief inference.
• Achieve 100% ToM accuracy on Hi-ToM with GPT-5.4/Qwen3.5.

Topics

• Theory of Mind
• Large Language Models
• Recursive Reasoning
• Nested Beliefs
• Perspective Construction
• Hi-ToM Benchmark

Best for: AI Engineer, Research Scientist, AI Scientist, NLP Engineer, Machine Learning Engineer

Related on AIssential

• Recursive Think-Answer Process to Stop "Thinking" Earlier — Integrating confidence-aware self-correction and RL-aware distillation improves LLM reasoning and efficiency.
• Do Large Language Models Possess a Theory of Mind? A Comparative Evaluation Using the Strange Stories Paradigm — GPT-4o demonstrates human-comparable Theory of Mind capabilities in text-based evaluations, outperforming smaller LLMs.
• OSCToM: RL-Guided Adversarial Generation for High-Order Theory of Mind — OSCToM uses RL-guided adversarial generation to create complex ToM benchmarks, significantly improving LLM performance on recursive belief conflicts.
• Think Thrice Before You Speak: Dual knowledge-enhanced Theory-of-Mind Reasoning for Persuasive Agents — Enhancing LLM persuasive dialogue requires explicitly modeling mental state dependencies using dual knowledge for robust Theory of Mind.
• LLM Reasoning Is Latent, Not the Chain of Thought — LLM reasoning is best understood as latent-state trajectory formation, not explicit chain-of-thought.
• ReTreVal: Reasoning Tree with Validation and Cross-Problem Memory for Large Language Models — Combining structured exploration, iterative refinement, explicit validation, and persistent memory significantly enhances LLM multi-step reasoning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.