Explore Theory of Mind: Program-guided adversarial data generation for theory of mind reasoning

TL;DR

ExploreToM introduces an A*-guided framework to generate diverse, challenging theory-of-mind data via a domain-specific narrative language, enabling robust evaluation and targeted training for LLMs. By separating story structure from lexical realization and grounding ground truth through a dedicated state tracker, it reveals deep weaknesses in current models (0%–9% accuracy on ExploreToM data) and demonstrates substantial downstream gains (up to +27 points on ToMi) when used for fine-tuning. The approach also exposes underlying skills needed for ToM, notably robust state tracking and data that explicitly requires theory of mind, while providing a scalable, adversarial benchmark resistant to data leakage. Overall, ExploreToM offers a principled, extensible path to both stress-test and systematically improve theory-of-mind reasoning in large language models, with broad implications for training data design and benchmark validity.

Abstract

Do large language models (LLMs) have theory of mind? A plethora of papers and benchmarks have been introduced to evaluate if current models have been able to develop this key ability of social intelligence. However, all rely on limited datasets with simple patterns that can potentially lead to problematic blind spots in evaluation and an overestimation of model capabilities. We introduce ExploreToM, the first framework to allow large-scale generation of diverse and challenging theory of mind data for robust training and evaluation. Our approach leverages an A* search over a custom domain-specific language to produce complex story structures and novel, diverse, yet plausible scenarios to stress test the limits of LLMs. Our evaluation reveals that state-of-the-art LLMs, such as Llama-3.1-70B and GPT-4o, show accuracies as low as 0% and 9% on ExploreToM-generated data, highlighting the need for more robust theory of mind evaluation. As our generations are a conceptual superset of prior work, fine-tuning on our data yields a 27-point accuracy improvement on the classic ToMi benchmark (Le et al., 2019). ExploreToM also enables uncovering underlying skills and factors missing for models to show theory of mind, such as unreliable state tracking or data imbalances, which may contribute to models' poor performance on benchmarks.

Figures (13)

• Figure 1: ExploreToM finds challenging stories for a given language model by searching through the space of stories supported by its domain-specific language for mental state tracking (), sampling $k$ supported actions at a time (shown as a node, $k=2$ in the example). Difficulty evaluation (simplified in the figure as easy, medium, hard) of each partial story is done through automatically generated questions with reliable ground-truth answers thanks to our tracking procedure.
• Figure 2: Overview of ExploreToM's story generation procedure. We first sample a plausible story context using an LLM (shown in A and B). Topics discussed, location changes of objects and people, and object state updates, may all be required to track in order to pass our theory of mind tests. We then search for difficult story structures (i.e., the raw story points) by sampling and analyzing different orders in which these actions may be performed using A* search (shown in C, and Fig. \ref{['fig:dsl']}). This ensures that the resulting stories will all be challenging tests for models, and may be used for further improvement. Finally, these story structures (nodes #1-4) are iteratively infilled, one story action at a time, using a language model, yielding a natural-sounding story. Infilled stories are used as training data; benchmarking is done with story structures since they have the highest reliability.
• Figure 3: Accuracy on ExploreToM's story structures when increasing the number of actions or people involved. Accuracy is computed across all story structure settings. Difficulty of ExploreToM-generated stories tends to increase or stay similar when increasing the number of actions. A story with greater number of people suggests similar or lower difficulty, possibly because when fixing the number of actions there are fewer actions per person (see details in App. \ref{['app:counterintuitive_phenomenon']}).
• Figure 4: ExploreToM-8B accuracy when evaluating on ExploreToM-generated data with more people $p$ and/or more actions $a$ than seen during training ($p\!<\!5, a\!<\!5$). Performance remains high when adding several actions and/or up to two people.
• Figure 5: ToMi accuracy when training with ExploreToM-generated data with different proportions of interesting questions (i.e., questions potentially requiring theory of mind to answer). Here, all variants are fine-tuned with 85000 story structure samples for 1 epoch.