Language Models Might Not Understand You: Evaluating Theory of Mind via Story Prompting

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Getachew, Nathaniel, Saparov, Abulhair
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910164226932736
author Getachew, Nathaniel
Saparov, Abulhair
author_facet Getachew, Nathaniel
Saparov, Abulhair
contents We introduce StorySim, a programmable framework for synthetically generating stories to evaluate the theory of mind (ToM) and world modeling (WM) capabilities of large language models (LLMs). Unlike prior benchmarks that may suffer from contamination in pretraining data, or rely on an LLM for generation, StorySim produces novel, compositional story prompts anchored by a highly controllable Storyboard, enabling precise manipulation of character perspectives and events. We use this framework to design first- and second-order ToM tasks alongside WM tasks that control for the ability to track and model mental states. Our experiments across a suite of LLMs show that most models achieve higher accuracy on WM tasks than on ToM tasks, and that models tend to reason more accurately when the subject of reasoning is a person rather than an inanimate object. Additionally, our framework enabled us to find evidence of heuristic behavior and an over-reliance on earlier events in the story. All code for generating data and evaluations is freely available.
format Preprint
id arxiv_https___arxiv_org_abs_2506_19089
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Language Models Might Not Understand You: Evaluating Theory of Mind via Story Prompting
Getachew, Nathaniel
Saparov, Abulhair
Computation and Language
Artificial Intelligence
We introduce StorySim, a programmable framework for synthetically generating stories to evaluate the theory of mind (ToM) and world modeling (WM) capabilities of large language models (LLMs). Unlike prior benchmarks that may suffer from contamination in pretraining data, or rely on an LLM for generation, StorySim produces novel, compositional story prompts anchored by a highly controllable Storyboard, enabling precise manipulation of character perspectives and events. We use this framework to design first- and second-order ToM tasks alongside WM tasks that control for the ability to track and model mental states. Our experiments across a suite of LLMs show that most models achieve higher accuracy on WM tasks than on ToM tasks, and that models tend to reason more accurately when the subject of reasoning is a person rather than an inanimate object. Additionally, our framework enabled us to find evidence of heuristic behavior and an over-reliance on earlier events in the story. All code for generating data and evaluations is freely available.
title Language Models Might Not Understand You: Evaluating Theory of Mind via Story Prompting
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2506.19089