Evaluating Spatial Understanding of Large Language Models

TL;DR

This paper investigates how well large language models understand and reason about spatial relationships through specially designed navigation tasks, revealing both capabilities and limitations in their implicit spatial knowledge.

Contribution

The study introduces natural-language navigation tasks to evaluate LLMs' spatial reasoning, highlighting variability in performance and analyzing the nature of their errors.

Findings

LLMs show variable performance across different spatial structures

Errors reflect both spatial and non-spatial factors

LLMs implicitly capture some aspects of spatial structure

Abstract

Large language models (LLMs) show remarkable capabilities across a variety of tasks. Despite the models only seeing text in training, several recent studies suggest that LLM representations implicitly capture aspects of the underlying grounded concepts. Here, we explore LLM representations of a particularly salient kind of grounded knowledge -- spatial relationships. We design natural-language navigation tasks and evaluate the ability of LLMs, in particular GPT-3.5-turbo, GPT-4, and Llama2 series models, to represent and reason about spatial structures. These tasks reveal substantial variability in LLM performance across different spatial structures, including square, hexagonal, and triangular grids, rings, and trees. In extensive error analysis, we find that LLMs' mistakes reflect both spatial and non-spatial factors. These findings suggest that LLMs appear to capture certain aspects of spatial structure implicitly, but room for improvement remains.