Exploring Spatial Intuitions and Analogical Reasoning in Large Language Models

This research explores how Large Language Models (LLMs), despite lacking physical embodiment, process and reflect abstract spatial intuitions similar to human cognition. Through a synthesis of psycholinguistic experiments and analogical reasoning tasks, we explore how LLMs mirror human associations between spatial concepts and abstract ideas. Drawing from spatial cognitive foundations, the study assesses the models' ability to use and generate analogies and how these analogical prompts influence their responses. The findings reveal overlaps between human responses and LLM behavior, highlighting both the potential and limitations of LLMs in navigating the intricate relationship between language, abstraction, and embodied cognition.