Communicating understanding of physical dynamics in natural language


Our ability to share abstract knowledge with others is a defining feature of modern human intelligence. What information do people choose to include in such communication? Here we develop a novel physics-based video game to elicit natural language responses on how this game works to teach other people. We collected data from 238 participants and found that people explicitly described the latent physical properties of the game environment like mass and gravity in their responses. We also found that people who performed better in the game also produced responses that covered more latent physical properties. Taken together, our study provides novel insight into how people communicate their understanding of physical dynamics in natural language.

44th Proceedings of the Annual Meeting of the Cognitive Science Society, 2022