Proceedings of the Annual Meeting of the Cognitive Science Society

From childhood, people routinely explain each other’s behav-ior in terms of inferred mental states, like beliefs and desires.In many cases, however, people can also infer the mental statesof agents whose behavior we cannot see, such as when we in-fer that someone was anxious upon encountering a chewed-uppencil, or that someone left in a hurry if they left the door open.Here we present a computational model of mental-state attri-bution that works by reconstructing the actions an agent took,based on the indirect evidence that revealed their presence. Ourmodel quantitatively fits participant judgments, outperforminga simple alternative cue-based account. Our results shed lighton how people infer mental states from minimal indirect evi-dence, and provides further support to the idea that human The-ory of Mind is instantiated as a probabilistic generative modelof how unobservable mental states produce observable action.