What Moves in a Mysterious Way? A domain-general account of learning about animacy and causality

A fundamental debate within cognitive science concerns how infants, children, and adults learn about the motion properties of animates and inanimates. In this paper, we show that an associative-learning mechanism implemented as a neural network can use animacy relations to predict and discriminate between familiar and novel causal and noncausal events. This is possible because (1) animate objects are more similar to each other than to inanimate objects, (2) causal events are more likely to have animate agents and inanimate recipients than noncausal events, (3) noncausal events are more likely to have animate (i.e., self-propelled) “recipients”, and (4) causal and non-causal events correlate with different types of motions. The results suggest that the emergence of concepts for animacy and causality in infancy and beyond can be explained without theory-driven top-down processing or specialized modules.