Learning about actions in infancy without a rationality principle

Between 6 and 9 months of age, infants begin to differentiate between the actions of others that are “rational” with respect to goals and those that are not. According to the teleological stance theory, this behavior is underpinned by an innate, naive rationality principle; according to a statistical learning account, experience alone is sufficient to explain this behavior. We present a recurrent neural network implementation of statistical learning that incorporates relevant pre-experimental experience that serves to shape expectations about goal-directed action. This model replicates the looking-time patterns of infants at different points in developmental time, and it demonstrates that a rationality principle is not necessary to account for the extant data.