Abstract
Over the past 30 years, studies of learning in humans and other animal species have revealed powerful mechanisms that extract structured information from exposure to stimuli without the need for immediate reward or feedback. These implicit learning mechanisms are referred to as statistical learning because they are sensitive to the distributional information present in the ambient environment. Here we review the historical roots of statistical learning research as it evolved from the psychological and linguistic literature, later incorporating computational and cognitive neuroscience methods with the goal of fleshing out underlying mechanisms. We emphasize that statistical learning must be constrained by evolutionary and/or learning biases to function effectively and efficiently in a given context. We also discuss the component processes involved in statistical learning and how they may be implemented in the brain. Finally, we highlight the utility of studying statistical learning in early development. Immature neural systems lend themselves to an examination of their underlying components, which are more difficult to tease apart once they are consolidated as parallel interactive circuits in the mature brain.