Schapiro, A. 1 , Rogers, T. 2 , Cordova, N. 1 , Turk-Browne, N. 1 & Botvinick, M. 1
1 Department of Psychology and Princeton Neuroscience Institute, Princeton University
2 Department of Psychology, University of Wisconsin-Madison
Our experience of the world seems to divide naturally into discrete, temporally extended events, yet the mechanisms underlying the learning and identification of events are poorly understood. Research on event perception has focused on transient elevations in predictive uncertainty or surprise as the primary signal driving event segmentation. We report behavioral and neuroimaging evidence in favor of an alternative account in which event representations coalesce around clusters or ‘communities’ of mutually predicting stimuli. Through parsing behavior, fMRI adaptation, and multi-voxel pattern analysis, we demonstrate the emergence of event representations in a domain containing such community structure, but where transition probabilities (the basis of uncertainty and surprise) are uniform. We present a neural network model that is exposed to the same novel event structure as experiment participants. The model naturally and spontaneously exploits the shared predictive structure within events to form event representations that resemble those found in the brain. The model additionally provides an explanation for event segmentation behavior in terms of detection of large changes in the learned internal representation of a stimulus, as opposed to detection of changes in the transition probabilities of stimuli in the environment. This work links event representation to visual associative learning in novel ways, and suggests that common principles may underlie the representation of events and of object categories in semantic cognition.