Higher-order regions in the brain play a role in organizing perceptual input into semantically-meaningful schematic representations, and they do so at differing timescales up the cortical hierarchy. Such integration in higher-order regions critically depends on the availability of temporal structure over long timescales. The ability to extract temporal structure is particularly adaptive because it enables the generation of predictions about upcoming events. How does the brain use repeated experience in structured environments to anticipate what is likely to come next? Here, we examine how the brain can anticipate event boundaries in familiar sequences of actions. We used a naturalistic stimulus (a movie), in which regularities are present at multiple timescales, enabling us to identify multiple timescales of prediction in the brain. We hypothesized that the timescale of neural prediction would vary continuously, with progressively higher-order regions (e.g., prefrontal cortex) predicting further in the future than lower-order regions (e.g., visual cortex). To test this, we examined brain activity with fMRI while individuals watched a 90-second clip from the movie The Grand Budapest Hotel, six times. To uncover neural anticipation, we used a searchlight approach in which, for each region of interest, we fit a Hidden Markov Model to identify temporal shifts between multivariate activity patterns evoked by the first viewing of the movie clip compared to repeated viewings. We found temporal shifts in event patterns throughout the brain, such that neural activity patterns for repeated viewings shifted to preceded those of initial viewing. The scale of prediction varied hierarchically in a posterior (less prediction) to anterior (more prediction) fashion, with prediction reaching up to 12 seconds in the future. Furthermore, in a subset of regions exhibiting prediction, neural event boundaries shifted with repeated viewing to precede subjective event boundaries by 6-7 seconds. Together, these results demonstrate that prediction is ubiquitous in the brain, but the amount of prediction varies across the cortical hierarchy.
bioRxiv Subject Collection: Neuroscience