Hasson, U.
The University of Trento, Italy
The capacity for statistical learning enables rapid assessment of the degree of uncertainty in the environment, thus allowing calibration of predictive and bottom-up processing. While initial neurobiological studies conjectured that there might be a single a-modal system coding for the degree of input uncertainty, later research has shown that the neurobiological implementation is more complex. Several recent neuroimaging and behavioral studies from our group indicate the following: 1) rapid learning of statistical regularities in visual and auditory domains relies on different neurobiological systems with little overlap between them; 2) different systems track changes in uncertainty when uncertainty is manipulated via transition probabilities or marginal frequencies; 3) systems sensitive to regularity operate differently depending on the familiarity of the input tokens; 4) perceived changes in regularity are associated with activity in neural systems related to event segmentation; and 5) the brain is sensitive not only to statistical features of the input, such as its entropy, but also to the complexity of the mechanism generating the input. Taken together, these findings point to highly distributed brain systems that are relatively specialized for both modality and specific statistical features. There is no evidence for a core common neurobiological system implementing this capacity.