Setting the stage for statistical learning? Sensitivity to environmental statistics in early sensory processing

Our brains make sense of the world on a moment-by-moment basis despite its enormous complexity, largely because its overall statistical structure can be detected, learned, and generalized across experiences. Exposure to specific regularities (e.g., in speech) results in an unsupervised, incidental, form of learning, commonly known as statistical learning (SL). SL is well-established from a cognitive perspective and often assumed to require high-level cortical or hippocampal processing. However, accumulating evidence suggests that SL emerges much earlier in ascending sensory pathways. Despite this, our understanding of the forms it might take in subcortical sensory centres is relatively limited. Here, we review neuronal sensitivity to statistics in early sensory regions and ask how this sensitivity relates to SL. We feature examples of adaptive responses elicited by stimulus repetitions, omissions, changes in stimulus distribution, and more complex patterning, highlighting the interplay between adaptive coding and SL as manifestations of sensitivity to environmental statistics.