[PS-1.2] Common computational principles under statistical learning of visual and auditory structures

Szabó, B. T. 1, 2 , Márkus, B. 3 , Nagy, M. 3 & Fiser, J. 3

1 Pázmány Péter Catholic University
2 Hungarian Academy of Sciences
3 Central European University

Traditionally, statistical learning studies in the auditory domain are linked to language processing and hence to sequential or "temporal" structure learning, while visual statistical learning research is focused more on discovering general spatio-temporal patterns, which requires spatial structure learning. Is this dichotomy warranted or should auditory statistical learning also treated in the more general framework of domain-independent discovering of spatio-temporal patterns? As a first step to resolve this issue, we transferred the classical spatial statistical learning paradigm used in vision to the domain of audition, and assessed whether human adult's learning behavior with auditory patterns of simultaneously presented sounds are similar to those found in vision. In two experiments, we found that listeners chunk such acoustic stimuli based on co-occurrences and conditional probabilities of the presented compound sounds in a very similar manner to what they do in the visual domain. We modeled these behavioral results with a Bayesian chunk learner to show that our results fit in the framework of probabilistic learning. Our results suggests that, apart from different spatial and temporal resolution in the two domains, the underlying learning mechanisms for acquiring representations of spatio-temporal patterns are similar for audition and vision.