Brice, H. 1 , Bogaerts, L. 1 , Schwartz, N. 1 , Eisenbach, T. 1 & Frost, R. 1, 2, 3
1 The Hebrew University of Jerusalem
2 Basque Center on Cognition, Brain and Language
3 Haskins Laboratories
Most proposed mechanisms for statistical learning (SL) assume either pre-existing representations of the elements across which regularities are being learned, or else that forming representations of the elements of the input is a necessary precursor to learning about their co-occurrence (Frost, Armstrong, Siegelman, & Christiansen, 2015). Research has shown, however, that encoding visual elements and picking up on their co-occurrences are not two distinct and temporally modular processes, but rather interact to jointly determine performance in a visual SL task (Bogaerts, Siegelman & Frost, 2016).
Using both behavioural measures and pupillometry, we investigated the intriguing possibility that encoding individual abstract shapes in visual short-term memory during fast sequential presentation is facilitated if the shapes co-occur (B follows A) as compared to if their order of presentation is random. This encoding advantage is observed following only three repetitions of each stimulus, in absence of any evidence that the regularities themselves have been learned. This suggests that the existence of regularity is part and parcel of learning about individual stimuli.