Tosatto, L. 1 & Rey, A. 2
1 Aix-Marseille Université
2 CNRS
In most statistical learning experiments, participants have to retrieve regularities within a continuous flow of information. Several models suggest that we learn these regularities by grouping elements frequently encountered into chunks of information. Previous studies provided information about the extraction dynamics of short recurring sequences but the chunking mechanisms underlying long sequence learning have been less explored (except in the domain of motor sequence learning). In this study, we asked human participants to learn a repeated visuo-motor sequence composed of 9 elements. This same sequence was repeated 120 times and we studied the evolution of response times for each element of the sequence through the course of learning. Our results show that all participants divided the sequence into several chunks of 3 or less elements. Moreover, participants processing the same sequence divided that sequence into the same set of chunks. These data therefore suggest that chunking was relatively stable across participants, with the largest chunks being composed of a maximum number of three adjacent elements.