Schwartz, N. 1 , Siegelman, N. 1, 2 , Christiansen, M. H. 2, 4, 5, 6 & Frost, R. 1, 2, 3
1 The Hebrew University of Jerusalem, Israel
2 Haskins Laboratories, New Haven, CT, USA
3 BCBL, Basque Center of Cognition, Brain and Language, Spain
4 Cornell University, Ithaca, NY, USA
5 Aarhus University, Denmark
6 ARC Centre of Excellence for the Dynamics of Language, Canberra, Australia
Statistical learning (SL) is the ability to extract regularities from sensory input. This ability underlies a range of cognitive functions such as segmentation, categorization, and language learning. Although much research has examined SL with tasks that include only full regularity (TP of 1), most regularities in the world and especially in language are typically quasi-regular. To date, it is unclear whether sensitivity to quasi-regularities can be trained and improved through practice. This has important potential practical implications, with the possibility of enhancing language learning. The present study examines the impact of systematic and adaptive SL training on sensitivity to low transitional probabilities (TPs), using a probabilistic SRT task. We present data from 28 participants, demonstrating learning at various TP levels. We assessed individual differences in the minimum TP level that can lead to learning of co-occurrence of events, and measured the amount of training that was needed to achieve this level. We discuss the theoretical issues that arise from these results, as well as the methodological challenges of designing a training procedure that can differentiate between initial learning ability and the effect of training.