Mareschal, D. 1 , French, R. M. . 2 & Mareschal, D. 1
1 Birkbeck University of London
2 CNRS-LEAD/Université de Bourgogne
Statistical learning is an important prerequisite for language learning in infancy (Aslin, Saffran & Newport, 1998). Similar skills and abilities seem to underlie adult sequential and implicit learning. We present a single model of both these abilities. The Truncated Recursive Autoassociative Chunk eXtractor (TRACX) outperforms PARSER (Perruchet & Vintner, 1998) and the simple recurrent network (SRN; Cleeremans & McClelland, 1991) in matching human sequence segmentation data. The mechanism relies on the recognition of previously encountered subsequences (chunks) in the input rather than on the prediction of upcoming items in the input sequence. Here, we show that a developmental version of the model is consistent with the results of both Kirkham, Slemmer and Johnson (2002) and Marcovitch and Lewkowicz (2009). The former habituated infants at ages 2, 5, and 8 months old to a sequence of visual looming shapes and found a preference for a more random sequence at test in all ages. The latter used the same procedure but independently varied conditional probability and pair frequency information. In that study 4.5- and 8.5-month-old infants showed a preference but 2.5 month olds did not. Finally, we also present an web-based implementation of the model that makes the model more widely accessible to the research community. The simulator runs inside a browser window with all code executed locally on user?s computer.