Connectionist models in developmental cognitive neuroscience: Critical periods and the paradox of success

Abstract

Connectionist models have made significant contributions to understanding developmental phenomena, mainly by providing novel computational accounts of behavioral emergence and change. What is the fate of such models given the increasing interest in and information about the biological bases of development? We consider this issue with respect to the classical idea of a critical period for acquiring language. The standard view is that neurobiological developments on a strict maturational timetable create limits on language learning capacity. Computational analysis suggests the opposite: that learning itself creates neurobiological conditions underlying the “closing” of the critical period. The critical period example suggests how connectionist models can continue to provide a necessary level of analysis intermediate between behavior and brain. 2