ABSTRACT
This chapter reviews connectionist studies that address the issues surrounding the critical period hypothesis in second language learning. In an early study, Sokolik (1990) used learning rate to differentiate child learners from adult learners, while Elman (1993) and Rohde and Plaut (1999) used simple recurrent networks and relied on the manipulation of input and memory (hidden units) configuration to simulate children’s success, although the latter study found such manipulation is not necessary. Li (2009) and Vallabha and McClelland (2007), both used self-organizing networks to simulate Chinese–English bilinguals and Japanese–English interlanguage, respectively. Li focused on the input configuration to simulate the difference between early and late bilinguals, while Vallabha and McClelland focused on how L2 learners can get over L1-based phonology in acquiring L2.
Overall, these studies suggest that biological constraints as an explanation for the critical period effects may not be necessary. For example, maturational change on the part of the architecture is not necessary in simulating child–adult differences. Generally speaking, connectionist research supports the explanation that earlier neural commitment makes it harder to use the neural structure for learning the second language system.
