Computational Theories of Learning and Developmental Psycholinguistics
Jeffrey Heinz · Oxford University Press eBooks · 2016
There is a bright future for research which honestly integrates the insights of computational learning theories with the insights and methodologies of developmental psycholinguistics. With this aim in mind, this chapter surveys results in computational learning theory focusing on the many formal definitions of learning and the kinds of patterns which can and cannot be learned according to these definitions. The main takeaways are that the central problem of learning is generalization, that feasible learning can only occur when target classes of patterns are restricted and structured appropriately, and that debates pitting statistical learning against symbolic learning are largely misplaced since the real issue there is about which data presentations learners should succeed on.