Models of word reading: theory-building vs. data-fitting

old_uid3442
titleModels of word reading: theory-building vs. data-fitting
start_date2007/11/13
schedule11h-12h
onlineno
summaryI will discuss two (not the only) approaches to models of word reading: the distributed connectionist approach developed by myself, McClelland, Plaut and others, and the DRC approach developed by Coltheart and colleagues. The two approaches entail very different goals and criteria for success. Our goal has been to explore a set of basic principles thought to underlie many aspects of cognition and learning. The models are tools in the development of general theories. Advocates of the DRC approach are skeptical about discovering general principles; the main criterion for success is how many empirical phenomena a model can simulate. I'll review the strengths and weaknesses of each approach. The major conclusion is that the model-fitting approach results in overfitting and models that fail to generalize. The connectionist approach has its own limitations, but has proved more successful in identifying general principles that are not tied to details of individual implemented models.
oncancelExceptionnellement un mardi
responsiblesPélissier