X hits on this document

PDF document

The split-fovea model of visual word - page 10 / 78





10 / 78

Chapter 1. Introduction and Background


Connectionist Models

Connectionist models are an alternative to the Dual-Route model for visual word recognition and reading. Connectionist networks are computational models that try to be more brain-like than the more traditional box-and-arrow models (e.g. the Dual- Route model). The most basic components of these networks are abstract models of neurons called units. The connections between neurons are modelled by weighted connections between the units. The units are usually grouped into different layers. A typical feed-forward network generally has an input layer, a hidden layer and an output layer. Each unit in the input layer is connected to each unit in the hidden layer, which in turn is connected to each unit in the output layer (see Fig. 1.2). Activation from the input units spreads to the hidden and then to the output units. For an overview of connectionist modelling see for example O’Reilly and Munakata (2000).

A lot of research has been done in using connectionist networks for visual word recog- nition (for an overview see Christiansen and Chater (1999). The networks for reading are generally structured as that in figure 1.2 with an input, an output and a hidden layer. It is important to note that unlike the Dual-Route model of reading, the connectionist models have to use the same mechanism to read novel words and exception words.The models show that it is possible to have a reading system that uses the same system for all words, which is contrary to the claim made by Coltheart, who says that there have to be two routes. One of the big challenges for early connectionist models was to prove that the networks could replicate the behaviours of surface and phonological dyslexia if they were damaged in different ways. One of the first successfull connec- tionist models of reading was the one by Seidenberg and McClelland (1989), which was subsequently improved (Plaut et al., 1996; Harm and Seidenberg, 1999). These connectionist networks were not only able to correctly pronounce normal, exception and even non-words, but they also managed to simulate dyslexic behaviour. The fact that they manage to replicate surface and phonological dyslexia is an important part in

Document info
Document views276
Page views290
Page last viewedWed Jan 18 02:27:03 UTC 2017