A computational approach to dyslexic reading and spelling

This chapter discusses the ability of computational models to improve our understanding of dyslexic reading and writing. Connectionist models of the development of alphabetic reading and spelling in normal and dyslexic children are described. The models learn to associate representations of word pronunciations with spellings. The models learn to read and spell regular words more quickly than irregular items. When the computational resources available to such models are restricted, the models learn more slowly and fail to learn some of the irregular items in their vocabularies. The restricted models behave analogously to developmental dyslexics, and, crucially, have selective deficits with non-word processing although they do not show reduced sound-to-spelling or spelling-to-sound regularity effects. This is consistent with the experimental literature. Experimental evidence is reported that shows that both normal and dyslexic children of various ages have difficulties with reading and spelling particular word types that are similar to the problems experienced by the models on the same words. The good fit between model and data is taken as evidence that, throughout much of the relevant developmental period, the task facing children can be usefully viewed as a statistical one. The level of difficulty posed by particular words in spelling is well predicted by the extent to which those words conform to the relevant regularities of the language. Furthermore, the models resolves an apparent paradox in the experimental literature, for in their dyslexic forms they exhibit a selective deficit in non-word reading and spelling even though they do not show reduced sound-to-spelling regularity effects.