Connectionist Modelling of Reading Aloud

A new class of neural network models of reading aloud are presented. Unlike previous models, they are not restricted to monosyllabic words, require no complicated input-output representations such as Wickelfeatures and require no pre-processing to align the letters and phonemes in the training data. The best cases are able to achieve 100% performance on the Seidenberg & McClelland training corpus, in excess of 90% on pronounceable non-words and on damage exhibit symptoms similar to acquired surface dyslexia. Their simplicity, performance and room for improvement make them a promising basis for the grapheme-phoneme conversion route of a realistic dual route model of reading.