Attractor dynamics in word recognition: converging evidence from errors bynormal subjects, dyslexic patients and a connectionist model

People make both semantic and visual errors when trying to recognise the me aning of degraded words. This result mirrors the finding that deep dyslexic patients make both semantic and visual errors when reading aloud. We link the results with the demonstration that a recurrent connectionist network w hich produces the meaning of words in response to their spelling pattern pr oduces this distinctive combination of errors both when its input is degrad ed and when it is lesioned. The reason why the network can simulate the err ors of both normal subjects and patients lies in the nature of the attracto rs which it develops as it learns to map orthography to semantics. The key role of attractor structure in the successful simulation suggests that the normal adult semantic reading route may involve attractor dynamics. (C) 200 0 Elsevier Science B.V. All lights reserved.