A computational exploration of double dissociations: modes of processing instead of components of processing

Two connectionist models are reported that simulated the defining features of the double dissociation between phonological and surface dyslexia in word reading. One model was a feed-forward, three-layer perceptron, and the other included recurrent connections. Neither model contained an architectural separation of sublexical and lexical processes, nor of phonological and semantic processes. Analyses showed that the double dissociation was simulated because the control parameter input gain shifted the models between conjunctive and componential modes of processing. The dissociation was not simulated by any kind of damage to separate system components. The simulations are discussed in the context of current accounts of surface and phonological dyslexia.