Authors:

Patterson, K.E.

Reference:

Year of publication:

1990

CBU number:

2492

Abstract:

A neural net model (developed by Seidenberg & McClelland, 1989) computes phonological representations for alphabetic letter strings. Learning, over a series of training experiences with English words, corresponds to changes in weights on connections between units at different layers in the model. After training, the performance of the model simulates many features of the word-naming performance of adult English readers. "Lesioning" the trained model also yields at least some features of one prominent form of acquired alexia. The distributed representations used by this and other neural net models have important implications for modelling cognitive impairments that result from brain lesions.