Modelling Early Stages of Word Recognition: Case and the Peripheral Dyslexias

Abstract

This study investigates the visual representations used in the early stages of word recognition. In particular we look at the role of letter position information and the possible existence of visual units which are larger than single letters. Simulations here are intended to model the early stages of visual word recognition in both normal reading and in cases of peripheral dyslexia. In a feedforward network, letter units are coarse coded for position across words. The model uses separate sets of units to code upper and lower case letters. Exposure to all upper and all lower case words trains the letter units to activate a common set of local phonological units

In the simulation of normal reading we examine the finding that word naming errors are more likely to occur in the middle of words as opposed to the ends. This result is simulated by the model due to the nature of its representations of letters and their positions. We also examined effects produced by coarsening location coding: something that may occur in patients with poor reading associated with damage to the parietal lobe. With coarse location coding visual reading errors increase, particularly for central letter positions, and greater numbers of migration errors occur when two words are presented. These deficits are discussed in relation to the neuropsychological syndrome of attentional dyslexia.