A new model of letter string encoding: simulating right neglect dyslexia.

Introduction Cognitive models of how oral reading is normally accomplished are typically developed and tested on the basis of experimental data gathered from skilled readers. Increasingly, however, information from other sources is playing an important role in the elaboration of such models. Detailed analyses of the eeects of focal brain lesions on skills such as oral reading can place limits on the possible types of relationships that might exist among hypothesized processing components see, e.g., Shallice, 1988. Of special interest are analyses of the relationship between the target words patients are attempting to read and the types of errors they produce, which m a y implicate breakdown of semantic, orthographic anddor phonologic processing. A second methodology that has begun to play an important role in the elaboration of cognitive theories of reading involves the development and testing of computational models. Implemented computer simulations of oral reading have demonstrated the feasibility of such hypothesized processing details as the distinction between serial assembly of sub-lexical units for unfamiliar words and parallel access to stored orthographic units for familiar words Coltheart et al., 1993. Frequently, neuropsychological data are used to test computational models, which should beable to bedegraded with simulated lesions" to reproduce clinical patterns Coltheart et al. This paper describes a theoretical model of some aspects of oral reading that was developed to be consistent with experimental data from normal subjects and with existing neurobiological theory. Computer simulations based on this model can be lesioned" to reproduce an error pattern obtained from a group of adult patients with left hemisphere lesions. The development of this model forced explicit consideration of a number of issues concerning normal word processing, especially the important issue of how letter order is encoded in words. We will describe the genesis of the model, including the motivation for modiications made during its development, to illustrate the value of computational modeling for investigating cognitive processes and their impairments. The error pattern at issue here is one in which patients substitute orthographically related words w ords with substantial overlap of the targets' letters for the words they are trying to read. The pattern of orthographic substitution is not random; rather, letters on the side of the target word opposite the patients' hemispheric lesions are much more likely be lost from the response than are letters from positions on the same side of the word as the lesion. A pattern of …