 |
|
|
Acquired deep dyslexia is a reading disorder characterized by the coincidence of semantic, visual, and mixed semantic-and-visual reading errors. Apart from this central pattern of coinciding errors a number of other core symptoms are associated with deep dyslexia as well, one of which is the so-called concreteness effect - namely a clear advantage in processing concrete words compared to abstract ones. The paralexias in response to abstract target words, typically, tend to be more concrete then the original stimuli. Most accounts of symptom origin in deep dyslexia assume that at least two functionally and locally distinct processing components of the cognitive system responsible for the processing of written words have to be damaged so as to give rise to such a complex pattern of impairments. In contrast to these classic approaches to deep dyslexia Hinton and Shallice (1991), and Plaut and Shallice (1993) describe a connectionist model which shows that lesions to a strictly non-modular attractor neural network which was previously trained to map from orthography to semantics may result in erroneous output patterns which bear striking resemblance to reading errors of deep dyslexic patients. Apart from showing the characteristic combination of error types, i.e. semantic, visual, and mixed errors, the damaged neural network also proved to be susceptible to the relative degree of the input word's concreteness. One critical point in this connectionist approach to the concreteness effect in deep dyslexia is the idiosyncratic structure of the corpus the network was trained to process. The structure of the word list is such that it does not allow an unambiguous decision on whether the observed effects of concreteness on the network's processing performance actually reflect the properties of the artificial neural network or, alternatively, if they might be attributable primarily to particularities of the corpus the network was trained with. The aim of this present simulation study is, in the first place, to replicate the results Plaut and Shallice published in their papers (Plaut & Shallice, 1993; Plaut, 1995), and, secondly, to investigate the assumed dependence from the corpus design by training a second neural network with a somewhat altered word list. The results of this alternative procedure strongly support Plaut and Shallice's interpretation of the data. It was shown in the context of the present study that the influence of the input word's degree of concreteness on the network's reading performance is an actual consequence of the connectionist model's most fundamental functional principles. The network produced the three aspects that make up the concreteness effect in deep dyslexia after lesions, independent of the respective corpus design. References Hinton, G. E. & Shallice, T. (1991). Lesioning an Attractor Network: Investigations of Acquired Dyslexia. Psychological Review, 98, 74-95. Plaut, D. C. (1995). Double Dissociation without Modularity: Evidence from Connectionist Neuropsychology. Journal of Clinical and Experimental Neuropsychology, 17, 291-321. Plaut, D. C. & Shallice, T. (1993). Deep Dyslexia: A Case Study of Connectionist Neuropsychology. Cognitive Neuropsychology, 10, 377-500. Many thanks to Professor Rainer Bösel, who provided me with invaluable support and guidance while the work reported here was carried out. |