Marchand, Y. 1, 2 & Damper, R. . 3
1 Department of Psychology. Dalhousie University. Halifax, Canada.
2 Faculty of Computer Science. Dalhousie University. Halifax, Canada.
3 School of Electronics and Computer Science. University of Southampton. Southampton, United Kingdom.
The goal of the present study was to determine if Pronunciation by Analogy (PbA), which is a computational method for converting letters to sound in text-to-speech systems, could be used to simulate the core reading deficits found in surface and phonological dyslexia. Surface dyslexia is characterized by the following cardinal symptoms: an ability to read regular words and nonwords but serious difficulty reading irregular words, with a strong tendency to “regularize” them (e.g., “have” read like “gave”). Phonological dyslexia is characterized by ease of reading both regular and irregular words and poor reading of nonwords. A literature review was conducted and case studies from 12 individuals (6 people with surface dyslexia and 6 people with phonological dyslexia) were identified. To reproduce these diverse reading impairments, the methodology consisted of selectively degrading certain component parts of the PbA architecture. PbA exploits the phonological knowledge inherent in a dictionary of words with corresponding pronunciations. The underlying key idea is that a pronunciation for an unknown word is derived by matching substrings of the input to substrings of known words in a lexicon. The PbA model consists of four components: the lexical database, the matcher that compares the target input to all the words in the database, the graph data structure (i.e. a pronunciation lattice) and the decision function that selects the “best” pronunciation among the set of possible ones. It was possible to successfully simulate the reading deficits of these 12 individuals with dyslexia by partially damaging both the segmentation of the matcher and the decision function. Furthermore, phonological and surface dyslexia differentiated from each other by the window size involved in the faulty segmentation procedure. Indeed, a larger size window appeared to be necessary to reproduce the characteristics of individuals with phonological dyslexia compared to surface dyslexia.