Implementing the “Simple” model of reading deficits: A connectionist investigation of interactivity

The “Simple model” of reading proposes that reading deficits can be understood in terms of single or combined contributions of comprehension and phonological impairments [1]. Phonological or semantic deficits have been effectively tested in isolation in computational models of reading, and their results map onto the predictions of the simple model. However, the subtle interactions between impaired representations have not yet been fully explored. We implemented a connectionist triangle model of reading which learned to map between orthographic, phonological, and semantic representations of words for a large sample of monosyllabic words in English. During learning, the model was impaired in terms of its ability to stably represent phonological or semantic information. The model was then tested on a variety of word and picture naming tasks. The model replicated previous studies of dyslexia resulting from a phonological deficit, and comprehension difficulties resulting from a semantic deficit. However, we also found that impairment at either phonological or semantic representations affected the fidelity of representations throughout the model, indicating that pure deficits in reading will be difficult to observe. The implemented model provides constraints on the extent to which phonological and semantic information may interact in order to explain reading and its deficits.