Division of Labor between Semantics and Phonology in Normal and Disordered Reading Development across Languages

In order to study how statistical patterns in different languages can shape the division of labor in the reading system, we trained two models with the same basic architecture and learning parameters: One was trained to read Chinese, the other English. In Simulation 1, we examined the role of semantics in the early development of reading by comparing results from training with and without input from semantics. Semantic input had relatively modest effects for learning to read English, mainly improving performance on exception words. In Chinese, the influence of semantics was much greater and much more widespread: all types of words were learned more quickly, although the benefit was still greater for words with atypical spelling-to-sound correspondences. In Simulation 2, we simulate the effect of developmental deficits in phonological and semantic processing on the development of reading. Consistent with data concerning individual differences in reading skill, phonological deficits had a much greater impact on English, whereas semantic deficits had a more serious impact on learning to read Chinese. The results demonstrate that differences in the division of labor among readers of different writing systems can be explained in terms of the statistical properties of the writing systems and their interaction with generic associative learning mechanisms.