A “Triangle Model” of Chinese Reading

In this paper we present two connectionist models of reading, using a parallel distributed processing framework that has been applied to English, to examine the extent to which such models can also account for developmental performance in Chinese. Simulation 1 was trained to map from orthography to phonology for a large corpus of stimuli, and simulates the frequency, regularity, consistency effects and interactions among them. Simulation 2 was trained to map among spelling, meaning, and sound for a smaller set of items, and captures basic effects of orthographic transparency and family size. Although the computational models used here are very similar to the English models, a very different developmental pattern emerged, such that mappings from orthography to semantics were learned more rapidly than mappings from orthography to phonology. The results show how qualitative differences in the development of reading skill across writing systems can arise from the functioning of the same general learning mechanisms.