Mixtures of Experts: As an Attempt to Integrate the Dual Route Cascaded and the Triangle Models for Reading English Words

Abstract

An implementation of neural network models for reading English words aloud is proposed. Since 1989, there has been existing a debate in neuropsycholgy and cognitive science about the models of reading. One is the Dual Route Cascaded model, another is the Triangle model. Since there exist arbitrary variables in both models, it was difficult to decide which model would be appropriate to explain the data from psychological experiments and neuropsychological evidence. Therefore, in order to provide a solution of this debate, an attempt to integrate both models was attempted. By introducing the Mixtures of Experts Network model, a solution to overcome the arbitrariness of both models could be given. The Mixtures of Experts Network model could include both models as a special case. From the Mixtures of Experts network's point of view, the difference between the Dual Route Cascaded model and the Triangle model would be considered as a quantitative difference of the dispersion parameters.