Grapheme Generation in Learning to Read English Words

摘要

A learning system for the complete process of learning to read aloud English words consists of three major steps: alignment, mapping learning, and grapheme generation. Such a system can be used to construct prototypes of gradint machines for English or other lanugages quickly and automatically. Almost all previous learning systems form reading (such as NETtalk) dealt only with mapping learning. In a previous paper, we discussed a basic alignment algorithm and several improvements that effectively accomplish the alignment and mapping learning simultaneously. In this paper, we focus on grapheme generation. We present several criteria for constructing graphemes, and demonstrate that graphemes generated improve the predictive accuracy in learning to read aloud.