Language model of Chinese character recognition and its application

This paper presents a 5-gram combined model that can reflect features of Chinese and Chinese character recognition based on introducing several kinds of Markov language models. The major feature of this model is that it captures both forward and backward statistical characters of one word. The model contains three traditional "trigram components", a "cache component" which reflects short-term patterns of word use, and a "3g-gram component" based on a new classification method that is fast and automatic. Experiment on a 1,500,000-word corpus shows significant improvement achieved by the proposed model.