Language Model for Chinese Character Recognition with Dense Errors

摘要

In this paper we present a new language model that intends to raise recognition rate when there are dense errors in sentences. Based on language models brought forward previously such as 5-gram combined model and variable length language model, this language model make use of the candidates of errors and short-term information. Previous language models including 5-gram combined language model can effectively correct errors when they distribute evenly, but this model plans to correct dense errors also. In the end, we make experiments and get encouraging result.