Skip to Main Content
In this paper, we present a distant BI-Gram model, which extended the regular BI-Gram model by considering the distance information and weight parameters, in order to describe the long-distance restrictions among the Chinese sentence. The extraction of the statistical information and weight parameters of this language model is discussed. Based on this work, the word combination strength and spread are employed to extract the recurrent word combinations, i.e. collocations. The distant BI-Gram model and collocation are applied to a statistic-based post-processing system for improving the recognition performance of Chinese characters. The experimental results show that by employing these two language models, the post-processing system achieves a higher improvement performance.