A statistical method for Uyghur tokenization

Tokenization is very important for Uyghur language processing. Tokenization of Uyghur, an agglutinative language, is quite different from other languages such as Chinese and English. In this paper we propose a two-steps statistical tokenization method for Uyghur. Two related factors, the feature template scheme and the manually tokenized corpora, are also discussed. The preliminary experiment results demonstrate that the proposed method is effective: the F-measure of tokenization reaches 88.9% in the open test.