EXPERIMENTS AND ANALYSES OF CHINESE PART OF SPEECH TAGGING BASED ON NON SUPERVISION TRAINING

Probability parameter obtaining is one of the two main study directions of part of speech tagging based on statistics. In this paper, emphasis is laid on non supervision model study, and probability parameters are obtained by training using untagging corpus. A non supervision training tagging model——HMM Basic is implemented. Experiments on Chinese part of speech tagging from different initial models and training sets are made, and the influence on the tagging performance as a result of the selections of the training set size and the initial model is discussed. And the existent problems are also analysed.