A Practical Tagger for Chinese Corpora

In this paper, we propose an automatic Chinese tagger having the accuracy ranged from 96% to 98% depending on the types of texts. Since large fully tagged Chinese corpus is not available, the relaxation labeling method is first adopted to select the statistically most plausible parts-of-speech for words which are categorically ambiguous. The performance of the relaxation labeling method is not satisfactory, hence we propose a hybrid approach which combines the relaxation labeling method with a rule-based method. The two methods complement each other. The accuracy of the relaxation labeling method is increased by 7%, because the statistically problematic ambiguities were resolved by the rules.