Combining Support Vector Machines, Border Revised Rules and Transformation-based Error-driven Learning for Chinese Chunking

In research work, we found that grammatical information in the Modern Chinese Grammar Information Dictionary is very effective to revise chunk border. So the Modern Chinese Grammar Information Dictionary used to extract the chunk Border Revised Rules (BRR). In this paper, a new method of chunking is proposed--combined with BRR and TBL, SVM used for chunking. We reduced the number of SVM feature vector, and use SVM for chunking. Since the most chunk border error can be corrected by BRR, we reduced the number of feature vectors to shorten the training and chunking time of SVM. Finally, the data should be trained and modified again by TBL to obtain further the accuracy and the recall rate of improvement. We compare our method with method that combined with SVM and TBL. The experimental results show the method improves the precision and recall rates, while also reducing working hours.