Multiple TreeBanks Integration for Chinese Phrase Structure Grammar Parsing Using Bagging

We describe our method of traditional Phrase Structure Grammar (PSG) parsing in CIPS-Bakeoff2012 Task3. First, bagging is proposed to enhance the baseline performance of PSG parsing. Then we suggest exploiting another TreeBank (CTB7.0) to improve the performance further. Experimental results on the development data set demonstrate that bagging can boost the baseline F1 score from 81.33% to 84.41%. After exploiting the data of CTB7.0, the F1 score reaches 85.03%. Our final results on the official test data set show that the baseline closed system using bagging gets the F1 score of 80.17%. It outperforms the best closed system by nearly 4% which uses a single model. After exploiting the CTB7.0 data, the F1 score reaches 81.16%, demonstrating further increases of about 1%.