A hybrid approach for chinese pronunciation-translated person names recognition
Yan Liang, Yaoting Zhu · 2008
Pronunciation-translated person names (PPN) bring ambiguities to Chinese word segmentation. In this paper, we regard PPN recognition as a binary classification problem. We propose a hybrid approach that combines conditional random fields (CRF) model and support vector machines (SVM) model for the task of recognizing PPN. The experiments show that the performance of the hybrid model is better than either the CRF model or the SVM model. With regard to the analyses of the results individually generated by the CRF model and the SVM model, we also apply some appropriate rules to the hybrid model in order to prune errors. According to our overall experiments, the hybrid method with rules achieves a high precision in the final results, which demonstrates that our hybrid model is effective.