The sensitive feature selection for both English and Chinese text chunking

Traditional text chunking approach is to identify many phrases using only one model, and the same features are used to identify these phrases too. So the helpful features of each phrase are ignored. In fact, different phrases have different helpful features. In this paper, the concept of ¿sensitive feature¿ is proposed, and the sensitive features of eleven English types and seven Chinese types of phrases are selected by dynamic comparison strategy. Through testing on the Multi-agent chunking model, the selected English and Chinese sensitive features are both effective.