Abstract: In Chinese natural language processing，semantic labeling is an important research domain.In the past researches，most label them in a manual way；in this paper，an automatically model extracting approach is applied to perform semantic labeling in Chinese，and is used to solve the problems such as classifying words and the relationships within compound nominalizations.The experiment results show that the labeling of word category has an accuracy of 81.640 5%，and the labeling of the relation within compound nominalization has also an improved result.