"A no B no C" is the most typical noun phrase and frequentcy in Japanese. This paper proposed a method automatically to generate dependency rules. In this method, we obtain 3 types of rules using part of speech,semantic attributes and both part of speech and semantic attributes. in the experiments, these rules were applied to the dependency analysis. As result, we obtain 86.8% accuracy rate. From these result, it was found that the proposed method obtains more accurate than ordinary method.