This paper proposed a method automatically to generate dependency rules for "A no B no C", the most typical noun phrases and frequently appeared in Japanese (similar to "C of B of A" in English), using a semantic attribute system currently developed. In this method, 3 kinds of semantic dependency rules such as one-, two-, tree-dimensional rules are independently generated from a noun phrase data-base. In the experiments, the generated rules. were applied to the dependency analysis (in the order of 1-, 2- and 3-dimensional rules) resulting in the recall rate of 96.0%, precision rate of 88.4%, accuracy rate of 85.1%. From these result, it was found that the method yields accurate rules for dependency analysis.
Noun Analysis, Dependency Rule, Machine Learning, Generalization