Semantic Labeling of Compound Nominalization in Chinese

This paper discusses the semantic interpretation of compound nominalizations in Chinese. We propose four coarse-grained semantic roles of the noun modifier and use a Maximum Entropy Model to label such relations in a compound nominalization. The feature functions used for the model are web-based statistics acquired via role related paraphrase patterns, which are formed by a set of word instances of prepositions, support verbs, feature nouns and aspect markers. By applying a sub-linear transformation and discretization of the raw statistics, a rate of approximately 77% is obtained for classification of the four semantic relations.