Character-Sense Association and Compounding Template Similarity: Automatic Semantic Classification of Chinese Compounds
Chao-jan Chen · Meeting of the Association for Computational Linguistics · 2004
This paper presents a character-based model of automatic sense determination for Chinese compounds. The model adopts a sense approximation approach using synonymous compounds retrieved by measuring similarity of semantic template in compounding. The similarity measure is derived from an association network among characters and senses, which is built from a formatted MRD. Adopting the taxonomy of CILIN, a system of deep semantic classification (at least to the small classes) for V-V compounds is implemented and evaluated to test the model. The experiment reports a high precision rate (about 38% in outside test and 61% in inside test) against the baseline one (about 18%).