Inducing Fuzzy Classes for Chinese Polysemic Verbs via Subcategorization Information

Abstract

This paper describes the application of Fuzzy k-Means, a derivant of k-Means that may assign an item to more than one cluster, in the task of inducing fuzzy classes for Chinese polysemic verbs. The probability distributions over subcategorization frames of 60 Chinese verbs, among which there are 40 polysemic ones and 20 monosemic ones are first acquired, and then these verbs are clustered into fuzzy classes. Evaluation and post-hoc analysis show that a combined measure of purity and pairwise precision can better estimate the clustering performance, and although to a certain extent syntactic behaviors of verbs have their counterparts of meaning components underlying, syntactic behaviors of verbs cannot be easily predicted from a single semantic level, at least for Chinese polysemic verbs.