Japanese dependency analysis using fuzzy support vector machines

This paper introduces fuzzy support vector machines (FSVMs) for Japanese dependency analysis. Japanese dependency analysis based on support vector machines (SVMs) has been proposed and has achieved high accuracy. While regular SVMs try to find a decision hyperplane from two distinct classes of the input examples, FSVMs apply a fuzzy membership to each input example such that different examples can make different contributions to the decision hyperplane. For nonlinear classification problem, FSVMs can achieve good performance by reducing the effect of outliers. In this paper, a new fuzzy membership function is proposed to Japanese dependency analysis. We train an initial classifier with a small training set. The fuzzy membership is calculated by the distance from each input example to the initial hyperplane. In addition, we employ Nivre´s algorithm for Japanese dependency analysis since it parses a sentence in linear-time. Experiments using the Kyoto University Corpus show that the parser using Nivre´s algorithm outperforms the previous systems, and the proposed FSVMs improve the already excellent performance of SVMs for Japanese dependency analysis.