<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-1303">
  <Title>Japanese Dependency Structure Analysis Based on Support Vector Machines</Title>
  <Section position="2" start_page="0" end_page="18" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Dependency structure analysis has been recognized as a basic technique in Japanese sentence analysis, and a number of studies have been proposed for years. Japanese dependency structure is usually defined in terms of the relationship between phrasal units called 'bunsetsu' segments (hereafter &amp;quot;chunks~). Generally, dependency structure analysis consists of two steps. In the first step, dependency matrix is constructed, in which each element corresponds to a pair of chunks and represents the probability of a dependency relation between them. The second step is to find the optimal combination of dependencies to form the entire sentence.</Paragraph>
    <Paragraph position="1"> In previous approaches, these probabilites of dependencies axe given by manually constructed rules. However, rule-based approaches have problems in coverage and consistency, since there are a number of features that affect the accuracy of the final results, and these features usually relate to one another. null On the other hand, as large-scale tagged corpora have become available these days, a number of statistical parsing techniques which estimate the dependency probabilities using such tagged corpora have been developed(Collins, 1996; Fujio and Matsumoto, 1998). These approaches have overcome the systems based on the rule-based approaches.</Paragraph>
    <Paragraph position="2"> Decision Trees(Haruno et al., 1998) and Maximum Entropy models(Ratnaparkhi, 1997; Uchimoto et al., 1999; Charniak, 2000) have been applied to dependency or syntactic structure analysis. However, these models require an appropriate feature selection in order to achieve a high performance. In addition, acquisition of an efficient combination of features is difficult in these models.</Paragraph>
    <Paragraph position="3"> In recent years, new statistical learning techniques such as Support Vector Machines (SVMs) (Cortes and Vapnik, 1995; Vapnik, 1998) and Boosting(Freund and Schapire, 1996) are proposed. These techniques take a strategy that maximize the margin between critical examples and the separating hyperplane. In particular, compared with other conventional statistical learning algorithms, SVMs achieve high generalization even with training data of a very high dimension. Furthermore, by optimizing the Kernel function, SVMs can handle non-linear feature spaces, and carry out the training with considering combinations of more than one feature.</Paragraph>
    <Paragraph position="4"> Thanks to such predominant nature, SVMs deliver state-of-the-art performance in real-world applications such as recognition of hand-written letters, or of three dimensional images. In the field of natural language processing, SVMs are also applied to text categorization, and are reported to have achieved  high accuracy without falling into over-fitting even with a large number of words taken as the features (Joachims, 1998; Taira and Haruno, 1999).</Paragraph>
    <Paragraph position="5"> In this paper, we propose an application of SVMs to Japanese dependency structure analysis. We use the features that have been studied in conventional statistical dependency analysis with a little modification on them.</Paragraph>
  </Section>
class="xml-element"></Paper>