<?xml version="1.0" standalone="yes"?>
<Paper uid="E99-1026">
  <Title>Japanese Dependency Structure Analysis Based on Maximum Entropy Models</Title>
  <Section position="5" start_page="201" end_page="201" type="concl">
    <SectionTitle>
4 Conclusion
</SectionTitle>
    <Paragraph position="0"> This paper described a Japanese dependency structure analysis based on the maximum entropy model. Our model is created by learning the weights of some features from a training corpus to predict the dependency between bunsetsus or phrasal units. The probabilities of dependencies between bunsetsus are estimated by this model. The dependency accuracy of our system was 87.2% using the Kyoto University corpus.</Paragraph>
    <Paragraph position="1"> In our experiments without the feature sets shown in Tables 1 and 2, we found that some basic and combined features strongly contribute to improve the accuracy. Investigating the relationship between the number of training data and the accuracy, we found that good accuracy can be achieved even with a very small set of training data. We believe that the maximum entropy framework has suitable characteristics for overcoming the data sparseness problem.</Paragraph>
    <Paragraph position="2"> There are several future directions. In particular, we are interested in how to deal with coordinate structures, since that seems to be the largest problem at the moment.</Paragraph>
  </Section>
class="xml-element"></Paper>