<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-1611">
  <Title>Paraphrasing Japanese noun phrases using character-based indexing</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions and future work
</SectionTitle>
    <Paragraph position="0"> This paper proposed a novel approach to extract paraphrases of a Japanese noun phrase from a corpus. The proposed method adopts both information retrieval techniques and natural language processing techniques. Unlike past research, the proposed method uses Kanzi (ideograms) characters as index terms and retrieves paraphrase candidates in a set of passages. The retrieved candidates are then filtered out based on syntactic and semantic constraints.</Paragraph>
    <Paragraph position="1"> The method was evaluated by a test set of 53 noun phrases, and paraphrases were extracted for 46 cases.</Paragraph>
    <Paragraph position="2"> These paraphrases were evaluated subjectively by three independent judges. The quantitative evaluation suggests that the performance needs to be further improved for fully automatic query expansion in information retrieval, but is usable to help users modify their queries by suggesting possible paraphrases.</Paragraph>
    <Paragraph position="3"> From a qualitative point of view, the proposed method could extract paraphrases which cannot be obtained by previous approaches such as syntactic transformation and word substitution. Considering characteristics of Japanese word formation by using character-based indexing enables us to obtain novel paraphrases.</Paragraph>
    <Paragraph position="4"> The performance of the current system needs to be improved for fully automatic paraphrasing. One direction is introducing more precise filtering criteria. The current system adopts only dependency analysis of bunsetu. We need case analysis as well, to capture relations among the bunsetu. Integrating syntactic transformation into the proposed method is another research direction to explore.</Paragraph>
    <Paragraph position="5"> In this paper, we evaluated output paraphrases subjectively. Task oriented evaluation should be also conducted. For example, effectiveness of phrase expansion in information retrieval systems should be investigated.</Paragraph>
    <Paragraph position="6"> Q Len. #Para. (1) (2-a) (2-b) (2-c) (3) Acc. Prec.</Paragraph>
  </Section>
class="xml-element"></Paper>