<?xml version="1.0" standalone="yes"?>
<Paper uid="W96-0107">
  <Title>Automatic Extraction of Word Sequence Correspondences in Parallel Corpora</Title>
  <Section position="8" start_page="86" end_page="86" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> A method for obtaining translation dictionary from parallel corpora was proposed, in which not only word-word correspondences but arbitrary length word sequence correspondences are extracted.</Paragraph>
    <Paragraph position="1"> This work is originally motivated for the purpose of improving the performance of our translation pattern extraction from parallel corpora \[Kitamura &amp; Matsumoto 95\], in which translation patterns are extracted by syntactically analyzing both Japanese and English sentences and by structurally matching them. Some discrepancy is caused by poor quality of translation dictionary. This is why we tried to pursue a way to obtain better translation dictionary from parallel corpora. We believe that the proposed method gives results of good performance compared with previous related work.</Paragraph>
    <Paragraph position="2"> The translation pairs obtained through our method are directly usable as the base resource for MT systems based on translation memory \[Lehmann 95\].</Paragraph>
    <Paragraph position="3"> We hope to acquire better translation patterns by combining the current results with our work of structural matching for finding out fine grained correspondence.</Paragraph>
  </Section>
class="xml-element"></Paper>