<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-1610">
  <Title>Learning Domain-Specific Transfer Rules: An Experiment with Korean to English Translation</Title>
  <Section position="7" start_page="1" end_page="1" type="concl">
    <SectionTitle>
6 Conclusion and Future Work
</SectionTitle>
    <Paragraph position="0"> In this paper we have described the design of an MT system based on lexico-structural transfer rules induced from parsed bitexts. In a small scale experiment with Korean to English translation, we have demonstrated a substantial improvement over three baseline systems, including a nearly 20% improvement in the preference rate for our system over Babelfish (which was not trained on our corpus). Although our experimentation was aimed at Korean to English translation, we believe that our approach can be readily applied to other language pairs.</Paragraph>
    <Paragraph position="1"> It remains for future work to explore how well the approach would fare with a much larger training corpus. One foreseeable problem concerns the treatment of lengthy training sentences: since the number of transfer rule candidates generated grows exponentially with the size of the parse tree pairs, refinements will be necessary in order to make use of complex sentences; one option might be to automatically chunk longer sentences into smaller units.</Paragraph>
  </Section>
class="xml-element"></Paper>