<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1077">
  <Title>Tree-to-String Alignment Template for Statistical Machine Translation</Title>
  <Section position="9" start_page="614" end_page="614" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper, we introduce tree-to-string alignment templates, which can be automatically learned from syntactically-annotated training data.</Paragraph>
    <Paragraph position="1"> The TAT-based translation model improves translation quality significantly compared with a state-of-the-art phrase-based decoder. Treated as special TATs without tree on the source side, bilingual phrases can be utilized for the TAT-based model to get further improvement.</Paragraph>
    <Paragraph position="2"> It should be emphasized that the restrictions we impose on TAT extraction limit the expressive power of TAT. Preliminary experiments reveal that removing these restrictions does improve translation quality, but leads to large memory requirements. We feel that both parsing and word alignment qualities have important effects on the TAT-based model. We will retrain the Chinese parser on Penn Chinese Treebank version 5.0 and try to improve word alignment quality using log-linear models as suggested in (Liu et al., 2005).</Paragraph>
  </Section>
class="xml-element"></Paper>