<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-1033">
  <Title>Synchronous Binarization for Machine Translation</Title>
  <Section position="6" start_page="262" end_page="262" type="evalu">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> Modeling reorderings between languages has been a major challenge for machine translation. This work shows that the majority of syntactic reorderings, at least between languages like English and Chinese, can be ef ciently decomposed into hierarchical binary reorderings. From a modeling perspective, on the other hand, it is bene cial to start with a richer representation that has more transformational power than ITG or binary SCFG. Our work shows how to convert it back to a computationally friendly form without harming much of its expressiveness. As a result, decoding with n-gram models can be fast and accurate, making it possible for our syntax-based system to overtake a comparable phrase-based system in BLEU score. We believe that extensions of our technique to more powerful models such as synchronous tree-adjoining grammar (Shieber and Schabes, 1990) is an interesting area for further work.</Paragraph>
  </Section>
class="xml-element"></Paper>