<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3104">
  <Title>Quasi-Synchronous Grammars: Alignment by Soft Projection of Syntactic Dependencies</Title>
  <Section position="6" start_page="28" end_page="29" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> With quasi-synchronous grammars, we have presented a new approach to syntactic MT: constructing a monolingual target-language grammar that describes the aligned translations of a source-language sentence. We described a simple parameterization 6For German - English MT, one would use a German English QCFG as above, but an English - German channel model. In this arguably inappropriate comparison, Figure 4 shows, the Model 4 channel model produces slightly better word alignments than the QG.</Paragraph>
    <Paragraph position="1">  ages). The QCFG consistently beat one GIZA++ model and was close to the other.</Paragraph>
    <Paragraph position="2"> with gradually increasing syntactic domains of locality, and estimated those parameters on German-English bitext.</Paragraph>
    <Paragraph position="3"> The QG formalism admits many more nuanced options for features than we have exploited. In particular, we now are exploring log-linear QGs that score overlapping elementary trees of T2 while considering the syntactic configuration and lexical content of the T1 nodes to which each elementary tree aligns.</Paragraph>
    <Paragraph position="4"> Even simple QGs, however, turned out to do quite well. Our evaluation on a German-English word-alignment task showed them to be competitive with IBM model 4--consistently beating the German-English direction by several percentage points of alignment error rate and within 1% AER of the English-German direction. In particular, alignment accuracy benefited from allowing syntactic breakages between the two dependency structures.</Paragraph>
    <Paragraph position="5"> We are also working on a translation decoding using QG. Our first system uses the QG to find optimal T2 aligned to T1 and then extracts a synchronous tree-substitution grammar from the aligned trees.</Paragraph>
    <Paragraph position="6"> Our second system searches a target-language vocabulary for the optimal T2 given the input T1.</Paragraph>
  </Section>
class="xml-element"></Paper>