<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-1021">
  <Title>Multilingual Dependency Parsing using Bayes Point Machines</Title>
  <Section position="8" start_page="165" end_page="166" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have successfully replicated the state-of-the-art results for dependency parsing (McDonald et al., 2005a) for both Czech and English, using Bayes Point Machines. Bayes Point Machines have the appealing property of simplicity, yet are competitive with online wide margin methods.</Paragraph>
    <Paragraph position="1"> We have also presented first results for dependency parsing of Arabic and Chinese, together with some analysis of the performance on those languages. null In future work we intend to explore the discriminative reranking of n-best lists produced by these parsers and the incorporation of morphological features. null  sizes. Graph shows average of five samples at each size and measures accuracy against the development test set.</Paragraph>
  </Section>
class="xml-element"></Paper>