<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1516">
  <Title>Strictly Lexical Dependency Parsing</Title>
  <Section position="8" start_page="157" end_page="157" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> To the best of our knowledge, all previous natural language parsers have to rely on part-of-speech tags. We presented a strictly lexicalized model for dependency parsing that only relies on word statistics. We compared our parser with an unlexicalized parser that employs the same probabilistic model except that the parameters are estimated using gold standard tags in the Chinese Treebank.</Paragraph>
    <Paragraph position="1"> Our experiments show that the strictly lexicalized parser significantly outperformed its unlexicalized counter-part.</Paragraph>
    <Paragraph position="2"> An important distinction between our statistical model from previous parsing models is that all the parameters in our model are conditional probability of binary variables. This allows us to take advantage of similarity-based smoothing, which has not been successfully applied to parsing before.</Paragraph>
  </Section>
class="xml-element"></Paper>