<?xml version="1.0" standalone="yes"?>
<Paper uid="A00-2014">
  <Title>The Effectiveness of Corpus-Induced Dependency Grammars for Post-processing Speech*</Title>
  <Section position="6" start_page="107" end_page="107" type="concl">
    <SectionTitle>
5 Conclusion and Future Directions
</SectionTitle>
    <Paragraph position="0"> ity to improve sentence accuracy of our speech system. To achieve balance between precision and coverage of our corpus-induced grammars, we have expanded the RM sentences with templates for expressions like dates and times. The grammars extracted from these expanded sentences gave increased RM2 coverage without sacrificing even 1% of the sentence accuracy. We are currently expanding the number of templates in our grammar in an attempt to obtain full coverage of the RM2 corpus using only template-expanded RM sentences. We have recently added ten semantic templates to the grammar and have improved the coverage by 9.19% without losing any sentence accuracy. We are also developing a stochastic version of CDG that uses a statistical ARV, which is similar to a supertag (Srinivas, 1996).</Paragraph>
    <Paragraph position="1"> The ability to extract ARV/ARVP grammars with varying degrees of specificity provides us with the ability to rapidly develop a grammar with the abil-</Paragraph>
  </Section>
class="xml-element"></Paper>