<?xml version="1.0" standalone="yes"?>
<Paper uid="P94-1004">
  <Title>HIDDEN UNDERSTANDING MODELS OF NATURAL LANGUAGE</Title>
  <Section position="10" start_page="30" end_page="31" type="concl">
    <SectionTitle>
8 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have demonstrated the possibility of automatically learning semantic representations directly from a training corpus through the application of statistical techniques.</Paragraph>
    <Paragraph position="1"> Empirical results, including the results of an ARPA  evaluation, indicate that these techniques are capable of relatively high levels of performance.</Paragraph>
    <Paragraph position="2"> While hidden understanding models are based primarily on the concepts of hidden Markov models, we have also shown their relationship to other work in stochastic grammars and probabilistic parsing.</Paragraph>
    <Paragraph position="3"> Finally, we have noted some limitations to our current approach. We view each of these limitations as opportunities for fta~er research and exploration.</Paragraph>
  </Section>
class="xml-element"></Paper>