<?xml version="1.0" standalone="yes"?>
<Paper uid="A97-1003">
  <Title>High Performance Segmentation of Spontaneous Speech Using Part of Speech and Trigger Word Information</Title>
  <Section position="9" start_page="14" end_page="15" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have shown that using neural networks for automatically segmenting turns in conversational speech into small clauses reaches a level of less than 5% error rate and achieves good precision/recall performance as measured by an F-score of more than .85.</Paragraph>
    <Paragraph position="1"> These results outperform those obtained by other methods as reported in the literature.</Paragraph>
    <Paragraph position="2">  Future work on this problem includes issues such as optimizing the set of POS tags, adding acoustic/prosodic features to the neural network, and using it for pro-drop languages like Spanish to assess the relative importance of POS vs. trigger word weights and to examine the performance of the system for languages where POS tags may not be as informative as they are for English.</Paragraph>
  </Section>
class="xml-element"></Paper>