<?xml version="1.0" standalone="yes"?>
<Paper uid="H94-1034">
  <Title>Tagging Speech Repairs</Title>
  <Section position="11" start_page="191" end_page="191" type="concl">
    <SectionTitle>
9. Conclusions
</SectionTitle>
    <Paragraph position="0"> This paper described a method of detecting repairs that uses a part-of-speech tagger. Our work shows that a large percentage of speech repairs can be detected, and corrected prior to parsing. Prosodic clues can be easily incorporated into our statistical model, and we are currently investigating methods of automatically extracting simple prosodic features in order to further improve the performance of the algorithm.</Paragraph>
    <Paragraph position="1"> Our algorithm assumes that the speech recognizer produces a sequence of words and identifies the presertce of word fragments. With the exception of identifying fresh starts, all other processing is automatic and does not require additional hand-tailored transcription. We will be incorporating this method of detecting and correcting speech repairs into the next version of the TRAINS system, which will use spoken input.</Paragraph>
  </Section>
class="xml-element"></Paper>