<?xml version="1.0" standalone="yes"?>
<Paper uid="N04-4018">
  <Title>Improving Automatic Sentence Boundary Detection with Confusion Networks</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> Detecting sentence structure in automatic speech recognition provides important information for language processing or human understanding. Incorporating multiple hypotheses from word recognition output can improve overall detection of SUs in comparison to prediction on a single hypothesis. This is especially true for CTS, which suffers more from word errors and can therefore benefit from considering alternative hypotheses.</Paragraph>
    <Paragraph position="1"> Future work will involve a tighter integration of SU detection and word recognition by including SU events directly in the recognition lattice. This will provide opportunities to investigate the interaction of automatic word recognition and structural metadata, hopefully resulting in reduced WER. We also plan to extend these methods to additional tasks such as disfluency detection.</Paragraph>
  </Section>
class="xml-element"></Paper>