<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-2023">
  <Title>Summarizing Speech Without Text Using Hidden Markov Models</Title>
  <Section position="6" start_page="91" end_page="91" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have shown a novel way of using continuous HMMs for summarizing speech documents without using any lexical information. Our model generates an optimal summary by decoding the state lattice, where states represent whether a sentence should be included in the summary or not. This model is able to take the context and the previous decisions into account generating better summaries. Our results also show that speech can be summarized fairly well using acoustic/prosodic features alone, without lexical features, suggesting that the effect of ASR transcription errors on summarization may be minimized by techniques such as ours.</Paragraph>
  </Section>
class="xml-element"></Paper>