<?xml version="1.0" standalone="yes"?>
<Paper uid="P04-1086">
  <Title>Using Conditional Random Fields to Predict Pitch Accents in Conversational Speech</Title>
  <Section position="9" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We used CRFs with new measures of collocational strength and new phonological factors that capture aspects of rhythm and timing to model pitch accent prediction. CRFs have the theoretical advantage of incorporating all these factors in a principled and efficient way. We demonstrated that CRFs outperform HMMs also experimentally. We also demonstrated the usefulness of some new probabilistic variables and phonological variables. Our results mainly have implications for the textual prediction of accents in TTS applications, but might also be useful in automatic speech recognition tasks such as automatic transcription of multi-speaker meetings. In the near future we would like to incorporate reliable acoustic information, controlling for individual speaker difference and also apply different discriminative sequence labeling techniques to pitch accent prediction task.</Paragraph>
  </Section>
class="xml-element"></Paper>