<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0607">
  <Title>Manual Annotation of Opinion Categories in Meetings</Title>
  <Section position="9" start_page="59" end_page="60" type="concl">
    <SectionTitle>
6 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> In this paper we performed an annotation study of opinions in meetings, and investigated the effects of speech. We have shown that it is possible to reliably detect opinions within multi-party conversations. Our consistently better agreement results with text+speech input over text-only input suggest that speech is a reliable indicator of opinions. We have also found that Annotator Uncertainty decreased with speech input. Our results also show that speech is a more informative indicator for negative versus positive categories. We hypothesize that this is due to the fact the people express their positive attitudes more explicitly. The speech signal is thus even more important for discerning negative opinions.</Paragraph>
    <Paragraph position="1"> This experience has also helped us gain insights to the ambiguities that arise due to sarcasm and humor.</Paragraph>
    <Paragraph position="2"> Our promising results open many new avenues for research. It will be interesting to see how our categories relate to other discourse structures, both at the shallow level (agreement/disagreement) as well as at the deeper level  (intentions/goals). It will also be interesting to investigate how other forms of subjectivity like speculation and intention are expressed in multi-party discourse. Finding prosodic correlates of speech as well as lexical clues that help in opinion detection would be useful in building subjectivity detection applications for multiparty meetings. null</Paragraph>
  </Section>
class="xml-element"></Paper>