<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-2022">
  <Title>Automatic Recognition of Personality in Conversation</Title>
  <Section position="5" start_page="86" end_page="87" type="concl">
    <SectionTitle>
4 Conclusion
</SectionTitle>
    <Paragraph position="0"> We showed that personality can be recognized automatically in conversation. To our knowledge, this is the first report of experiments testing trained models on unseen subjects. There are models for each dimension that perform significantly better than the baseline. Combinations of these models may be useful to identify important personality types in different NLP applications, e.g. a combination of extraversion, emotional stability and intellect indicates leadership, while low intellect, extraversion and agreeableness are correlated with perceptions of trustworthiness.</Paragraph>
    <Paragraph position="1"> One limitation for applications involving speech recognition is that recognition errors will introduce noise in all features except prosodic features, and prosodic features on their own are only effective in the extraversion model. However, our data set is relatively small (96 subjects) so we expect that more  increase, while rows 11-20 indicate evidence for the other end of the scale, e.g. introversion. training data would improve model accuracies and might also make additional features useful. In future work, we plan to integrate these models in a dialogue system to adapt the system's language generation; we will then be able to test whether the accuracies we achieve are sufficient and explore methods for improving them.</Paragraph>
  </Section>
class="xml-element"></Paper>