<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-2022">
  <Title>Automatic Recognition of Personality in Conversation</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> It is well known that utterances convey information about the speaker in addition to their semantic content. One such type of information consists of cues to the speaker's personality traits, typically assessed along five dimensions known as the Big Five (Norman, 1963):  Findings include that extraverts talk more, louder, and faster, with fewer pauses and hesitations, and more informal language (Scherer, 1979; Furnham, 1990; Heylighen and Dewaele, 2002; Gill and Oberlander, 2002). Neurotics use more 1 st person singular pronouns and negative emotion words, while conscientious people avoid negations and negative emotion words (Pennebaker and King, 1999). The use of words related to insight and the avoidance of past tense indicate intellect, and swearing and negative emotion words mark disagreeableness. Correlations are higher in spoken language, possibly especially in informal conversation (Mehl et al., in press).</Paragraph>
    <Paragraph position="1"> Previous work has modeled emotion and personality in virtual agents, and classified emotions from actor's speech (Andr'e et al., 1999; Liscombe et al., 2003). However, to our knowledge no one has tested whether it is possible to automatically recognize personality from conversation extracts of unseen subjects. Our hypothesis is that automatic analysis of conversation to detect personality has application in a wide range of language processing domains.</Paragraph>
    <Paragraph position="2"> Identification of leaders using personality dimensions could be useful in analyzing meetings and the conversations of suspected terrorists (Hogan et al., 1994; Tucker and Whittaker, 2004; Nunn, 2005).</Paragraph>
    <Paragraph position="3"> Dating websites could analyze text messages to try to match personalities and increase the chances of a successful relationship (Donnellan et al., 2004). Dialogue systems could adapt to the user's personality, like humans do (Reeves and Nass, 1996; Funder and Sneed, 1993). This work is a first step toward individual adaptation in dialogue systems.</Paragraph>
    <Paragraph position="4"> We present non-linear statistical models for ranking utterances based on the Big Five personality traits. Results show that the models perform significantly better than a random baseline, and that prosodic features are good indicators of extraversion. A qualitative analysis confirms previous findings linking language and personality, while revealing many new linguistic markers.</Paragraph>
  </Section>
class="xml-element"></Paper>