<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1314">
  <Title>Automatically Detecting Action Items in Audio Meeting Recordings</Title>
  <Section position="4" start_page="0" end_page="96" type="intro">
    <SectionTitle>
2 Related work
</SectionTitle>
    <Paragraph position="0"> Multi-party meetings have attracted a significant amount of recent research attention. The creation of the ICSI corpus (Janin et al., 2003), comprised of 72 hours of meeting recordings with an average of 6 speakers per meeting, with associated transcripts, has spurred further annotations for various types of information, including dialog acts (Shriberg et al., 2004), topic hierarchies and action items (Gruenstein et al., 2005), and &amp;quot;hot spots&amp;quot; (Wrede and Shriberg, 2003).</Paragraph>
    <Paragraph position="1"> The classification of individual utterances based on their role in the dialog, i.e. as opposed to their semantic payload, has a long history, especially in the context of dialog act (DA) classification.</Paragraph>
    <Paragraph position="2">  Research on DA classification initially focused on two-party conversational speech (Mast et al., 1996; Stolcke et al., 1998; Shriberg et al., 1998) and, more recently, has extended to multi-party audio recordings like the ICSI corpus (Shriberg et al., 2004). Machine learning techniques such as graphical models (Ji and Bilmes, 2005), maximum entropy models (Ang et al., 2005), and hidden Markov models (Zimmermann et al., 2005) have been used to classify utterances from multi-party conversations.</Paragraph>
    <Paragraph position="3"> It is only more recently that work focused specifically on action items themselves has been developed. SVMs have been successfully applied to the task of extracting action items from email messages (Bennett and Carbonell, 2005; Corston-Oliver et al., 2004). Bennett and Carbonell, in particular, distinguish the task of action item detection in email from the more well-studied task of text classification, noting the finer granularity of the action item task and the difference of semantics vs. intent. (Although recent work has begun to blur this latter division, e.g. Cohen et al. (2004).) In the audio domain, annotations for action item utterances on several recorded meeting corpora, including the ICSI corpus, have recently become available (Gruenstein et al., 2005), enabling work on this topic.</Paragraph>
  </Section>
class="xml-element"></Paper>