<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1079">
  <Title>Generating Overview Summaries of Ongoing Email Thread Discussions</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Background: Email Threads
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.1 Email Discussions supporting a Decision-
Making Process
</SectionTitle>
      <Paragraph position="0"> The focus of this paper is on email discussions supporting a group decision-making process. In contrast to studies on individual email usage (for an overview see: Ducheneaut and Bellotti, 2001), this research area has been less explored.</Paragraph>
      <Paragraph position="1"> Occasionally, such discussions end with an online vote. However, Ducheneaut and Belotti do note that voting is relatively infrequent and our own experience with our email corpora tends to support this.</Paragraph>
      <Paragraph position="2"> In general, we expect that these threads contain supporting discussions, and the actual decision might occur outside of the email medium, for example in a board meeting. What we hope to observe is that, for some issue discussed, candidate solutions and responses highlighting the pros and cons of a solution are introduced via email.</Paragraph>
      <Paragraph position="3"> Decision-making discussion threads occur frequently enough in environments which depend on professional usage of email. In the corpus we examined, 40% of the threads were decision-making discussions.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.2 Constraints on and Choice of a Corpus of
Email Discussions
</SectionTitle>
      <Paragraph position="0"> To collect a corpus of these threads, we placed a few constraints on the mailing list archives we found online.</Paragraph>
      <Paragraph position="1"> To begin with, we focused on threads from mailing lists that were set up to support organization activities as these often involve decision-making processes. As we are also interested in examining the role of dialogue, we required access to the email thread structure from which we can infer a basic dialogue structure.</Paragraph>
      <Paragraph position="2"> We chose to use the archives of the Columbia University ACM Student Chapter Committee as this group has organized several events and used email as their primary mode of communication outside of meetings. For practical reasons, it was relatively straightforward to obtain the necessary permissions to use the data, something that might be more difficult for other archives. Possible alternative corpora might be the mailing lists of organizing committees, for example that of a conference organizing committee or a steering group. Project-based mailing lists might also be potentially used, especially if the group participants have sufficient shared background to engage in discussions.</Paragraph>
    </Section>
    <Section position="3" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.3 Observations on Thread Structure
The Columbia University ACM Student Chapter
</SectionTitle>
      <Paragraph position="0"> Committee was made up of about 10 people. Upon initial examination of the data, we found that we could classify the threads of email according to its purpose. The set of group tasks facilitated by the email correspondence were: decision-making, information provision, requests for action and social conversation.</Paragraph>
      <Paragraph position="1"> However, it is natural for the group to engage in multiple tasks. Thus, we use the term &amp;quot;task shift&amp;quot; to refer to adjacent segments of the thread (comprised of emails) which reflect distinct group goals. In the corpus we use, we observe that these tasks usually occur sequentially. In some cases, a single email proposes more than one issue for discussion, and subsequent responses address each of these in turn.</Paragraph>
      <Paragraph position="2"> Intuitively, it makes sense to create a summary for a single task. Accordingly, we have designed our algorithm to accept only dialogue structures addressing a single group task. If discussions invoke short clarification questions, these should not be treated differently if the task remains the same. One supporting reason for this is the syntactic variation with which participants express disagreement. We have observed that disagreement is often expressed as a clarification question, or as a question which offers an alternative suggestion.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>