<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3402">
  <Title>Off-Topic Detection in Conversational Telephone Speech</Title>
  <Section position="3" start_page="0" end_page="8" type="intro">
    <SectionTitle>
2 Linguistics of Conversational Speech
</SectionTitle>
    <Paragraph position="0"> Cheepen (Cheepen, 1988) posits that speakers have two primary goals in conversation: interactional goals in which interpersonal motives such as social rank and trust are primary; and transactional goals which focus on communicating useful information or getting a job done. In a context where conversations are indexed and searched for information, we assume in this paper that users will be interested in the communicated information, rather than the way in which participants interact. Therefore, we assume that utterances with primarily transactional purposes will be most important, while interactional utterances can be ignored.</Paragraph>
    <Paragraph position="1"> Greetings and partings are the most predictable  type of interactional speech. They consistently appear at the beginning and end of conversations and follow a fairly formulaic pattern of content (Laver, 1981). Thus we hypothesize that: Utterances near the beginning or end of conversations are less likely to be relevant.</Paragraph>
    <Paragraph position="2"> Cheepen also defines speech-in-action regions to be segments of conversation that are related to the present physical world or the activity of chatting, e.g. &amp;quot;What lovely weather.&amp;quot; or &amp;quot;It is so nice to see you.&amp;quot; Since these regions mainly involve participants identifying their shared social situation, they are not likely to contain transactional content. Further, since speech-in-action segments are distinguished by their focus on the present, we hypothesize that: Utterances with present tense verbs are less likely to be relevant.</Paragraph>
    <Paragraph position="3"> Finally, small talk that is not intended to demarcate social hierarchy tends to be abbreviated, e.g. &amp;quot;Nice day&amp;quot; (Laver, 1981). From this we hypothesize that: Utterances lacking common helper words such as &amp;quot;it&amp;quot;, &amp;quot;there&amp;quot;, and forms of &amp;quot;to be&amp;quot; are less likely to be relevant.</Paragraph>
  </Section>
class="xml-element"></Paper>