<?xml version="1.0" standalone="yes"?>
<Paper uid="P99-1026">
  <Title>Understanding Unsegmented User Utterances in Real-Time Spoken Dialogue Systems</Title>
  <Section position="3" start_page="200" end_page="200" type="intro">
    <SectionTitle>
2 Problem
</SectionTitle>
    <Paragraph position="0"> A dilemma is addressed in this paper. First, it is difficult to identify utterance boundaries in spontaneous speech in real time using only pauses. Observation of human-human dialogues reveals that humans often put pauses in utterances and sometimes do not put pauses at utterance boundaries. The following human utterance shows where pauses might appear in an utterance.</Paragraph>
    <Paragraph position="1"> I'd like to make a reservation for a conference room (pause) for, uh (pause) this afternoon (pause) at about (pause) say (pause) 2 or 3 o'clock (pause) for (pause)</Paragraph>
  </Section>
class="xml-element"></Paper>