<?xml version="1.0" standalone="yes"?>
<Paper uid="H01-1028">
  <Title>Finding Errors Automatically in Semantically Tagged Dialogues</Title>
  <Section position="9" start_page="0" end_page="0" type="relat">
    <SectionTitle>
6. PREVIOUS WORK
</SectionTitle>
    <Paragraph position="0"> In Hirschman &amp; Pao [5], annotation was done by manual inspection of the exchanges in the dialogue. Each exchange was evaluated based on the portion of information &amp;quot;visible to the other party&amp;quot;. Errors and problems were identified manually and traced back to their point of origin. This is quite similar to our baseline manual annotation described in section 3.</Paragraph>
    <Paragraph position="1"> There have been other approaches to detecting and characterizing errors in HC dialogues. Danieli [2] used expectations to model future user utterances, and Levow [6][7] used utterance and pause duration, as well as pitch variability to characterize errors and corrections. Dybkjaer, Bernsen &amp; Dybkjaer [4] developed a set of principles of cooperative HC dialogue, as well as a taxonomy of errors typed according to which of the principles are violated. Finally, Walker et. al.</Paragraph>
    <Paragraph position="2"> [11][12] have trained an automatic classifier that identifies and predicts problems in HC dialogues.</Paragraph>
  </Section>
class="xml-element"></Paper>