<?xml version="1.0" standalone="yes"?>
<Paper uid="P99-1053">
  <Title>Integrating multiple knowledge sources</Title>
  <Section position="7" start_page="418" end_page="419" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> Traditionally, dialog has been treated as a series of single speaker utterances, with no systematic allowance for speech repairs and editing terms. Such a treatment cannot adequately deal with dialogs involving more than one human (as appear in machine translation or meeting analysis), and will not allow single user dialog systems to progress to more natural interactions. The simple set of rules given here allows speakers to collaborate to form utterances and prevents an interruption such as a backchannel response from disrupting the syntax of another speaker's utterance. Speech repairs are captured by parallel phrase structure trees, and editing terms are represented as separate utterances occurring inside other utterances.</Paragraph>
    <Paragraph position="1"> Since the parser has knowledge of grammar and the syntactic structure of the input, it can boost speech repair identification performance. In the experiments of this paper, the parser was able to increase the recall of a pre-parser speech identifier by 4.8%. Another advantage of giving speech repair information to the parser is that the parser can then include reparanda in its output and a truer picture of dialog structure can be formed. This can be crucial if a pronoun antecedent is present in the reparandum as in have the engine take the oranges to Elmira, urn, I mean, take them to Coming. In addition, this information can help a dialog system detect uncertainty and planning difficultly in speakers.</Paragraph>
    <Paragraph position="2"> The framework presented here is sufficient to describe the 3441 human-human utterances comprising the chosen set of TRAINS dialogs. More corpus investigation is necessary before we can claim the framework provides broad coverage of human-human dialog. Another necessary test of the framework is extension to dialogs involving more than two speakers.</Paragraph>
    <Paragraph position="3"> Long term goals include further investigation into the TRAINS corpus and attempting full dialog analysis rather than experimenting with small groups of overlapping utterances. Another long term goal is to weigh the current framework against a purely robust parsing approach (Ros~ and Levin, 1998), (Lavie, 1995) that treats out of vocabulary/grammar phenomena in the same way as editing terms and speech repairs. Robust parsing is critical to a parser  such as the one described here which has a coverage of only 62% on fluent utterances.</Paragraph>
    <Paragraph position="4"> In our corpus, the speech repair to utterance ratio is 14%. Thus, problems due to the coverage of the grammar are more than twice as likely as speech repairs. However, speech repairs occur with enough frequency to warrant separate attention. Unlike grammar failures, repairs are generally signaled not only by ungrammaticality, but also by pauses, editing terms, parallelism, etc.; thus an approach specific to speech repairs should perform better than just using a robust parsing algorithm to deal with them.</Paragraph>
  </Section>
class="xml-element"></Paper>