<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-1526">
  <Title>Online Statistics for a Unification-Based Dialogue Parser</Title>
  <Section position="5" start_page="0" end_page="198" type="ackno">
    <SectionTitle>
7 words.
</SectionTitle>
    <Paragraph position="0"> Since our motivation for unification parsing is to reveal semantics as well as syntax, we next evaluated Aug-TRIPS's production of correct interpretations at the sentence level, which require complete correctness not only of the bracketing structure but of the sense chosen for each word and the thematic  roles of each argument (Tetreault et al., 2004).</Paragraph>
    <Paragraph position="1"> For this task, we modified the probability model to condition on the senses in our lexicon rather than words. For instance, the words &amp;quot;two thousand dollars&amp;quot; are replaced with the senses &amp;quot;number numberunit money-unit&amp;quot;. This allows us to model lexical disambiguation explicitly. The model generates one or more senses from each word with probability P(sense|word,tag), and then uses sense statistics rather than word statistics in all other calculations.</Paragraph>
    <Paragraph position="2"> Similar but more complex models were used in the PCFG-sem model of Toutanova et al. (2005) and using WordNet senses in Bikel (2000).</Paragraph>
    <Paragraph position="3"> We used the Projector dialogues (835 sentences), which concern purchasing video projectors. In this domain, Aug-TRIPS makes about 10% more interpretation errors than TRIPS (Table 3), but when parsing sentences on which TRIPS itself makes errors, it can correct about 10% (Table 4).</Paragraph>
    <Paragraph position="4">  Our parser makes substantially fewer constituents than baseline TRIPS at only slightly lower accuracy. Tsuruoka et al. (2004) achieved a much higher speedup (30 times) than we did; this is partly due to their use of the Penn Treebank, which contains much more data than our corpora. In addition, however, their baseline system is a classic HPSG parser with no efficiency features, while our baseline, TRIPS, is designed as a real-time dialogue parser which uses hand-tuned weights to guide its search and imposes a maximum chart size.</Paragraph>
    <Paragraph position="5"> Acknowledgements Our thanks to Will DeBeaumont and four anonymous reviewers.</Paragraph>
  </Section>
class="xml-element"></Paper>