<?xml version="1.0" standalone="yes"?>
<Paper uid="P03-1022">
  <Title>A Machine Learning Approach to Pronoun Resolution in Spoken Dialogue</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> We presented a machine learning approach to pronoun resolution in spoken dialogue. We built upon a system we used for anaphora resolution in written text and extended it with a set of features designed for spoken dialogue. We refined distance features and used metrics from information retrieval for determining non-NP-antecedents. Inspired by the more linguistically oriented work by Eckert &amp; Strube (2000) and Byron (2002) we also evaluated the contribution of features which used the predicative context of the pronoun to be resolved. However, these features did not show up in the final models since they did not lead to an improvement. Instead, rather simple distance metrics were preferred. While we were (almost) satisfied with the performance of these features, the major problem for a spoken dialogue pronoun resolution algorithm is the abundance of pronouns without antecedents. Previous research could avoid dealing with this phenomenon by either applying the algorithm by hand (Eckert &amp; Strube, 2000) or excluding these cases (Byron, 2002) from the evaluation. Because we included these cases in our evaluation we consider our approach at least comparable to Byron's system when she uses only domain-independent semantics. We believe that our system is more robust than hers and that it can more easily be ported to new domains.</Paragraph>
    <Paragraph position="1"> Acknowledgements. The work presented here has been partially funded by the German Ministry of Research and Technology as part of the EMBASSI project (01 IL 904 D/2) and by the Klaus Tschira Foundation. We would like to thank Susanne Wilhelm and Lutz Wind for doing the annotations, Kerstin Sch&amp;quot;urmann, Torben Pastuch and Klaus Rothenh&amp;quot;ausler for helping with the data preparation. null</Paragraph>
  </Section>
class="xml-element"></Paper>