<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-1016">
  <Title>Resolving PP attachment Ambiguities with Memory-Based Learning</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have shown that our MBL approach is very competent in solving attachment ambiguities; it achieves better generalization performance than many previous statistical approaches. Moreover, because we can measure the r'elevance of the features using an information gain metric (IBI-IG), we are able to add features without a high cost in model selection or an explosion in the number of parameters.</Paragraph>
    <Paragraph position="1"> An additional advantage of the MBL approach is that, in contrast to the other statistical approaches, it is founded in the use of similarity-based reasoning.</Paragraph>
    <Paragraph position="2"> Therefore, it makes it possible to experiment with different types of distributed non-symbolic lexical representations extracted from corpora using unsupervised learning. This promises to be a rich source of extra information. We have also shown that task specific similarity metrics such as MVDM are sensitive to the sparse data problem. LexSpace is less sensitive to this problem because of the large amount of data which is available for its training.</Paragraph>
  </Section>
class="xml-element"></Paper>