<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3506">
  <Title>Catching Metaphors</Title>
  <Section position="9" start_page="46" end_page="47" type="concl">
    <SectionTitle>
8 Conclusion
</SectionTitle>
    <Paragraph position="0"> Metaphors are a ubiquitous phenomenon in language, and our corpus analysis clearly bears this out. It is somewhat gratifying that with a judicious combination of the available wide-coverage resources (WordNet, FrameNet, PropBank) we were able to build classifiers that could outperform the baseline even in the most skewed cases. Our results show the utility of our approach and more generally the maturity of the current NLP technology to make progress in attacking the challenging and important problem of interpreting figurative language.</Paragraph>
    <Paragraph position="1"> However, this is only the first step. As with all semantic extraction methods and technologies, the proof of utility is not in how good the extractor is but how much it helps in an actual task. As far as we can tell, this problem remains open for the entire semantic parsing/role labeling/extraction field despite the flurry of activity in the last four years. In the case of metaphor interpretation, we have some initial encouragement from the results published by (Narayanan, 1997) and others.</Paragraph>
    <Paragraph position="2">  Our classifier relies on PropBank senses, so we can use the high performance classifiers available for PropBank. The price is that we have to constructmappingsfrom FrameNetframestoPropBank senses. However, this is a one-time effort pursued by many groups, so this should not present a problem to extending our approach to cover all frames and metaphors. Additionally, we are in the process of linking the metaphor detector to a metaphor inferencesystem. Wehopetohaveinitialresultstoreport on by conference time.</Paragraph>
  </Section>
class="xml-element"></Paper>