<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0411">
  <Title>Preposition Semantic Classification via PENN TREEBANK and FRAMENET</Title>
  <Section position="8" start_page="3" end_page="3" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> Our approach to classifying prepositions according to the PENN TREEBANK annotations is fairly accurate (78.5% individually and 86.1% together), while retaining ability to generalize via class-based lexical associations. These annotations are suitable for  They target all of the TREEBANK function tags but give performance figures broken down by the groupings defined in the Treebank tagging guidelines. The baseline figure shown above is their recall figure for the 'baseline 2' performance. default classification of prepositions in case more fine-grained semantic role information cannot be determined. For the fine-grained FRAMENET roles, the performance is less accurate (70.3% individually and 49.4% together). In both cases, the best accuracy is achieved using a combination of standard word collocations along with class collocations in the form of WordNet hypernyms.</Paragraph>
    <Paragraph position="1"> Future work will address cross-dataset experiments. In particular, we will see whether the word and hypernym associations learned over FrameNet can be carried over into Treebank, given a mapping of the fine-grained FrameNet roles into the coarse-grained Treebank ones. Such a mapping would be similar to the one developed by Gildea and Jurafsky (2002).</Paragraph>
  </Section>
class="xml-element"></Paper>