<?xml version="1.0" standalone="yes"?>
<Paper uid="P03-1008">
  <Title>Syntactic Features and Word Similarity for Supervised Metonymy Resolution</Title>
  <Section position="8" start_page="13" end_page="13" type="concl">
    <SectionTitle>
8 Conclusions
</SectionTitle>
    <Paragraph position="0"> We presented a supervised classification algorithm for metonymy recognition, which exploits the similarity between examples of conventional metonymy, operates on semantic classes and thereby enables complex inferences from training to test examples.</Paragraph>
    <Paragraph position="1"> We showed that syntactic head-modifier relations are a high precision feature for metonymy recognition. However, basing inferences only on the lexical heads seen in the training data leads to data sparseness due to the large number of different lexical heads encountered in natural language texts. In order to overcome this problem we have integrated a thesaurus that allows us to draw inferences be- null Incorporating knowledge about particular PMWs (e.g., as a prior) will probably improve performance, as word idiosyncracies -- which can still exist even when treating regular sense distinctions -- could be accounted for. In addition, knowledge about the individual word is necessary to assign its original semantic class.</Paragraph>
    <Paragraph position="2"> tween examples with similar but not identical lexical heads. We also explored the use of simpler grammatical role features that allow further generalisations. The results show a substantial increase in precision, recall and F-measure. In the future, we will experiment with combining grammatical features and local/topical cooccurrences. The use of semantic classes and lexical head similarity generalises over two levels of contextual similarity, which exceeds the complexity of inferences undertaken in standard supervised word sense disambiguation.</Paragraph>
    <Paragraph position="3"> Acknowledgements. The research reported in this paper was supported by ESRC Grant R000239444.</Paragraph>
    <Paragraph position="4"> Katja Markert is funded by an Emmy Noether Fellowship of the Deutsche Forschungsgemeinschaft (DFG). We thank three anonymous reviewers for their comments and suggestions.</Paragraph>
  </Section>
class="xml-element"></Paper>