<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1190">
  <Title>Semi-Supervised Training of a Kernel PCA-Based Model for Word Sense Disambiguation</Title>
  <Section position="9" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have proposed a new composite semi-supervised WSD model based on the Kernel PCA technique, that employs both supervised and semi-supervised components. This strategy allows us to combine large amounts of cheap unlabeled data with smaller amounts of labeled data. Experiments on the hard-to-disambiguate verbs from the Senseval-2 English lexical sample task confirm that when the supervised KPCA model is insufficiently confident in its sense predictions, taking advantage of the semi-supervised KPCA model trained with the unlabeled data can help to give a better prediction. The composite semi-supervised KPCA model exploits this to improve upon the accuracy of the supervised KPCA model introduced by Wu et al. (2004).</Paragraph>
  </Section>
class="xml-element"></Paper>