<?xml version="1.0" standalone="yes"?>
<Paper uid="H05-1105">
  <Title>Using the Web as an Implicit Training Set: Application to Structural Ambiguity Resolution</Title>
  <Section position="8" start_page="841" end_page="841" type="concl">
    <SectionTitle>
4 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> We have shown that simple unsupervised algorithms that make use of bigrams, surface features and paraphrases extracted from a very large corpus are effective for several structural ambiguity resolutions tasks, yielding results competitive with the best unsupervised results, and close to supervised results.</Paragraph>
    <Paragraph position="1"> The method does not require labeled training data, nor lexicons nor ontologies. We think this is a promising direction for a wide range of NLP tasks.</Paragraph>
    <Paragraph position="2"> In future work we intend to explore better-motivated evidence combination algorithms and to apply the approach to other NLP problems.</Paragraph>
    <Paragraph position="3"> Acknowledgements. This research was supported by NSF DBI-0317510 and a gift from Genentech.</Paragraph>
  </Section>
class="xml-element"></Paper>