<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1601">
  <Title>Unsupervised Discovery of a Statistical Verb Lexicon</Title>
  <Section position="9" start_page="7" end_page="7" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have demonstrated that it is possible to learn a statistical model of verb semantic argument structure directly from unannotated text. More work needs to be done to resolve particular classes of errors; for example, the one reported above for the verb work. It is perhaps understandable that the dependents occurring in the obliques with and for are put in the same role (the head words should refer to people), but it is harder to accept that dependents occurring in the oblique on are also grouped into the same role (the head words of these should refer to tasks). It seems plausible that measures to combat word sparsity might help to differentiate these roles: backing-off to word classes, or even just training with much more data. Nevertheless, semantic role labeling performance improvements demonstrate that on average the technique is learning verb linking models that are correct.</Paragraph>
  </Section>
class="xml-element"></Paper>