<?xml version="1.0" standalone="yes"?>
<Paper uid="W05-0633">
  <Title>Semantic Role Labeling Using Lexical Statistical Information</Title>
  <Section position="5" start_page="215" end_page="215" type="evalu">
    <SectionTitle>
3 Results
</SectionTitle>
    <Paragraph position="0"> Table 1 shows the results on the test set. Problems are inherently related with the skewed distribution of role classes, so that roles which have a limited number of occurrences are harder to classify correctly.</Paragraph>
    <Paragraph position="1"> This explains the performance gap on the A0 and A1 roles on one hand, and the A2, A3, A4, AM- arguments on the other.</Paragraph>
    <Paragraph position="2"> One advantage of using a decision tree learning algorithm is that it outputs a model which includes a feature ranking, since the most informative features are those close to the root of the tree. In the present case, the most informative features were both distance/position metrics (distance and adjacency) and lexicalized features (head word and predicate).</Paragraph>
  </Section>
class="xml-element"></Paper>