<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-1005">
  <Title>Tagging for Learning: Collecting Thematic Relations from Corpus</Title>
  <Section position="13" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Current Status and Conclusions
</SectionTitle>
    <Paragraph position="0"> Based on a number of tagged sentences, the system determines that SHAREHOLDERS are recipients of PAY, while DIVIDENDS axe objects. This generalized lexical relation enables the semantic resolution of more difficult cases such as DIVIDEND PAYMENT and COMPANY PAID STOCK DIVIDEND.</Paragraph>
    <Paragraph position="1"> The implemented system using these techniques includes several elements: (1) morphology analysis - currently produces accurate results for all the required cases; (2) tagging - produces results for only 60% of the required examples; more detailed rules could improve this figure to about 70%; (3) rule forming - currently works only with dative verbs such as PAY and SELL.</Paragraph>
    <Paragraph position="2"> A number of important pieces of recent research have highlighted the power of co-occurrence information in text. In the techniques described here, we have extended this research to use co-occurrence information for discriminating thematic roles. These techniques combine data acquisition from a tagged corpus with relation-driven language analysis to derive thematic knowledge from the text.</Paragraph>
  </Section>
class="xml-element"></Paper>