<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0435">
  <Title>Memory-Based Named Entity Recognition using Unannotated Data</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> In the plain learning problem (i.e. using only annotated data), our system used only context and word-internal information. Type - token generalization was never beneficial, and gazetteers helped only for the German task.</Paragraph>
    <Paragraph position="1"> When using unannotated data, performance was improved in two ways: extra gazetteers were constructed by exploiting conjunctions, and words which appear mostly in capitalized form were set apart from those that do not.</Paragraph>
  </Section>
class="xml-element"></Paper>