<?xml version="1.0" standalone="yes"?>
<Paper uid="P03-1044">
  <Title>Counter-Training in Discovery of Semantic Patterns</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper we have presented counter-training, a method for strengthening unsupervised strategies for knowledge acquisition. It is a simple way to combine unsupervised learners for a kind of &amp;quot;mutual supervision&amp;quot;, where they prevent each other from degradation of accuracy.</Paragraph>
    <Paragraph position="1"> Our experiments in acquisition of semantic patterns show that counter-training is an effective way to combat the otherwise unlimited expansion in unsupervised search. Counter-training is applicable in settings where a set of data points has to be categorized as belonging to one or more target categories. The main features of counter-training are:  off, compared to the single-trained learner.</Paragraph>
  </Section>
class="xml-element"></Paper>