<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-0735">
  <Title>Single-Classifier Memory-Based Phrase Chunking</Title>
  <Section position="5" start_page="158" end_page="158" type="intro">
    <SectionTitle>
4 Discussion
</SectionTitle>
    <Paragraph position="0"> We found in the experiments that minor improvements on the default settings of TiMBL can be obtained by applying MVDM, particularly to the POS tags. A larger k generally improved accuracy to a certain extent. Unpacking the features did not give the expected improvement. Complex features, however, did, and seem a promising alley to go.</Paragraph>
  </Section>
class="xml-element"></Paper>