<?xml version="1.0" standalone="yes"?>
<Paper uid="J93-3004">
  <Title>Squibs and Discussions Co-occurrence Patterns among Collocations: A Tool for Corpus-Based Lexical Knowledge Acquisition</Title>
  <Section position="3" start_page="536" end_page="537" type="concl">
    <SectionTitle>
4. Conclusion
</SectionTitle>
    <Paragraph position="0"> The two pilot analyses presented here indicate that this approach can be a useful tool for the semi-automatic identification of underlying word senses and uses. Further, both analyses produced unanticipated but systematic results, indicating that this approach can provide a useful complementary perspective to traditional lexicographic methods.</Paragraph>
    <Paragraph position="1"> These analyses could be extended in several ways. First, statistics such as the mutual information index could be used to help identify the set of relevant collocations to be included in the factor analysis. Second, the corpus could be pre-processed by a  Computational Linguistics Volume 19, Number 3 grammatical tagger, making the collocations sensitive to grammatical category. Finally, lexical associations at a distance should be included in the analysis. In this regard, the analyses here have been restricted: they consider only collocations of adjacent words, with no regard for grammatical category, and with the input data identified simply on the basis of absolute frequency. However, even with these restrictions, factor analysis appears to be a powerful tool for identifying underlying patterns among collocations, reflecting some of the major senses and uses of a word.</Paragraph>
  </Section>
class="xml-element"></Paper>