<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-3243">
  <Title>On Log-Likelihood-Ratios and the Significance of Rare Events</Title>
  <Section position="10" start_page="8" end_page="8" type="concl">
    <SectionTitle>
7 Conclusions
</SectionTitle>
    <Paragraph position="0"> When we use Fisher's exact test to estimate pvalues, our new method for estimating noise for collections of rare events seems to give results that are quite consistent with our previous annecdotal experience in using LLR scores as a measure of word association. Using likelihood ratios to estimate p-values introduces a substantial amount of error, but not the orders-of-magnitude error that Dunning (1993) demonstrated for estimates that rely on the assumption of a normal distribution. However, since we have also shown that Fisher's exact test can be applied to this type of problem without a major computational penalty, there seems to be no reason to compromise in this regard.</Paragraph>
  </Section>
class="xml-element"></Paper>