<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0418">
  <Title>Identifying Events using Similarity and Context</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> Both of our criteria play essential roles in the event generation process. Using similarity alone would combine all clauses that, while similar on the surface, actually are referring to different types of occurrences. Using context alone would combine all clauses encountered in the same position relative to some other clause. Either approach allows excessive noise to contaminate the resulting events.</Paragraph>
    <Paragraph position="1"> Our event generation technique is an essential part of SPANIEL, our event correlation learning system. Without a means for combining nodes, the system would be unable to generalize between different authors' descriptions of occurrences, unless they used exactly the same terminology. While standardized lexicons have been investigated (Kamprath et al., 1998), their use has not become common. Therefore, competent event generation is required for success in detecting event correlations.</Paragraph>
    <Paragraph position="2"> In addition to their role in event correlations, events could be used as information extraction tools. Using multiple fillers from different nodes for two primary slots makes the event potentially useful as a pattern for extracting fillers for the third slot from specific documents.</Paragraph>
    <Paragraph position="3"> Our results show that this technique for generating events produces encouragingly coherent events and out-performs randomly grouping nodes. Our technique for exploiting the training texts during event generation is relatively simple. Possible future work includes testing other algorithms for combining nodes, such as standard clustering techniques. In summary, we have defined an interesting problem and provided useful insight into its solution, which furthers our research into learning event correlations.</Paragraph>
  </Section>
class="xml-element"></Paper>