<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1050">
  <Title>Learning Event Durations from Event Descriptions</Title>
  <Section position="8" start_page="399" end_page="399" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> In the research described in this paper, we have addressed a problem -- extracting information about event durations encoded in event descriptions -- that has heretofore received very little attention in the field. It is information that can have a substantial impact on applications where the temporal placement of events is important.</Paragraph>
    <Paragraph position="1"> Moreover, it is representative of a set of problems - making use of the vague information in text - that has largely eluded empirical approaches in the past. In (Pan et al., 2006), we explicate the linguistic categories of the phenomena that give rise to grossly discrepant judgments among annotators, and give guidelines on resolving these discrepancies. In the present paper, we describe a method for measuring inter-annotator agreement when the judgments are intervals on a scale; this should extend from time to other scalar judgments. Inter-annotator agreement is too low on fine-grained judgments. However, for the coarse-grained judgments of more than or less than a day, and of approximate agreement on temporal unit, human agreement is acceptably high. For these cases, we have shown that machine-learning techniques achieve impressive results.</Paragraph>
  </Section>
class="xml-element"></Paper>