<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-1017">
  <Title>Event-Based Extractive Summarization</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have introduced atomic events as a feature that can be automatically extracted from text and used for summarization, and described algorithms that utilize this feature to select sentences for the summary while minimizing the overlap of information in the output. Our experimental results indicate that events are indeed an effective feature, at least in comparison with words in the input texts that form the basis of many of current summarizers' feature sets. With all three of our summarization algorithms, we achieved a gain in performance when using events. This gain was actually more pronounced with the more sophisticated sentence selection methods, establishing that events also exhibit less interdependence than features based directly on words. The advantage was also larger in longer summaries.</Paragraph>
    <Paragraph position="1"> Our approach to defining and extracting events can be improved in many ways. We are currently looking at ways of matching connectors that are similar in meaning, representing paraphrases of the same event, and methods for detecting and prioritizing special event components such as time and loca-tion phrases. We are also considering merging information across many related atomic events to a more structured representation for each event, and allowing for partial matches between such structures and input sentences.</Paragraph>
  </Section>
class="xml-element"></Paper>