<?xml version="1.0" standalone="yes"?>
<Paper uid="P99-1042">
  <Title>Deep Read: A Reading Comprehension System</Title>
  <Section position="8" start_page="331" end_page="331" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have argued that taking reading comprehension exams is a useful task for developing and evaluating natural language understanding systems. Reading comprehension uses found material and provides humancomparable evaluations which can be computed automatically with a minimum of human annotation. Crucially, the reading comprehension task is neither too easy nor too hard, as the performance of our pilot system demonstrates.</Paragraph>
    <Paragraph position="1"> Finally, reading comprehension is a task that is sufficiently close to information extraction applications such as ad hoc question answering, fact verification, situation tracking, and document summarization, that improvements on the reading comprehension evaluations will result in improved systems for these applications.</Paragraph>
  </Section>
class="xml-element"></Paper>