<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-0601">
  <Title>Reading Comprehension Programs in a Statistical-Language-Processing Class*</Title>
  <Section position="4" start_page="0" end_page="0" type="evalu">
    <SectionTitle>
2 Results
</SectionTitle>
    <Paragraph position="0"> Figure 1 shows four of the results that bettered those of Deep Read. In the next section we discuss the techniques used in these programs.</Paragraph>
    <Paragraph position="1"> The performance of all the programs varied widely depending on the type of question to be answered. In particular, &amp;quot;why&amp;quot; questions proved the most difficult. (Deep Read observed the same phenomenon.) In Figure 2 we break down the results for system 3 in Figure 1 according to question type. This system was able to answer only 22% of the &amp;quot;why&amp;quot; questions correctly. Program 5, which had the most complicated scheme for handling &amp;quot;why&amp;quot; questions, answered 26% correctly.</Paragraph>
  </Section>
class="xml-element"></Paper>