<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1108">
  <Title>Event Extraction in a Plot Advice Agent</Title>
  <Section position="9" start_page="862" end_page="862" type="evalu">
    <SectionTitle>
5.4 Results
</SectionTitle>
    <Paragraph position="0"> The results are given in Table 5. The majority of the advice was rated overall as &amp;quot;fair.&amp;quot; Only one story was given &amp;quot;poor&amp;quot; advice, and a few were given &amp;quot;good&amp;quot; advice. However, most advice rated as &amp;quot;good&amp;quot; was the advice generated by &amp;quot;excellent&amp;quot; stories, which generate less advice than other types of stories. &amp;quot;Poor&amp;quot; stories were given almost entirely &amp;quot;fair&amp;quot; advice, although once &amp;quot;poor&amp;quot; advice was generated. In general, the teacher found &amp;quot;coarse-grained&amp;quot; advice to be very useful, and was very pleased that the agent could detect when the student needed to re-read the story and when a student did not need to write any more. In some cases the specific advice was shown to help provide a &amp;quot;crucial detail&amp;quot; and help &amp;quot;elicit a fact.&amp;quot; The advice was often &amp;quot;repetitive&amp;quot; and &amp;quot;badly phrased.&amp;quot; The specific advice came under criticism for often not &amp;quot;being directed enough&amp;quot; and for being &amp;quot;too literal&amp;quot; and not &amp;quot;inferential enough.&amp;quot; The rater noticed that &amp;quot;The program can not differentiate between an unfinished story...and one that is confused.&amp;quot; and that &amp;quot;Some why, where and how questions could be used&amp;quot; in the advice.</Paragraph>
  </Section>
class="xml-element"></Paper>