<?xml version="1.0" standalone="yes"?>
<Paper uid="N06-2047">
  <Title>Engineering Management The Chinese University of Hong Kong</Title>
  <Section position="7" start_page="187" end_page="187" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> This paper proposes the integration of two types of deep linguistic features, namely word dependencies and grammatical relations, in a ME framework to handle the RC task. Our system leverages linguistic information such as POS, word dependencies and grammatical relationships in order to extract the appropriate answer sentence for a given question from all available sentences in the passage.</Paragraph>
    <Paragraph position="1"> Our system achieves 44.7% and 73.2% HumSent accuracy on the Remedia and ChungHwa test sets respectively. This shows a statistically signi cant improvement over the reference performance levels, 40.3% and 70.6% on the same test sets.</Paragraph>
  </Section>
class="xml-element"></Paper>