<?xml version="1.0" standalone="yes"?>
<Paper uid="W00-0708">
  <Title>Memory-Based Learning for Article Generation</Title>
  <Section position="5" start_page="43" end_page="43" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> There has been considerable research on generating articles in machine translation systems (Gawrofiska, 1990; Murata and Nagao, 1993; Bond and Ogura, 1998; Heine, 1998).</Paragraph>
    <Paragraph position="1"> These systems use hand-written rules and lexical information to generate articles. The best cited results, 88% accuracy, are quoted by Heine (1998) which were obtained with respect to a very small corpus of 1,000 sentences in a restricted domain.</Paragraph>
    <Paragraph position="2"> Knight and Chander (1994) present an approach that uses decision trees to determine whether to generate the or alan. They do not consider the possibility that no article should be generated. On the basis of a corpus of 400K NP instances derived from the Wall Street Journal, they construct decision trees for the 1,600 most frequent nouns by considering over 30,000 lexical, syntactic and semantic features. They achieve an accuracy of 81% with respect to these nouns. By guessing the for the remainder of the nouns, they achieve an overall accuracy of 78%.</Paragraph>
  </Section>
class="xml-element"></Paper>