<?xml version="1.0" standalone="yes"?>
<Paper uid="A00-2026">
  <Title>Trainable Methods for Surface Natural Language Generation</Title>
  <Section position="7" start_page="199" end_page="200" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> This paper presents the first systems (known to the author) that use a statistical learning approach to produce natural language text directly from a semantic representation. Information to solve the attribute ordering and lexical choice problems-which would normally be specified in a large hand-written graxnmar-- is automatically collected from data with a few feature patterns, and is combined via the maximum entropy framework. NLG2 shows that using just local n-gram information can out-perform the baseline, and NLG3 shows that using syntactic information can further improve generation accuracy. We conjecture that NLG2 and NLG3 should work in other domains which have a complexity similar to air travel, as well as available an- null notated data.</Paragraph>
  </Section>
class="xml-element"></Paper>