<?xml version="1.0" standalone="yes"?>
<Paper uid="H89-1018">
  <Title>UNDERSTANDING SPONTANEOUS SPEECH</Title>
  <Section position="7" start_page="139" end_page="140" type="concl">
    <SectionTitle>
SUMMARY
</SectionTitle>
    <Paragraph position="0"> We aim to achive robust recognition by using a mixed strategy of syntax and semantics. Grammar is used locally to form phrases from words. The phrases are associated with meanings and semantic constraints are applied to sequences of meanings. This allows us to use grammar to guide the word search without insisting that the final results conform to the grammar. The focus is on the information to be transferred, phrases convey meanings.</Paragraph>
    <Paragraph position="1">  Sequences of meanings more naturally represent performance, particularly ellipsis and telegraphic style, than other mechanisms in use. Using semantics from all recognized parts of an utterance helps resolve ambiguous or ill-formed sections.</Paragraph>
  </Section>
class="xml-element"></Paper>