<?xml version="1.0" standalone="yes"?>
<Paper uid="J92-3001">
  <Title>Making DATR Work for Speech: Lexicon Compilation in SUNDIAL</Title>
  <Section position="8" start_page="264" end_page="265" type="concl">
    <SectionTitle>
7. Conclusion
</SectionTitle>
    <Paragraph position="0"> In the design of our lexicon compilation tool, we have shown how linguistic knowledge can be arranged in terms of a set of DATR structured base definitions that are  Computational Linguistics Volume 18, Number 3 portable across applications. Knowledge at the levels of morphology, syntax, and semantics combines in a single reusable DATR database. The fact that this knowledge is expressed in a high-level representation language does not limit its usefulness. On the contrary, it allows the designers of base definitions or application lexicons to think clearly about the structural relations that hold between objects in the representation and to maximize generalizations. Default inheritance allows generalizations to trickle down to specific instances, unless overridden. As a consequence, every good generalization captured during the design of structured base definitions represents labor saved during subsequent application-specific work.</Paragraph>
    <Paragraph position="1"> We have also shown how high-level knowledge can be entered by the lexicon builder at the appropriate conceptual level and then compiled into a lower level form appropriate for a chosen application. The system we describe produces two kinds of output: a term-encoded full lexicon for use in sentence interpretation and generation, and a lower level bit-encoded acceptance lexicon for use in lattice pruning during speech processing. The modular design of our system makes it particularly easy to exchange existing coding modules for new ones, thus allowing linguistic knowledge to be customized for a wide variety of speech or language applications.</Paragraph>
  </Section>
class="xml-element"></Paper>