<?xml version="1.0" standalone="yes"?>
<Paper uid="W91-0212">
  <Title>Redefining the &amp;quot;Level&amp;quot; of the</Title>
  <Section position="8" start_page="135" end_page="136" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> The notion that the level of &amp;quot;the word&amp;quot;, as structurally defined, is the appropriate starting place for semantic representation, has been implicit in the design of most lexical knowledge bases. The use of machine-readable dictionaries as a source of lexical knowledge reinforces this notion, at the cost of considerable descriptive loss. This discussion has also revealed a further detrimental assumption fostered by the use of the dictionary as the model for the lexicon: the idea that a single store of lexical knowledge, with a single lookup function and a unified structure, is the necessary mechanism for bringing word meanings into sentence analysis. On the contrary, the multi-faceted semantic behavior of the structural units we call &amp;quot;words&amp;quot; requires a base of knowledge consisting of multiple knowledge stores, each organized in a way that is appropriate to the knowledge it contains and the stage of processing to which it must contribute. The structural word can be used as the indexing key which relates the separate, differently structured stores, but it need not be the basic currency of all of &amp;quot;lexical&amp;quot; semantics.</Paragraph>
    <Paragraph position="1"> 9 One or more of these levels of distillation would presumably be equivalent to the single set of canonical graphs envisioned by Sowa (1984).</Paragraph>
  </Section>
class="xml-element"></Paper>