<?xml version="1.0" standalone="yes"?>
<Paper uid="A92-1011">
  <Title>The Acquisition of Lexical Knowledge from Combined Machine-Readable Dictionary Sources</Title>
  <Section position="3" start_page="80" end_page="81" type="intro">
    <SectionTitle>
2 Background
</SectionTitle>
    <Paragraph position="0"> Our point of departure is the Lexical Data Base (LDB) and LKB tools developed at the Computer Laboratory in Cambridge within the context of the ACQUILEX project. The LDB (Carroll, 1990) gives flexible access to MRDs and is endowed with a graphic interface which provides a user-friendly environment for query formation and information retrieval. It allows several dictionaries to be loaded and queried in parallel.</Paragraph>
    <Paragraph position="1">  Until recently, this facility has been used to extract information from combined MRD sources which included a main dictionary and a number of dictionaries derived from it. For example, the information needed to build LKB representations for English verbs (see below) was partly obtained by running LDB queries which combined information from the Longman Dictionary of Contemporary English (LDOCE) and two other dictionaries derived from LDOCE: LDOCE_Inter and LDOCE_Sem.</Paragraph>
    <Paragraph position="2"> LDOCE_Inter was derived by a translation program which mapped the grammar codes of LDOCE entries into theoretically neutral intermediate representations (Boguraev &amp; Briscoe, 1989; Carroll &amp; Grover, 1989).</Paragraph>
    <Paragraph position="3"> LDOCE_Sem was derived by extracting genus terms from dictionary definitions in LDOCE (Alshawi, 1989; Vossen, 1990). Figure 1 provides an illustrative example of an LDB query which combines these MRD sources.</Paragraph>
    <Paragraph position="4"> We used these LDB facilities for running queries from combined MRD sources which included more than one MRD -- i.e. LDOCE and the Longman Lexicon of Contemporary English (LLOCE), a thesaurus closely related to LDOCE.</Paragraph>
    <Paragraph position="5"> The LKB provides a lexicon development environment which uses a typed graph-based unification formalism as representation language. A detailed description of the LKB's representation language is given in papers by Copestake, de Paiva and Sanfilippo in Briscoe et al. (forthcoming); various properties of the system are also discussed in Briscoe (1991) and Copestake (1992). The LKB allows the user to define an inheritance network of types plus restrictions associated with them, and to create lexicons where such types are assigned to lexical templates extracted through LDB queries which give word-sense specific information. Consider, for example, the lexical template relative to the first LDOCE sense of the verb delight in (1) where sense specific information is integrated with a reference to the LKB type STRICT-TRANS-SIGII which provides a general syntactic and semantic characterization of strict transitive verbs. 1</Paragraph>
    <Paragraph position="7"> &lt;sense-id: idb-entry-no&gt; = &amp;quot;9335&amp;quot; &lt;sense-id:sense-no&gt; I &amp;quot;1&amp;quot;.</Paragraph>
    <Paragraph position="8"> When loaded into the LKB, the lexical template above will expand into a full syntactic and semantic representation as shown in Figure 2; this representation arises from integrating sense-specific information with the information structure associated with the type STRICT-TRANS-SIGNfi XThe type specification INDEF-0BJ in (1) corresponds to the LDB value Unaccusative (see Figure 1) and marks transitive verbs which are amenable to the indefinite object alternation, e.g. a book which is certain to delight them vs. a book which is certain to delight. Information concerning diathesis alternations is also derived from LDOCElnter. The value TRUE for the attribute reg-morph indicates that delight has regular morphology. OBJ and E-ItlJI~N are sorted variables for individual objects.</Paragraph>
    <Paragraph position="9"> 2According to the verb representation adopted in the LKB (Sanfilippo, forthcoming), verbs are treated as predicates of eventualities and thematic roles as relations between eventualities and individuMs (Parsons, 1990). The semantic content</Paragraph>
  </Section>
class="xml-element"></Paper>