<?xml version="1.0" standalone="yes"?>
<Paper uid="P87-1026">
  <Title>FLUSH: A Flexible Lexicon Design</Title>
  <Section position="2" start_page="186" end_page="186" type="metho">
    <SectionTitle>
A. Compound Lexemes
1. Word Sequences
</SectionTitle>
    <Paragraph position="0"> Word sequences are phrases such as &amp;quot;by and large&amp;quot; and &amp;quot;let alone&amp;quot; that must be treated as compound words because there is little hope in trying to determine their meaning by examining their components. Internally, these word sequences may or may not be grammatical (e.g., &amp;quot;kick the bucket&amp;quot; is internally grammatical, but &amp;quot;by and large&amp;quot; is not).</Paragraph>
    <Paragraph position="1"> Because type of compound lexeme is very specific, a separate category exists for each word sequence under the general category of word-sequence. Lexical constraints are placed on the different constituents of the word-sequence relation by dominating them by the appropriate simple lexeme. This is one method that can be used to establish constraints on compound lexemes, and it is used throughout the compound lexeme hierarchy.</Paragraph>
  </Section>
  <Section position="3" start_page="186" end_page="186" type="metho">
    <SectionTitle>
2. Lexical Relations
</SectionTitle>
    <Paragraph position="0"> Lexical relations include compound lexical entities such as &amp;quot;pick up&amp;quot; and &amp;quot;sell out&amp;quot; that can appear in a variety of surface forms, but have some general relationship among their simple lexeme constituents. Compound lexemes such as verb-particles (&amp;quot;pick up&amp;quot;), verb-prepositions (&amp;quot;take to&amp;quot;), and helper-verbs (&amp;quot;get going&amp;quot;) all fall into the category of lezical relations. In contrast to the individual subcategories of word sequences, there are many entries that fall underneath each individual subcategory of lexical relations. Most of the entries under these subcategories, however, share constituents with other entries, which makes generalizations possible. For example, Figure 1 shows how all verb-particles that have up as the par- null ticle (e.g., &amp;quot;pick up&amp;quot;, &amp;quot;throw up&amp;quot;, &amp;quot;look up') are represented. null This generalization in representing seemingly specific phrases is what makes FLUStt extensible. If a new verb-particle with up as the particle is added to the system (e.g., &amp;quot;hang up&amp;quot;), it inherits everything except the verb from the structure above it--that is, the general properties of verb-particle relations are inherited (such as the transposition of the particle with the object &amp;quot;it&amp;quot;), and the specific properties of verb-particles having the preposition &amp;quot;up&amp;quot; (the constraint on the preposition itself, and possibly some default semantics for the particle) are inherited.</Paragraph>
  </Section>
  <Section position="4" start_page="186" end_page="186" type="metho">
    <SectionTitle>
3. Linguistic Relations
</SectionTitle>
    <Paragraph position="0"> Linguistic relaiions are invoked according to constraints on their constituents, where the constituents may be simple lexemes, compound lexemes, or syntactic structures. An example occurs in the sentence &amp;quot;John was sold a book by Mary&amp;quot; where the object of the preposition is the main actor of the event described by the verb. This condition occurs only when the whole verb'is in the passive form (constraint 1) and the preposition in the modifying prepositional phrase is by (constraint 2).</Paragraph>
    <Paragraph position="1"> Linguistic relations are difficult to represent for two reasons: their constituents are not always simple lexemes and usually there are additional constraints on each constituent. It has been found, however, that a great deal of generality can be extracted from most of the linguistic relations to make accessing them easier. The best example of a linguistic relation is the class of the modifying prepositioval phrases. In some instances, prepositional phrases modify noun phrases and verb phrases in almost the same way (e.g., &amp;quot;The man on the hill is a skier&amp;quot; and &amp;quot;We had a picnic on the hil?'). In other cases prepositional phrases modify noun phrases and verb phrases in completely different ways (e.g., &amp;quot;The man by the car is my father.&amp;quot; and &amp;quot;The boy was hit by the car.&amp;quot;). FLUSH is able to represent both types of linguistic relation by having more than one level of generic representation. Figure 2 shows the general modifying relation (mod.rel) at the first level below compound-lexeme. Prepositional phrases that are homogeneous across noun phrases and verb phrases are represented underneath this category. Below rood.tel in Figure 2 are the verb-adjunct (va) and noun-post-modifier (npm) categories, which make up the second level of generic representation. Prepositional phrases that modify verb phrases and noun phrases differently are represented underneath these categories.</Paragraph>
    <Paragraph position="2"> As an example, in Figure 2 the rood-tel category has the more specific modifying relation mod-rel-zzz.from underneath it, which is a modifying relation where the preposition in the modifier is prep-from. Example uses of this prepositional phrase are found in the sentences: &amp;quot;The man arrived from New York&amp;quot; and &amp;quot;The woman from Boston is my aunt&amp;quot;.</Paragraph>
  </Section>
class="xml-element"></Paper>