<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1195">
  <Title>Tony.Veale@UCD.ie</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Broad-coverage lexical knowledge-bases like WordNet (Miller et al., 1990) generally contain a large number of compound terms, many of which are literal in composition. These compounds are undoubtedly included for a reason, yet the idea that literal compounds might actually be essential to WordNet's usefulness may strike some as heretical on at least two fronts: first, the lexicon is a finite resource, while the space of compounds is potentially infinite; and at any rate, literal compounds can be created as needed from purely compositional principles (Hanks, 2004). However, these retorts are valid only if we view WordNet as a dictionary, but of course it is much more than this. WordNet is a lexical ontology, and ultimately, ontologies derive a large part of their functionality from their structure.</Paragraph>
    <Paragraph position="1"> So, while the meaning of literal compounds like Greek-deity and animal-product may well be predictable from compositional principles alone, such concepts still serve an important organizational role in WordNet by adding much needed structure to the middle ontology. Having conceded the importance of such compounds, one is forced to address the issues of completeness and consistency that then arise from their inclusion.</Paragraph>
    <Paragraph position="2"> Completeness suggests that we strive to include as many literal compounds as are sensible, if they enhance the organization of the ontology or if there is evidence that they are in common usage in the language. Systematicity is a related issue that arises when a group of existing compounds suggests that another should also exist for the ontology to be consistent. For instance, the existence of Greek-deity, Greek-alphabet and Hebrew-alphabet leads to the hypothesis that Hebrew-deity should also exist if WordNet is to be both consistent and symmetric in its treatment of different cultural groupings.</Paragraph>
    <Paragraph position="3"> Indeed, because literal compounds like these arise from the yoking together of two different ontological branches into one, compounding represents an important contextualization device in the design of ontologies, allowing lexical elements to be logically grouped into clusters or families that share important dimensions of meaning. This clustering facilitates both automated reasoning by machines (such as the determination of semantic similarity based on taxonomic distance) and effective browsing by humans. Sometimes this yoking results in a compound that, following Boden (1990) and Wiggins (2003), deserves to be called &amp;quot;creative&amp;quot;, because it exhibits both novelty and value. Novelty can be measured along either a psychological or a historical dimension, while utility is a reflection of the uses to which a compound can be put. For instance, a new compound may have utility as a clustering node when added to the middle ontology if its appropriate hyponyms can be identified.</Paragraph>
    <Paragraph position="4"> Alternately, a new compound may represent an alternate nominalization of an existing concept (e.g., see Vendler's (1967) insights about nominalization, and Lynott and Keane's (2003) application of these insights to compound generation).</Paragraph>
    <Paragraph position="5"> In this paper we present a process of ontological exploration to identify those areas of the lexicon that can contribute to, and may in turn benefit from, the invention of new compound terms. Since the discovery of new compound terms is essentially a process of creative exploration, we frame our discussion within the theoretical framework of creative computation. Within this framework two approaches to validating new compounds are presented: internal validation determines whether the ontology itself provides evidence for the sensibility of a new compound, while external validation uses web-search to find evidence that the compound already exists outside the ontology. We then go on to show how these different strategies create a validation gap that can be exploited to identify the small number of truly creative compounds that arise.</Paragraph>
  </Section>
class="xml-element"></Paper>