Squibs and Discussions: WordNet Nouns: Classes and Instances

If you were to say “Women are numerous,” you would not wish to imply that any particular woman is numerous. Instead, you would probably mean something like “The class of women contains numerous instances.” To say, on the other hand, “Rosa Parks is numerous,” would be nonsense. Whereas the noun woman denotes a class, the proper noun Rosa Parks is an instance of that class. As Quirk et al. (1985, page 288) point out, proper nouns normally lack number contrast. This important distinction between classes and instances underlies the present discussion of WordNet nouns. Some nouns are understood to refer to classes; membership in those classes determines the semantic relation of hyponymy that is basic for the organization of nouns in WordNet (WN). Other nouns, however, are understood to refer to particular individuals. In many cases the distinction is clear, but not always. The distinction to be discussed here is between words ordinarily understood as referring to classes and words ordinarily understood as referring to particular individuals and places. In the literature on knowledge representation, the classic discussion of this distinction is provided by Woods (1975). The distinction was not drawn in initial versions of WN (Miller 1990; Fellbaum 1998), which used the “is a” relation in both cases. That is to say, both “A heroine is a woman” and “Rosa Parks is a woman” were considered to be occurrences of the “is a” relation and were encoded in the WN database in the same manner. Requests to incorporate a distinction between classes and instances have come from ontologists, among others. In their discussion of WN, for example, Gangemi et al. (2001) and Oltramari et al. (2002) complain about the confusion between individuals and concepts. They suggest that if there was an “instance of” relation, they could distinguish between a concept-to-concept relation of subsumption and an individual-to-concept relation of instantiation. That is, essentially, the suggestion we follow in the present work, but in some cases the distinction was not easy to draw. Incorporating this distinction was resisted at first because WN was not initially conceived as an ontology, but rather as a description of lexical knowledge. WN includes verbs, adjectives, and adverbs in addition to nouns. Although no ontology was intended, the organization of nouns in WN bore many similarities to an ontology. As the importance of ontology became more apparent, requests to convert the WN noun hierarchy could no longer be ignored. Version 2.1 of WN takes a step in that direction: the Tops file is reorganized to have a single unique beginner: entity. In a reasonable ontology, however, all terms might be expected to conform to the membership relation of set theory and would not contain individuals or placenames. The confounding of classes and instances in WN posed a problem. The obvious way to solve