PROTOTYPE SEMANTICS: THE ENGLISH WORD LIE Linda Coleman and Paul Kay University of California, Berkeley The meaning of the word lie ('prevaricate') consists in a cognitive prototype to which various real or imagined events may correspond in varying degrees. This view contrasts with the familiar one in which word meanings consist of sets of necessary and sufficient conditions, and distinguish discretely between instances and non-instances. The relevance of the notions of prototype and gradience in semantics has previously been established in physical and sensory lexical domains. The present paper shows that these notions are also relevant in abstract and social domains. Results are reported from an experiment which supports this view.* Received notions of word meaning in linguistics are based on the idea of the semantic feature or component:1 semantic features are discrete properties (or relations), and they contrast discretely with one another. The meaning of a word is represented as a set of features, possibly with a single member. Variations on this standard theme differ in that some see the set of discretely contrasting features as having no further structure (as in, the writings of anthropological semanticists such as Wallace & Atkins 1960, in Chomsky 1965:214, and in the earlier writings of Jerrold Katz), while others believe that additional (e.g. treelike) structure is imposed (cf. Weinreich 1966, Katz's later writings, or Leech 1974:149 ff.) We are not concerned here with these subvariations within the generally accepted theory of discretely contrasting semantic components, since we intend to challenge the very notion of the discrete semantic feature. The ensemble of theories built upon this fundamental notion have been aptly characterized by Fillmore 1975 as checklist theories.2 Since the simplest case of such a theory is that in which no additional structure is imposed on the set of discretely contrasting features, it will suffice to show that, even in this simplest form, the checklist theory is inferior to our approach in accounting for the kind of cases that have been selected to exemplify it. (A similar demonstration in the case of color vocabulary is given by Kay & McDaniel 1978.) According to the checklist view, the definition of a semantically complex word (or sometimes a single 'sense' of such a word) consists of a set offeatures or components such that a given object (physical or otherwise) is aptly labeled by the word just when it possesses the property named by each feature in the definition. The list of features amounts to a set of necessary and sufficient conditions which a thing must satisfy in order to be an instance of the category * We wish to express thanks for detailed comments on earlier drafts to Charles Fillmore, Willett Kempton, George Lakoff, Geoffrey Nunberg, Robert Randall, Joel Sherzer, Eve Sweetser, and Kenneth Wächter. We are indebted to Alan Sonafrank for computational assistance. 1 When we speak of a 'semantic component' in this paper, we never intend it in the sense of a 'semantic section of a grammar' 2 See Fillmore's article for a seminal discussion of the prototype concept in linguistics, and for references to the earlier literature in a number of fields. Black (1954:24-5) develops a very similar view. 26 PROTOTYPE SEMANTICS27 labeled by the word. Applicability of a word or other linguistic label to a thing is thus a matter of 'yes or no', not of 'more or less.' Either the thing satisfies the list of necessary and sufficient conditions, or it does not. No partial fit of word to object is countenanced. By contrast, our prototype view of word meaning attempts to account for the obvious pretheoretical intuition that semantic categories frequently have blurry edges and allow degrees of membership. On this view, applicability of a word to a thing is in general not a matter of 'yes or no', but rather of 'more or less'. This has been shown experimentally in the domain of color (e.g. Berlin & Kay 1969, Kay & McDaniel 1978) and in several other lexical domains (Kempton 1977, Rosch 1975, Labov 1973). It has also been supported by a variety of linguistic arguments; cf. Kay 1978, Fillmore 1975, and Lakoff 1972. In the philosophy of language, Putnam has recently...
Stored only when both papers are in the corpus — 383 citations exist upstream.