THE HYPOTHESIS

For optimum semantic description of the verbs of a natural language we suggest using a mixture of the componential and definitional methods - minimizing the drawbacks associated with each method. We divide the lexical verbs of a language into two groups: nuclear and non-nuclear. Componential definitions of nuclear verbs are generated from systems of primitive semantic features; non-nuclear verbs are defined in terms of nuclear verbs (or of already defined non-nuclear verbs) utilizing the full grammatical possibilities of the language in the formulation of these definitions.¹

For componential definition of nuclear verbs only a fairly small number of rather general systems of semantic features are required. A primitive feature will normally recur in the semantic descriptions of several nuclear verbs; a system of such features may also occur in the grammatical description of the language.

For a quick, rough and ready illustration of nuclear and non-nuclear verbs consider look and stare in English. Look is a nuclear verb, whereas stare is non-nuclear; stare could semantically be defined as, say, look hard - that is, the meaning of stare could be defined in terms of the meanings of look and hard, related together through a typical verb-plus-adverb construction. Look cannot in the same way be defined in terms of some more general verb; however, look does show some semantic similarities of various types to a number of other nuclear verbs - for instance, look is to listen as see is to hear - and these similarities can be captured through appropriate componential descriptions of the verbs.

This method of semantic description eliminates the disadvantages associated with the pure componential and pure definitional approaches. Great use is made of grammatical relations in formulating the definitions of non-nuclear words. Vicious circles are eliminated. Pairs of words that are synonyms or near synonyms are explicitly related through being either (1) non-nuclear words defined in terms of the same nuclear word, or (2) a nuclear word, and a non-nuclear word defined in terms of that nuclear word, or (3) two nuclear words whose componential descriptions involve a common feature or features. Antonyms are similarly explicitly related through being, or being defined in terms of, nuclear words whose semantic descriptions have a common core of features and differ only in choosing different features from a certain semantic system. Further, only a small number of rather general semantic features are required; there is no need for ad hoc features that each occur only in the semantic description of a single word.

The division of the verbs of a language into nuclear and non-nuclear sets is not in any sense an arbitrary division, but an entirely natural one. The nuclear verbs can be componentially described in terms of a small number of rather general features. To also provide completely componential descriptions of non-nuclear verbs we would have to set up additional, ad hoc systems, whose features would each occur only in a single semantic description (cf. the discussion of bilin^yu and bumiran^yu). And nuclear words must be componentially described since it appears that they are not susceptible to definition in terms of other nuclear words, in the way that non-nuclear verbs are.

Suppose that there were a language which had the requirement that its lexicon contain an absolute minimum number of verbs. Such a language need not contain any non-nuclear verbs. In place of a putative non-nuclear verb it could simply use a ‘definition’: thus instead of stare it could have look hard. The language would, however, have to contain a full set of nuclear verbs, since nuclear items cannot be replaced by definitions as can non-nuclear verbs. Dyalŋuy behaves almost exactly like this.

We have noted that Dyalŋuy is extremely parsimonious, that it has as small a vocabulary as possible consistent with it being possible to express in Dyalŋuy everything that can be expressed in Guwal. Guwal has several hundred nuclear verbs, and in addition several thousand non-nuclear verbs. Dyalŋuy has a full set of nuclear verbs (in a one-to-one correspondence with the set of Guwal nuclear verbs) but it has almost no non-nuclear items.

Since nuclear verbs cannot be defined in terms of other lexical words, we can see that it is essential for Dyalŋuy to have a direct correspondent for each Guwal nuclear verb, if it is to function as a full natural language. However, it need not have any non-nuclear words: Guwal non-nuclear verbs can in Dyalŋuy be ‘defined’ in terms of one or more nuclear verbs, and the grammatical relations of the language. When a single-word Dyalŋuy correspondent of a Guwal non-nuclear verb is elicited, the nuclear verb that is the ‘head’ of the full definition is given.

Dyalŋuy occasionally allows itself the luxury of a direct correspondent for a Guwal non-nuclear verb. But most often in such cases one dialect has a direct correspondent for a non-nuclear word and the other does not, making do with the usual ‘definition in terms of nuclear words’. For instance, in the Mamu dialect nuclear verb bawalbin was given as the Dyalŋuy correspondent of two Guwal verbs: nuclear yanu ‘go’ and non-nuclear manman^yu ‘shift camp ’; a fuller specification of the meaning of manmanyu was given by the Dyalŋuy ‘definition’ naŋguŋunu bawalbin ‘go from the camp’. The Dyirbal dialect also has Dyalŋuy bawalbin corresponding to yanu but in this instance has a non-nuclear Dyalŋuy verb, bargan^yu as direct correspondent of manman^yu. In the case of the non-nuclear verb gunin^yu ‘search’, Mamu has a direct Dyalŋuy correspondent banman ‘search’ whereas Dyirbal Dyalŋuy just defines it as n^yuɽimarind^yan^yu ‘look repeatedly’ (n^yuɽiman is in both dialects the Dyalŋuy correspondent of nuclear buɽan ‘look’ and also of eleven other non-nuclear Guwal verbs).

The Dyalŋuy correspondents of some of the more frequent verbs concerned with motion, presenting data which is directly suggestive of the mixed componential-plus-definitional approach.

Before discussing Guwal-Dyalŋuy correspondences in detail, we first make the following assumptions:

(1) if a Guwal word and a Dyalŋuy word are in one-to-one correspondence then a relation of synonymy exists between them: they have the same semantic description;

(2) if several Guwal words are in a many-to-one relation to a Dyalŋuy word then a relation of hyponymy exists between them: the semantic descriptions of the Guwal words each include the semantic description of the Dyalŋuy word (that is, the Dyalŋuy semantic description is the highest common factor of the Guwal semantic descriptions).

Here ‘synonymy’ and ‘hyponymy’ relations are recognized between words in two different languages (rather than between words all in one language, as is usual) - or rather, between words in two different ‘realizations’ of the same underlying semantic system.

It should be noted that assumption (2) does not apply to certain instances of metaphoric-type Dyalŋuy naming (see Dixon 1968 a.246 ff.); these however all concern nouns. The class of verbs is far more semantically complex and highly structured than other classes; it is thus the ideal domain in which to test the effectiveness of any method of semantic description. The method outlined is, in the present paper, applied only to description of Dyirbal verbs.²

¹ Nuclear words tend to have greater frequency than non-nuclear items. This is not to say that the least frequent nuclear verb is more frequent than the most common non-nuclear one; rather that in almost every case a certain nuclear word will have greater frequency than non-nuclear words that are related to it (for instance, by being defined in terms of it). Compare with Weinreich’s suggestion (1962.37) that ‘one way of progressively reducing the defining meta-language in richness would be to require that the definition of a term X be formulated only in words of frequency greater than that of X’. However, relative frequency is not a criterion in the present method, merely an empirical accompaniment of the nuclear/non-nuclear distinction.

² Nouns, with the exception of subsets of kinship and agegroup (boy, man, etc.) terms, do not lend themselves to componential analysis and definition in the way that verbs do - see Dixon 1968a for detailed discussion; the nuclear/non-nuclear distinction is not, generally, very useful in the semantic description of adjectives (see Dixon, forthcoming, a, for discussion).