This paper takes up various problems of Gorum verb morphology and, more important, of its semantic categories. These relate specifically to the interconnections found between so-called causativization and transitivity. The notion ‘intransitivity’, is shown to be inadequate to characterize the data, and can be replaced by a semantic feature – which we call affectedness – of the subject NP of a primary verb in a tree. Higher sentence construction is not adequately handled by the ambiguous – or various – notions of causativity. What happens in Gorum can be more adequately described in terms of displacement of the head NP in a sentence, and this is shown to be directly related to transitivity, causation and perhaps volition. For, an additional distinction operating in the Gorum verbal system is that of internally motivated action versus the absence of such motivation. This distinction – and its correlation with the feature affected – is also touched on.

Two models of generative grammar are now available, the interpretative semantic model and the generative semantic model. The latter model differs sharply from Chomsky's ‘standard model’ in that its base component generates semantic objects, namely predicates and their arguments, whereas the base component of the ‘standard model’ generates syntactic objects, e.g. noun phrases and verb phrases. The interpretative semantic model is essentially the ‘standard model’ to which have been added rules for interpreting surface structures. (For an interpretative account of pronouns and negation, see Jackendoff (1968, 1969). A criticism of Jackendoff's proposals is to be found in Lakoff (1970)). This paper offers an informal account of aspect in Russian which, it will be argued, indicates that the many problems raised by generative semantics must be tackled by any linguist interested in ‘explanatory adequacy’.

1. Adverbial subordinate clauses introduced by instead of and rather than exhibit certain unique syntactic and semantic properties. I will propose an account of the interaction between the structure and the meaning of sentences containing these two clause types, and suggest some explanations for the rather clear-cut variation which exists among speakers with respect to these sentences.

In the present study I am adducing arguments against the assumption that there is a discrete level of grammar where all syntactic operation has stopped and the assignment of phonetic values to the then existing structure (the output of transformations) begins.

The concept of ‘surface structure’ has received the following definitions in Chomsky's various works: ‘a proper bracketing of the linear, temporarily given sequence of elements, with the paired brackets labelled by category names (that is a labelled tree diagram, with such categories as Sentence, Noun Phrase, Verb Phrase, Noun, and a small number of others serving as labels)' (Chomsky, 1966: 21), ‘the organization into categories and phrases that is directly associated with the physical signal’ (Chomsky, 1968:25). These definitions take the physical event of speech as a starting point, and refer to the organization of the sentence on this level (cf. Chomsky, 1965a: 15). One finds, on the other hand, that in the Chomskyan system ‘surface structure’ is also defined in terms of derivation, as the last item in a finite sequence of phrase-markers, the output of the transformations (Chomsky, 1970: 53 f., 87), and such surface structures ‘are mapped into phonetic representations by the phonological rules’ (ibid.: 54) and are assigned intonational contours by rules of phonological interpretation (ibid.: 77). This is summed up in the formula ‘phonology: Pn→phonetic representation’ (ibid.: 87, where Pn is the surface structure, the last phrase-marker).

Linguists concerned with the historical evolution of a language have for some time been aware of the implications for their discipline of the distinction between deep and surface structure implicit, of course, in the writings of such Cartesians as Cordemoy and in the Grammaire générale et raisonnée of Port-Royal, but largely ignored by historical linguists until it re-emerged as crucial within the generative-transformational approach to language study. It is now evident that historical linguistics should not be exclusively concerned with surface-structure changes – changes which are particularly clear and well-documented within the field of Romance linguistics – but should also consider what changes, if any, have taken place in the deep structure underlying the surface representation. The present paper will look at various recent analyses of the interrelationship between these two levels and will go on to suggest, largely on the basis of evidence drawn from the evolution of the verb system from Latin to French, that there are in fact at least three valid levels of analysis and that the historical linguist will be primarily concerned with neither the deepest nor the most superficial of these.