In a lengthy footnote to his article ‘Pronominalization, negation, and the analysis of adverbs’ (1970) George Lakoff makes the following interesting observation:

All known selectional restrictions on the occurrence of lexical items are finite in scope. Usually they apply in the same sentence (subject–verb and verb–object selections) and sometimes one sentence away (verb–verb selections; i.e.force requires an activity verb in its complement) (160).

The term ‘agentive’ is a familiar one in current discussions of the syntax of English (and other languages). Although most who use it seem, at first glance, to be referring to more or less the same semantic feature, the term is employed in a variety of ways. For instance, Fillmore (1968: 24) talks of an ‘agentive case’; Gruber (1967: 943) has ‘agentive verbs’; Lyons and others speak of ‘agentive nouns’; while Halliday (1967: 196), although he does not use the term ‘agentive’, distinguishes a feature of clauses which is clearly related to the notion of agentivity. Apart from this disagreement as to what, precisely, the term ‘agentive’ is to be predicated of, there is a further divergence of opinion over which nouns (or verbs, etc.) are to be considered agentive. For instance, Lyons marks the surface subject of see as agentive (1968: 387). Fillmore, on the other hand, assigns this to the dative case, and Gruber classifies see as a non-agentive verb. Again, Lyons and Halliday take up what are in effect incompatible positions with regard to the status of the prisoners in John marched the prisoners. In view of this indeterminacy, it seems worthwhile to attempt a critical examination of the notion of agentivity, in the hope that a clearer characterization of it might emerge.

The convention discussed in the following pages made its debut, as will be apparent, in the course of work on Catalan morphology and phonology. It is clear, however, that the TYPE of data discussed is by no means restricted to this language, and the problem arising in its interpretation is a general one.

A sentiment which is expressed in a number of recent works in linguistics, e.g. Emonds (1970) and Kisseberth (1970) is that there are certain generalizations about languages which cannot be captured by devices already found in linguistic theory, such as rule ordering and the various collapsing conventions (cf. Chomsky & Halle, 1968). In this paper I will present evidence which suggests that there are generalizations about languages which are not capturable even by a linguistic theory which includes a number of these recent plausible revisions. I will also give an indication of how one might go about capturing these generalizations, given the assumption that they are not spurious ones.

In his study of Japanese phonology, McCawley introduces an interesting typological classification of languages according to the kinds of rules needed for describing the distribution of prosodic features like stress and pitch (1968: 58–61). Whereas Trubetzkoy had made a simple distinction between ‘syllablecounting’ languages and ‘mora-counting’ languages, McCawley suggests that we should make two independent distinctions:

(a) accordig to the ‘unit of phonological distance’ (the unit in terms of which the location of any accent is calculated) between ‘syllable-counting’ and ‘mora-counting’ languages;

(b) according to the ‘prosodic unit’ (the unit the actually bears the accent) between ‘syllable languages’ and ‘mora languages’.

Traugott and Waterhouse (1968: 302) have made the interesting suggestion that the adverbials already and yet should be generated from a set of semantic features associated with the perfective, and that still may be related to the progressive in a similar way. The present paper hopes to show that still and anymore are actually more related to the PERFECTIVE than previously supposed, and that a description of these forms in structural semantic terms, particularly with respect to what Chafe (1970) calls the distribution of new information, helps to clarify the meaning of the perfective inflection.