D. A. Cruse's ‘Some thoughts on agentivity’ (1973: 11–23) makes the useful distinction of [volitive] and [agentive] NPs. Recognizing this difference is the first step in avoiding some of the difficulties encountered by Jackendoff (1972), Fischer and Marshall (1969) and others in their discussion of how an NP is characterized as [volitive] in a clause and how some predicates can characterize an NP in an embedded clause as [volitive]. These discussions have been relatively inconclusive because of the very common assumption that [ ±volitive] or something like it should be represented as a feature of predicates, or predictable from one. If one believes that lexical selectional features are semantic (as McCawley (1968) argues), one arrives at the kind of ‘referential’ criterion for volitivity that Cruse rightly wishes to shun and that other writers have found so dismaying (see e.g. Jackendoff, 1972: 221). The way to avoid this is to suppose that [volitive] is not a ‘feature’ at all, but rather abbreviates an assumption entailed (or ‘presupposed’) by saying certain things. In this way, some sentences can be marked as contradictory, but many will simply seem queer outside of an explicating context, set of special assumptions, and so on. It follows that a great many sentences starred in recent grammatical literature as inherently bad regardless of context ought not to be. We mark: I asked John to be tall.

Linguists are constantly invoking a relation of subordination of one clause to another and there seems to be agreement among linguists, about their native tongues at least. Yet the reason for the existence of the relation and the consistency of judgments concerning it is not well understood. The study of subordination and of argumentation for assigning the relation can profitably be made from at least two viewpoints. One viewpoint is paradigmatic or transformational. Evidence for subordination derives from an asymmetry in certain transformational relationships. Non-subordinating constructions on a pair of clauses show identical properties in each clause when the whole construction undergoes transformation, while the same transformations display non-identical properties in the clauses of subordinating constructions. The other viewpoint is syntagmatic or textual. One might even call this viewpoint semantic. Subordinating constructions are restricted in their co-occurrence within coreference patterns of a text in a way that corresponds to an intuitive feeling that a subordinate clause is not the topic of the text at the point where it occurs. Thus one might state the pattern by saying that the common element of meaning of all subordination markers is to indicate that the clause they operate on is subordinate in topic to the non-marked clause. However, the precise account will be in cerms of coreference patterns.