Head of a Projection

In recent work in generative grammar, one finds two approaches concerning the projection of heads onto phrases. On the one hand, in the Government-Binding/Minimalist Program tradition it has commonly been assumed that a projection inherits all the properties of its head and nothing else, a principle that Brody (1998) refers to as Uniqueness. On the other hand, in Head-Driven Phrase Structure Grammar (HPSG) and its predecessors it is explicitly assumed that features of a complement of X can project to XP, so a projection of X ends up richer than its head.

Therefore, whether Uniqueness holds between a projection and its head is an issue that needs to be resolved. So far, within minimalism only a conceptual argument has been presented in favor of Uniqueness, that of Chomsky's (1995) theory of bare phrase structure. In this squib I present two additional empirical arguments in favor of Uniqueness, drawn from well-known properties of causative constructions and noun incorporation.

1 Uniqueness

The principle that a phrase has a unique head figures prominently in Brody 1998 and Chomsky 1995. In Brody 1998:371 it is expressed as in (1) (Brody's (5)).

(1) Uniqueness

Every phrase is projected by a unique category.

(1) ensures that if a head X selects for a ZP, the resulting phrase will have X as its only head, not a hybrid of X and Z. Chomsky (1995:244) reaches the same conclusion by means of the following conceptual argumentation: Take two terms, α and β, each a bundle of features (including categorial features), which are merged, forming the set {α, β}. The question now is how the label of the set is going to be defined or, in other words, what features, including the categorial features, are going to project. Chomsky proposes to deduce the answer from set theory. The possibility that the label is defined as the intersection of α and β is rejected as meaningless. The possibility that it might be defined as the union of α and β is also discarded, because it would lead to an incoherent label if the features of α and β are incompatible. The only possibility left is that the label of the set must be either α or β. Thus, only α or β projects. (Incidentally, notice that this conclusion should hold whether α or β is a maximal projection or whether both α and β are heads, with one adjoined to the other, because any application of Merge gives rise to a set.)

There is an alternative to (1), according to which compatible features of a nonhead could percolate. Let me present the following abstract scenario: Suppose that α has a feature of type f and value f₁. β, selected by α or adjoined to α via head-to-head movement, also has a feature of type f but of a different value, f₂. Presumably, f₂ should not project. Suppose further that β has an additional feature of type g but α does not. It could be assumed that g would project, so the label of the set {α, β} could be α_[f1],[g], that is, α with f₁ and g as sublabels. In (2a) I present a hypothetical example in which a head Y has adjoined to a head X via head-to-head movement, X^max has projected, but a feature of Y has also percolated to Xmax. In (2b) I consider the scenario in which both α and β have features of type f. Could the resulting projection have two instances of f? If we do not assume Uniqueness, it should.

(2) The label of {α, β} includes compatible features of a non-head.

1. a.

2. b.

Let me briefly discuss an analysis in which (2) has been assumed implicitly. In Chomsky 1995:315 transitive predicates are built in four steps. In the first step a lexical verbV merges with the object, assigns a θ-role, and projects a VP. In the second step VP merges with a light verbv and v projects a vP. In the third step vP merges with the external argument, assigns a θ-role, and projects again. Finally V incorporates into v, forming the conglomerate Vb = {v, V}. This is shown in (3).

In (3) v assigns a θ-role to the subject and accusative...