I now turn to the external argument of the verb. In what way is this argument external?
In section 2.2 we established the claim that it is external to the domain formed by the verb and its internal argument(s): it is thus external to the domain o f "special meaning", to the domain of change of state and measuring out (Tenny 1994) and to the temporal path which is criterial of the event (Ramchand 1997). The external argument is thus never a measurer, and never participates in idiom-formation. Its role in the event structure o f the verb is to cause, originate or bring about the event (LRH 1995, HK
1997). It is interpreted as initiating the change of state which determines the event, but never forming part of it.
A number of linguists have assumed that the external argument is external to the VP at some level of representation (e.g., argument structure). In particular, Williams (1981) assumed that the external argument is located externally to the maximal projection o f the verb, and Marantz (1984) took this idea further, and argued that the external argument is not an argument of the verb (but rather, o f the VP).
I assume here that there are no intermediate levels of representation, and that externality is given by the syntactic position in which an argument is generated. Where, then, is the external argument generated: within the VP or not?
Theories differ on this point. Some theories, like Borer (1994), Harley (1995), Kratzer (1996) and Ramchand (1997) assume that the external argument is added by a functional head whose denotation has to do with aspectual properties, causation etc. A different line, originating with Koopman and Sportiche (1988), is known as The VP- internal subject hypothesis (VISH). Based on evidence from floating quantifiers, they argue that the subject has to be base-generated in a projection within the VP. The overt position where we find it at S-structure may be the result of movement. ^ ^
What, then, is correct? Is there a way to reconcile the syntactic evidence concerning the VP-intemal status of the subject with our semantic intuition, that this argument is excluded from a certain internal domain, which is defined by the verb and its internal argument? I believe that the theory o f VP-shells, suggested originally in Larson (1988), could do this. VP-shells were originally designed to accommodate ditransitive and locative verbs (such as give X to Y or put X on Y): such verbs are problematic for standard phrase-structure theory; first, in that they seem to be at odds with the binary-branching requirement (Kayne 1984) and second, in that there seems to be no way to represent the internal hierarchy between the two internal arguments (cf. Brass and Lasnik 1986; see discussion in section 2.4.1). The solution, as suggested by
Larson, was to split up the VP into a multi-layered shell. The lower argument, goal or location, is generated as the complement o f the verb, while the theme is generated as its specifier:
(26)
Theme
PP
Goal/Location
This structure enables the verb to accommodate both internal arguments, as well as distinguish between them hierarchically. According to Larson the agent is now generated at the specifier of an upper VP, which is located on top o f the lower VP which contains the internal arguments:
(27)
Theme
PP
Goal/Location
The VP-shell thus consists o f two verb positions: the lower one, where the lexical VP is generated, and the upper, empty one, into which it moves. The most important consequence of the VP-shell for our purpose is that it offers a way to capture our intuitions about the organization o f the VP: on the one hand, the subject is part of the maximal projection of the VP, as required by the VISH. On the other hand, it is also external to the lower VP, where the internal arguments are generated. The VP now consists of two domains: the lower VP, which is the domain o f the internal arguments, of special meaning, etc., and the upper VP, which is external to that domain, and is the domain o f the external argument.
Further modification of the VP-shell is introduced by Hale and Keyser (1993, 1997a). They assume that the specifier o f the upper VP is the thematic position of the external argument (agent/causer). The fact that the NP is in this particular configuration, [y p NP [y VP ] (an NP at the specifier of a verb which takes as a complement another VP) makes it interpreted as an external argument.
Adopting Larson's proposal, Chomsky (1995, 1997) takes the upper V head to be a functional head: it is, simultaneously, the head projecting the agent and the head which is responsible for checking accusative case features. This upper head is put in small italics, v, to distinguish it diacritically from the lower V head. Following HK (1993), Chomsky assumes that v exists in all verbs which have an external argument (including unergatives, which are taken to be hidden transitives), that is, all verbs except unaccusatives, which have neither an object nor an external argument.
The gist o f Burzio's (1986) generalization is thus retained under the assumption of v: the same head theta-marks the external argument and checks accusative case (see
cV\o.fHrpt 6 ,
discussion o f the status of EG in^section 5.7). v is thus both a functional head, i.e., it checks object case features after AgrO is dispensed with (Chomsky 1995), and a lexical head, since it specifier is a thematic position, and it is within a maximal projection which is labelled VP.
To sum up, here is the VP structure I will be assuming (ditransitives are to be discussed in the next section):
(28) a. transitive:
Collins (1997) suggests that unaccusatives, too, have a small v. His argument is based on sentences such as (i), in which the verb must have raised to an upper verbal position, that is, v:
unergative: vP NP V c. unaccusative: VP NP V
The transitive VP thus has two domains: internal and external. Unergatives have only an external argument, and unaccusatives - only an internal argument.