The Null Case theory of PRO (Chomsky and Lasnik (1991), Martin (1992)) comes tantalizingly close to eliminating the need for the PRO-Theorem (Chomsky (1981)). This is significant, since Binding Theory requires several ad hoc stipulations to deduce that PRO is ungoverned. In particular, if the distribution of PRO can be deduced without appealing to Binding Theory, the role of government in the module can be eliminated. This is a central goal of the Minimalist Program of Chomsky (1995) and one of the most important consequences of the Null Case theory of PRO. However, a well-known distinction between Control with whether versus if, shown in (1), poses a serious problem for an entirely Case-theoretic account of the distribution of PRO. If PRO need only be in a Null Case position, both (1a) and (1b) should be grammatical.
(1) a. I don't know whether PRO to leave b.*I don't know if PRO to leaveIn this paper, I argue that (1) is best accounted for by well-established constraints on movement, such as the Minimal Link Condition of Chomsky (1995), rather than the PRO Theorem. I propose that PRO is a clitic which must cliticize onto an appropriate host. The absence of a functional category capable of licensing PRO in the infinitival clause forces clitic climbing. I argue that, in obligatory control structures such as (1), the closest c-commanding head which can license PRO is the matrix AGR. Following Kayne (1989/1992), I take _if_ to be a head located in C and _whether_ to be a wh-phrase in CP-Spec. The fact that movement of PRO is allowed across an empty C in (1a), but is prohibited across a filled C in (1b), follows straightforwardly from RM/MLC, much like in Kayne's (1989) treatment of standard clitic climbing. I assume that PRO, being anaphoric, is marked as identical in reference to an NP in the same minimal domain. Thus cliticization of PRO to AGR results in control by the NP in the specifier of AGR. Note that if AGRo is present, movement to AGRs is blocked, which results in obligatory object control. Hence my analysis explains the strict locality of the controller, as well as the contrast in (1).
As discussed in Kayne (1992), there are Romance-internal contrasts between French, which prohibits (1b), and Italian/Spanish/Galician, which allow such constructions.
(2) a.*Je ne sais pas si PRO aller au cinema (FR) I NEG know not if to go to the cinema b. Non so se PRO farli (IT) NEG know if to do-them
Kayne provides convincing arguments that _se/si_ corresponds to English if not whether, hence should be taken to be a head. Note, in particular, that Italian se blocks climbing of overt clitics.
(3) *Non li so se PRO fare (cf. (2b))
Under my analysis, the grammaticality of (2b) indicates the presence of a c-commanding functional category capable of licensing PRO internal to the infinitival clause, in which case PRO need not move past _se_. If obligatory control is simply the result of an "anaphoric" element ending up in the same minimal domain as its "controller", constructions such as (2b), in the languages that allow it, should exhibit different control properties. I present data which show this to be true for both Italian and Galician. Consider the following example from Italian where PRO can bind a plural anaphor, despite the fact that the only possible controlling NP is singular.
(4) Non so [se PRO incontrar-ci qui] NEG know-1.sgl if to meet-ourselves here
Similar properties are displayed by other languages that allow (2b). I suggest that the category responsible for licensing PRO in such contexts is a "point of view" phrase along the lines of that proposed by Uriagereka (1995). If the "point of view" of the embedded PRO must include the matrix subject, but not necessarily be identical to it, the looser control witnessed in (4) can plausibly be derived. I also argue that PRO is licenced in this way with the class of English infinitivals that alternate with for-infinitivals. PRO can be the subject of _to meet_, which requires a plural subject, in (5a), but not (5b), although the so-called controller is singular in both cases.
(5) a. I want [PRO to meet] b. *I tried [PRO to meet]
Thus, languages show two types of control-infinitivals -- those which can license PRO internally and those which force PRO to move to a higher clause -- yielding two distinct types of control.
In the paper, I argue that the analysis of (1)-(3) in Kayne (1992) is conceptually and empirically problematic. I also discuss Kayne's important attempt to relate (2b) to the infinitive- clitic order, showing that the present analysis can at least partially derive his generalization. Here, I concentrate on Galician -- a language which presents numerous difficulties for Kayne's analysis.