EPP in control infinitives

The analysis assumes that a finite T carries an [EPP] feature which drives A-movement. But what about the kind of infinitival [T to] constituent found in control clauses? we assumed that infinitival to never has an [EPP] feature, and hence that the PRO subject of a control clause like that bracketed in (61a) below remains in situ in spec-VP as in (61b), rather than raising to spec–TP as in (61c):

We noted Baltin’s (1995) claim that the in-situ analysis (61b) under which PRO remains in situ would account for why wanna-contraction is possible in such sentences (yielding They don’t wanna see you), since there would be no PRO intervening between want and to. However, Baltin’s argument is not entirely convincing. After all, if intervening null constituents block to from criticizing onto want and if control clauses are CPs, why doesn’t the intervening null complementizer in (61b,c) block wanna-contraction? If we answer this question by suggesting that to first criticizes onto the null complementizer and then the two of them together subsequently criticize onto want, we can handle the relevant data by supposing that only an overt subject in spec-TP (like who at the relevant stage of derivation in ∗Who don’t they want to see you?) in spec-TP blocks cliticization of to onto C, not a null subject like PRO in (61c). What weakens the contraction argument still further is that this kind of contraction is idiosyncratic to the verb WANT (and indeed to the form want rather than wants, wanted, or wanting) rather than being associated with all control predicates, and this has led some linguists to suggest that wanna should simply be listed in the lexicon as an idiosyncratic form of want rather than being the product of a cliticization operation. (See Boeckx 2000 for an alternative account of wanna-contraction.)

In short, the wanna-cliticization argument for saying that PRO remains in situ is potentially flawed. Indeed, there seems to be counter-evidence in support of claiming that PRO does in fact move to spec-TP in control infinitives (and hence that control to has an EPP feature). Part of the evidence comes from the syntax of constituents like those italicized in (62) below which have the property that they are construed as modifying a bold-printed antecedent which is not immediately adjacent to them in the relevant structure:

Both in (62a) is a floating quantifier (and each/all can be used in a similar fashion); myself in (62b) is a floating emphatic reflexive; and personally in (62c) is an argument-oriented adverb (construed as modifying an argument, in this case he). In each sentence in (62), the italicized expression is construed as modifying the bold-printed subject of the clause. Contrasts such as those in (63) and (64) below:

suggest that a floating modifier must be c-commanded by its bold-printed antecedent.

In the light of the requirement for a floating modifier to be c-commanded by its antecedent, consider the syntax of the bracketed clauses in the following sentences:

In each of these examples, the bracketed clause is a control clause containing a PRO argument. In each case, PRO is the thematic complement of a passive participle (viz. betrayed/indicted/accused). Hence, if control to has no [EPP] feature and PRO remains in situ, the TP in the bracketed infinitive complement in (65b) will have the skeletal structure (66a) below, but if control to has an [EPP] feature, this will trigger movement of PRO to become the structural subject of to – as in (66b):

Given the requirement for a floating emphatic reflexive to be c-commanded by its antecedent, and given that PRO is the intended antecedent of themselves in (66), it is clear that (66a) cannot be the right structure, since PRO does not c-command themselves in (66a). By contrast, movement of PRO to spec-TP in (66b) means that PRO will indeed c-command themselves, so correctly predicting that (66b) is grammatical.

Let’s therefore follow Chomsky (1998, 1999, 2001) in positing that control to does indeed have an [EPP] feature, triggering raising of PRO to spec-TP. Let’s also follow Chomsky in positing that PRO is assigned null case by agreement with a c-commanding T with null (non-finite) tense in much the same way as subjects in tensed clauses are assigned nominative case by agreement with a c-commanding T which has finite (present or past) tense. More specifically, let’s assume that to in control infinitives contains not only an abstract non-finite tense feature, but also abstract -features; and let’s further suppose that null case assignment can be characterized informally as follows:

See Stowell (1982) and Martin (2001) on the tense properties of control to, and Martin (2001) for evidence that control to has agreement features; but see Bowers (2002) for a different analysis of the case-marking of PRO subjects.

In the light of these assumptions, consider the derivation of the bracketed control clause in:

Decide is a control predicate (as we see from the fact that (68) is paraphraseable as They have decided that they will help you, and from the fact that decide does not allow an expletive subject in a sentence like ∗There has decided to be an enquiry). Given the VP-Internal Subject Hypothesis, the PRO subject of the bracketed infinitive clause will originate in spec-VP, as the specifier of help you. More specifically, the derivation proceeds as follows. The verb help merges with its complement you, and the resulting V-bar help you in turn merges with its PRO subject to form the VP PRO help you. Merging control to with this VP forms the TP to help you. Let’s suppose that since PRO refers back to they in (68), PRO (as used here) carries the interpretable features [3-Pers, Pl-Num]; let’s also suppose that PRO enters the derivation with an unvalued case feature [u-Case]. In addition, let’s assume that control to carries an interpretable non-finite-tense feature [Nf-Tns] (denoting an irrealis event which has not yet happened but may happen in the future), and also has uninterpretable (and unvalued) person/number features. Finally (for the reasons given above), let’s assume that to carries an [EPP] feature in control clauses. Given all these assumptions, merging to with [VP PRO help you] will form the T-bar (69) below (simplified by showing only features on constituents of immediate concern to us):

Since to is the highest head in the structure and is active (by virtue of its uninterpretable -features), it serves as a probe which searches for a goal to value and delete its -features. Since to c-commands PRO and PRO is active by virtue of its uninterpretable case feature, PRO can serve as a goal for the probe to. The unvalued -features on the probe are assigned the same third-person-plural values as those on the goal by Feature-Copying (7) and are deleted by Feature-Deletion (14). The unvalued case feature on PRO is assigned the value [Null-Case] by Null Case Assignment (67) and deleted by Feature-Deletion (14). Since PRO is a definite pronoun, the [EPP] feature of to is deleted by movement of PRO to spec-TP in accordance with the EPP Generalization (45iii). The result of applying these various operations is to derive the TP (70) below (simplified in a number of ways, e.g. by showing the trace of PRO simply as t rather than as a deleted copy of PRO):

The resulting TP is subsequently merged with the null non-finite complementizer which introduces control clauses. As required, the structure which will serve as input to the semantic component will contain only (bold-printed) interpretable features – all uninterpretable features having been deleted.

A question of incidental detail which arises from the assumption made above that control T assigns null case to a nominal or pronominal expression which it c-commands is why T in (69) cannot assign null case to the pronoun you which is the object of the transitive verb help, since T c-commands you as well as PRO. One answer to this question is that, transitive verb phrases are phases, and hence the Phase Impenetrability Condition (55) allows a T probe to locate a goal on the edge of a transitive VP (like the PRO subject of the VP in (69) above), but not to locate a goal in the c-command domain of a transitive verb (hence not a pronoun like you in (69) since this is c-commanded by the transitive phase head help). A second answer is that the Earliness Principle requires you to be assigned case as early as possible in the derivation; and given our assumption that a transitive head assigns accusative case to a noun or pronoun expression which it c-commands, it follows that the case feature carried by you will be valued as accusative (and, we suppose, deleted) at the stage of derivation where it is merged as the complement of the transitive verb help: and our Feature Inactivation Hypothesis (52) tells us that once its case feature is deleted, you thereafter becomes inactive for agreement.

We suggested that a finite T has an [EPP] feature which triggers movement of the closest active matching goal to spec-TP, in conformity with the Attract Closest Principle. Here we have suggested that control to likewise carries an [EPP] feature triggering movement of the closest active goal to spec-TP. This suggests that we should look to see whether there is some property which finite T and control T share in common which will account for why both of them have an [EPP] feature. One possibility suggested by Chomsky (1999, p. 6) is ‘to associate EPP with -completeness’. What this would mean is that T has an [EPP] feature only if it has a complete set of -features – an idea explored in Nasu (2001, 2002). On the assumption that in a language like English where T probes agree in person and number but not gender with appropriate goals T is -complete if it carries person and number features, we can say that a finite T is -complete by virtue of carrying person and number features, and the same is true of control to under the formulation of Null Case Assignment given in (67) above. However, a natural question to ask in relation to the - completeness analysis of [EPP] is whether raising to also has an [EPP] feature, and if so whether it is -complete or not.