On Infinitives and Floating Quantification

1 The Multiple Quantifier Diagnostic

Hornstein (2001) develops a diagnostic test based on the properties of floating quantifiers (FQs) to show that obligatory control (OC) structures involve an extended A-chain (the movement theory of control (MTC); see Hornstein 1999). This diagnostic is based on Sportiche's (1988) idea that FQs are residues of an A-movement, adjacent to a trace of a moved NP.¹ Hornstein observes that examples in which an NP is associated to more than one FQ—and, in general, to more than one quantifier—are odd (see Hornstein's examples (89b-c) and (91b)).

(1)

1. ??{All} The men {all} have all eaten supper.

2. b. ??{Both} The men {both} have both eaten supper.

Raising structures containing two FQs (or one DP-internal quantifier and one FQ) are odd as well, since they contain only one A-chain (Hornstein's examples (91a) and (93b)).

(2)

1. a. ??{All} The men {all} seem to have all eaten supper.

2. b. ??{Both} The men {both} seem to have both eaten supper.

Since each NP chain can include at most one quantifier (be it DP-internal or floating), quantifiers can be used as a diagnostic to detect A-chains. I will dub this the multiple quantifier diagnostic (MQD).

Hornstein argues that if OC can be analyzed in terms of movement, OC structures containing more than one quantifier are expected to have the same status as sentences (1) and (2). He then shows that this expectation is borne out (Hornstein's examples (95c) and (95d)).

(3)

1. a. ??All the men hope to have all eaten supper (by 6).

2. b. ??Both the men hope to have both eaten supper (by 6).

On the other hand, structures involving two A-chains each containing one FQ are predicted to be fully legitimate, since they do not violate the constraint prohibiting the presence of more than one quantifier per chain. In Hornstein's approach, nonobligatory control (NOC) does not involve A-movement. The optional controller and the infinitival subject—pro, in Hornstein's view—are not part of the same A-chain. Thus, the prediction follows that each A-chain can contain one FQ without violating any constraint on the distribution of FQs. The following examples (Hornstein's examples (96a) and (96b)) show that this prediction is borne out:

(4)

1. a. The men all thought that all dancing with Mary was fun.

2. b. The men both thought that both dancing with Mary was fun.

I will now show that the MQD is able to shed new light on some long-standing questions concerning the nature of OC.

2 Floating Quantifiers and Control Theories

Descriptively, OC has been shown to include two different subclasses of phenomena, which Landau (2000) has dubbed exhaustive and partial control (EC and PC). The former class includes predicates requiring that the denotation of the infinitival subject be strictly identical to the denotation of the controller. The latter class includes predicates that do not require that the denotation of the infinitival subject and the denotation of the controller be identical; in PC, the denotation of the subject of the infinitival predicate need only include the denotation of the controller.

While Hornstein (2003) and Boeckx and Hornstein (2006) claim that both EC and PC must be analyzed in terms of A-movement,² Landau (2003), rejecting the hypothesis that OC is movement and claiming that it is best explained in terms of Agree, points out that PC cannot be explained in terms of movement, since a DP and its traces must be referentially strictly identical. Barrie (2004) and Barrie and Pittman (2004) propose a different solution to explain PC in terms of the MTC, hypothesizing that in PC, A-chains split at LF in order not to violate the θ-Criterion, and copies are reinterpreted (as pro).³ Finally, Wurmbrand (1998, 2001, 2003, 2004) and Cinque (2004, 2006) hypothesize that restructuring involves A-movement.⁴ In their view, EC ("perfect control," as Wurmbrand labels it) coincides with restructuring; PC ("imperfect control," in Wurmbrand's terminology) corresponds to nonrestructuring infinitives and is the only authentic instance of OC—at least in Cinque's theory...