Against the Movement Theory of Control: Another Argument from Icelandic
Jim Wood · Linguistic Inquiry · 2012
In this squib, I present an empirical argument from Icelandic that the movement theory of control (MTC) is on the wrong track—or at best, does not achieve what it purports to achieve. I depart from previous discussion, which has focused on the fact that Icelandic PRO bears case, and show that the MTC would require violating locality constraints on A-movement that raising infinitives do not violate. Since the goal of the MTC is to account for the properties of control sentences on the basis of mechanisms independently needed within the Minimalist Program, this difference in the locality domains of raising and control dependencies constitutes another challenge for the MTC.The conclusion from the recent debate on Icelandic obligatory control (OC) (Boeckx and Hornstein 2006, Sigurðsson 2008, Bobaljik and Landau 2009, Boeckx, Hornstein, and Nunes 2010b) is often taken to be that Icelandic shows that OC is not movement. However, this is not quite accurate. What Icelandic shows is that control is not Case-driven movement (unless Case has nothing to do with case). This is so for two reasons. First, in their reply to Bobaljik and Landau (2009), Boeckx, Hornstein, and Nunes (2010b) fail to address the strongest argument against Case-driven movement: that the nominative case borne by PRO is the same structural nominative as that seen in finite clauses, even by Boeckx and Hornstein’s (2006) criteria (including, crucially, full agreement with passive participles).1 No convincing reason to believe otherwise has been presented.Second, even in their own analysis, Boeckx, Hornstein, and Nunes (2010b) make it clear that it is agreement with a Ф-complete probe that is important, not Case valuation. The need to enter an Agree relation with a Ф-complete probe is what drives movement of “quirky”-Case-marked DPs into θ-positions (under Enlightened Self- Interest; Hornstein 1999), and movement into θ-positions wipes out the Case value of the moving DP.2Therefore, a notational variant of Boeckx, Hornstein, and Nunes’s (2010b) analysis might say that DPs need “Ф-complete valuation” rather than Case valuation, and then draw some strong formal connection between certain morphological case values and agreement with different Ф-complete probes, while maintaining the position that control sentences are derived by A-movement. While raising and control would then be similar in that they both involve A-movement out of an infinitive, they would differ precisely where Boeckx, Hornstein, and Nunes (2010a,b) claim they do: only the latter involves movement into a θ-position. This claim might seem unfalsifiable, as has been noted (see Sigurðsson 2008:418–419), but it could, in principle, be correct.However, I will argue that even this version of the MTC cannot be right for Icelandic. The facts leading to this conclusion have to do with the locality domains for movement rather than the question of PRO being case-marked. It can be shown that if control were movement, it would in some cases involve movement out of a domain from which no other kind of movement can take place, whether A-movement or Ā-movement.Icelandic control infinitives can often occur with an optional (case-marked) pronoun það ‘it’, a construction analyzed by Thráinsson (1979:110–111) as object extraposition (1).3As Thráinsson shows, Ā-extraction can occur out of a control clause only when það is not present, as illustrated for topicalization in (2).We thus see a case where Ā-extraction is impossible out of a control clause—a case that under the MTC would have to allow A-movement out of that same clause. A-movement past this pronoun would be unexpected. If the pronoun is the head of the clause (e.g., a D taking a CP sister), then A-movement out of the clause would involve an A-over-A violation. If the pronoun is not the head of the clause (e.g., a DP in Spec,CP), then A-movement past it would violate Minimality; assuming that the pronoun c-commands the moving DP/PRO, it would thus be closer to any landing site than the latter.It has been pointed out in connection with this that standard raising infinitives do not allow það (Andrews 1990, Roehrs 2005), presumably at least in part for the reason just cited.However, there is another difference between the structure of these raising infinitives and that of control infinitives: the former lack the complementizer að ‘to/that’, and it could be claimed that this independent difference might explain the impossibility of the optional pronoun. However, this is refuted by aspectual verbs, which, as shown by Sigurðsson (1989:68–71), are like standard raising verbs in that they preserve the “quirky” case assigned by the embedded verb (see also Thráinsson 1986). In (4), the embedded verb leiðast ‘bore’ assigns dative case to its subject, which is preserved under A-movement past the aspectual verb.These aspectuals do not allow an extraposition það either, so its impossibility in (3) cannot be directly attributed to the lack of the infinitive marker að.The impossibility of an optional pronoun in raising contexts suggests that standard Case-driven A-movement is not possible past such a pronoun. This is consistent with the fact that A-movement is in general governed by stricter locality constraints than Ā-movement.Since neither Ā-movement nor A-movement (of the sort seen in raising infinitivals) is possible past the optional pronoun, the MTC would seem to have two options for analyzing sentences like (1) when the pronoun is present.4 First, one could tie the availability of movement past the pronoun in control sentences to the fact that it is targeting a θ-position. Second, one could analyze such cases as instances of nonobligatory control (NOC), which in the MTC involves a null pro that is not related to its antecedent by movement (Boeckx, Hornstein, and Nunes 2010a:195ff.). In this section, I will claim that the first option is untenable and the second has no independent support.Turning to the first possibility, one might suppose that the optional pronoun is nonthematic.5 Thus, A-movement past it might be possible if such movement were thematically driven, rather than Case-driven. This would explain why standard raising contexts do not allow such a pronoun while control contexts do. However, the serious problem faced by this approach, from the standpoint of the MTC, is that this pronoun can show up with “quirky” dative or genitive case, and the same paradigm obtains (Thráinsson 1979:110–111, 228–229). In (6), the optional pronoun is dative, and, as (7) shows, extraction is still impossible when it is present.This is a problem because the MTC crucially ties Icelandic “quirky” case to θ-positions. The reason that a PRO—or DP-trace, in the MTC—can have a different case from the controller is that control involves movement into a θ-position, θ-positions are involved in assigning “quirky” Case, and so the Case value of the moving DP is wiped out when it moves into a θ-position. But this account of “quirky” case and the distribution of case in control clauses leads to the conclusion that the optional pronoun in (6) is in a θ-position. This is further supported by the fact that the choice of case on the pronoun depends on the lexical verb (dative for fresta ‘postpone’, accusative for ákveða ‘decide’, etc.); fresta ‘postpone’ marks nonclausal direct objects in the dative as well (example from Thráinsson 1979:228).Therefore, one cannot say that movement past the pronoun is possible because the latter is nonthematic—it appears to be thematic insofar as its case marking is considered (and case marking is the domain of thematicity that is relevant to the MTC’s explanation of the Icelandic facts).The second possibility open to the MTC is to analyze the examples with the optional pronoun as NOC. In this case, they simply would not be relevant to the MTC, since as noted by Boeckx, Hornstein, and Nunes (2010a:195), the analyses of OC and NOC do not, in principle, need to be unified. To see this, consider the English examples in (9).6In (9a), which exemplifies OC, PRO cannot be controlled by the non-c-commanding antecedent Olaf. (9b), however, has in addition a reading where the therapist decided that giving up vodka was the best course of action for Olaf. This kind of control by a non-c-commanding antecedent is a diagnostic for NOC. However, when a wh-phrase is extracted, as in (9c), this reading is lost. A possible explanation for this fact is that (9b) is ambiguous between OC and NOC, but that extraction is only possible in OC contexts; so (9c), with extraction, only allows the OC reading/structure. Extending this reasoning to the Icelandic data in (1) and (6) would involve saying that the optional pronoun is only compatible with NOC structures; thus, extraction is impossible.This possibility is hard to refute completely, since the NOC readings are generally a superset of OC readings. However, it should be pointed out in this respect that sentences with the optional pronoun do not pass positive tests for NOC. First, the sentence in (10) only allows a reading where the therapist quits drinking.That non-c-commanding antecedents of this sort generally cannot control PRO in these constructions is supported by (11). The floating quantifier einn ‘alone’ agrees with PRO in number, gender, and case.PRO would pick up its number and gender features from its controller, and its case clause-internally. If Jóns could serve as a non-c-commanding controller of PRO, the floating quantifier einn would be grammatical, with the reading that Jon’s mother decided that he would go alone, contrary to fact. In cases of NOC, however, the form of the quantifier can vary with the intended referent of PRO. It can be “speaker-inclusive,” in Sigurðsson’s (2008:424) terms, so that a female speaker can use the feminine form and a male speaker the masculine form.7In NOC contexts such as this one, the form of the quantifier can vary, unlike in (11), where only the thematic subject of the matrix verb can determine the number and gender features of PRO and the quantifier. Similarly, in the NOC sentence in (13), either the c-commanding faðir Maríu ‘Mary’s father’ or the non-c-commanding Maríu ‘Mary’s’ can control PRO, with concomitant differences in the availability of masculine and feminine quantifier forms.Second, PRO cannot be controlled long-distance in cases like (1) and (6), whether the pronoun is present or not. In (14), the embedded reciprocal requires a plural, feminine, local antecedent. Since þær ‘they.fem.nom’ cannot control PRO to provide one—only þjálfarinn ‘the coach’ can control PRO—the sentence is ungrammatical.8In contrast, the NOC sentence in (15) allows þær ‘they.FEM.NOM’ to control PRO long-distance.Third, there is no alternation between a null pronoun and an overt argument the way there is with at least some NOC contexts (see (16b)).As pointed out by an anonymous reviewer, the impossibility of (16a) does not make the strongest argument, since infinitives headed by að never allow an overt subject in Icelandic.9 However, the general point is that there does not seem to be any positive evidence that sentences like (1) and (6) involve NOC. Therefore, the problem of accounting for the silence, distribution, and interpretation of PRO remains for such cases.The main goal of the MTC is to understand the silence and interpretation of PRO in terms of mechanisms that are independently needed: interpretation comes from θ-positions, and PRO is silent because of chain reduction of A-chains; copies of A-moved elements are usually silent. The facts presented here show that this account of the silence of PRO cannot work in general without saying that the locality domain for A-movement into a θ-position is different from the domain for A-movement into a non-θ-position. This movement would violate either the A-over-A Principle, if the optional pronoun is taken to head the infinitive, or Minimality, if the pronoun occupies a position outside of the infinitive.The domains of control dependencies in these cases are more reminiscent of those spanned between overt pronouns and their antecedents than of those spanned by A-movement relations; a movement approach to control like that found in Kayne 2002 is conceivable, though we still need an account of when, where, and why PRO is silent.10 Therefore, even if control is movement, we cannot simply account for it on the basis of independently understood principles, as the MTC aims to do. Even if these facts were derivable on the movement approach—for example, by moving the embedded DP out of the infinitive before extraposition—why extraction is not possible would be a mystery.This argument from locality is independent of the question of whether the landing site is a θ-position or not, which, according to Boeckx, Hornstein, and Nunes (2010b:112), is the “single difference” that “should underlie the reported differences between control and raising.” It is not at all clear how one could exploit a θ-position landing site to account for this locality difference, especially given that the pronoun is apparently thematic in the relevant way.11 The conclusion drawn here is a conservative one, but an important one: we do not yet fully understand the difference between raising and control, and we cannot account for the properties of control constructions on the basis of mechanisms independently needed for A-movement.