Hornstein (1999) put forward two thought-provoking ideas that generated a stimulating debate on control: (a) A-movement out of CP complements of control verbs is a design feature of Universal Grammar as first suggested by Kuno (1976), and (b) obligatory control (OC) is an instance of A-movement. This article presents new evidence from Kirundi (Bantu) that supports (a) but defies (b), a paradox that is only apparent. Four sets of facts are discussed: antilocality in promise-constructions, control obviation in inverse OC constructions, passivization in transitive expletive OC constructions, and OC in long-distance inversion constructions. These facts are shown to challenge the movement account of OC while supporting (a) and the PRO-based account.
For over a decade, an insightful discussion has been going on between two competing syntactic views on obligatory control (OC).1 The first view holds that OC derives from coindexation (Agree) between a controller in the matrix clause and a PRO in the subject of the complement clause, as in (1b). See notably Borer 1989, Landau 2001, 2003, 2006, 2007, 2008, and Bobaljik and Landau 2009, among others.
Mary hopes to win.
[TP Maryi hopes [CP[TP PROi to win]]].
The second view holds that PRO does not exist; instead, OC derives from A-movement of the controller out of the infinitival CP complement. See notably Bowers 1973, 2008, Martin 1996, O’Neill 1997, Hornstein 1999, 2001, Manzini and Roussou 2000, Boeckx and Hornstein 2003, 2004, 2006a,b, and Takano 2010.2 This theory, dubbed the movement theory of control (MTC) by Boeckx and Hornstein, contrasts not only with an “Agree alone” theory of control but with any PRO-based account of OC.
Under the MTC, a sentence like (1a) is derived as in (2). The subject Mary merges with the embedded verb win and checks its external θ-role. It then raises to the matrix Spec,vP to check the external θ-role of the verb hope. Finally, it raises to the matrix Spec,TP to check its uninterpretable features against T.3 Under the minimalist view of traces as copies, all intermediate copies are traces, and the resulting chain has only one Case and two θ-roles.
(2) [TP Mary [VP Mary v+hopes [CP[TP Mary to [VP Mary v+win]]]]]
In this article, I present novel data from Kirundi that challenge the MTC account. I show that [+control] CPs are indeed transparent to A-movement, yet OC is not an instance of raising.
The article is organized as follows. In section 2, I discuss OC in Kirundi promise/permit-constructions, showing that an MTC account of the ‘promise’ reading violates the Minimal Link Condition. I refer to this as the locality puzzle. In section 3, I discuss control obviation in the inverse OC construction. The latter defies the MTC’s claimed virtue of deriving the complementary distribution between OC PROs and NOC (nonobligatory control) PROs. I refer to this as the distribution puzzle. In section 4, I examine passivization in transitive expletive OC constructions and show that the MTC fails to predict convergent passivized constructions of this type. I refer to this as the passivization puzzle. Section 5 is devoted to long-distance inversion in OC constructions, showing that raising out of the infinitival CP complement of a control verb is indeed permitted but the raised argument is not the controller. In section 6, I offer conclusions.
The Locality Puzzle
Hornstein (1999) notes that English promise-sentences like (3) pose problems for the MTC. Raising of John from the embedded Spec,TP to the matrix Spec,TP over the DP Mary violates Rosenbaum’s (1967) Minimal Distance Principle (MDP), which, in minimalist terms, falls under the Minimal Link Condition (MLC).
(3) Johni promised Maryj [CP PRO (=ti/*j) to leave].
Boeckx and Hornstein (2003) derive the refractory behavior of (3) from an allegedly special status of promise, following C. Chomsky’s (1969) observation that children show a significant delay in acquiring the control properties of promise.
For Boeckx and Hornstein, “what makes promise special is the existence of a ( possibly optional) null preposition heading the object” (2003:274). This null preposition (Pnull) renders promise similar to raising predicates with an experiencer,4 by allowing the subject of the embedded verb to raise across the object of Pnull without violating the MLC. I show below that the Pnull hypothesis cannot extend to the Kirundi counterpart of promise.5
Furthermore, the applied argument (Ben) in (4b) is always overt, never null, and asymmetrically c-commands the theme (Th) argument, as the following facts show.
Since Ben asymmetrically c-commands Th in (9), it may bind a variable within Th, but not conversely. The contrast between (7a) and (7b) follows. Furthermore, demotion of Subj in passive constructions renders Ben the closest goal to be probed and raised to Spec,TP, in accord with the MLC. The contrast between (8a) and (8b) follows.
With that asymmetry in mind, we can now tackle OC in the promise-construction illustrated in (10), a replica of (4b) with a nonfinite CP complement in lieu of the DP object. Importantly, (10) conveys two readings corresponding to the English promise/permit pair: the embedded null subject may refer to either the matrix Subj Petero or the Ben umwâna.
The question for the MTC is, how can the two readings be derived under a raising account?
The ‘permit’ reading is easily derived: the controlling argument umwâna ‘child’ raises from the position of PRO to some position in the matrix clause—say, Spec,ApplP in (9). The derivation obeys the MLC and the ‘permit’ reading follows.
The problem arises with the ‘promise’ reading. This requires raising of the controlling argument Petero from the PRO position in (10) to the matrix Spec,TP over the Ben argument umwâna merged in Spec,ApplP of the matrix verb. Yet this derivation is not viable, as it leads to an MLC violation. Incidentally, merging Ben below CP to avoid an MLC violation is ruled out on independent grounds. As shown by the passive sentences (11a–c) derived from (10), only Ben may raise to Spec,TP, as in (11a). CP-raising over Ben fails, as in (11b); Ben must be removed, as in (11c), an indication that Ben has a blocking effect on CP-raising and, by extension, a blocking effect on any constituent within CP.8
Therefore, the MTC account is unable to derive the ‘promise’ reading without violating the MLC. On the other hand, the PRO-based account easily derives both the ‘promise’ and ‘permit’ readings by allowing coindexation of ∆ (= PRO) in (10) with either the matrix Subj or Ben, under the structure in (9).9
The Distribution Puzzle
According to Boeckx and Hornstein (2003:272–273) “a virtue of the movement theory of control, perhaps its most alluring property, is that it completely accounts for the distribution of (obligatory) controlled PRO. In effect, one finds such PROs in precisely the configurations allowed by movement.” In this section, I show that this virtue is challenged by control obviation in Kirundi inverse OC constructions. I begin with an overview of the inversion construction, known as subject-object reversal in the literature on Bantu.
Sentence (12a) illustrates a neutral SVO sentence in Kirundi. The verb contains the italicized morpheme -ra-, dubbed “antifocus” in Ndayiragije 1999. This morpheme occurs only in root clauses and asserts the truth value of the proposition. Thus, (12a) may be uttered in out-of-theblue contexts as an answer to questions like ‘What happened?’ or yes/no questions such as ‘Did John read a book?’ or ‘Is it the case that John read a book?’, but not as an answer to wh-questions like ‘What did John read?’ or ‘Who read a book?’. Furthermore, the antifocus marker -ra- may be dropped, as shown in (12b). This omission triggers a contrastive focus reading on a postverbal constituent, the object in this case. (12b) can only be uttered in answer to a question like ‘What did John read?’ or ‘Did John read books or newspapers?’; it cannot be used to answer ‘Who read books?’ or ‘Did John read books?’.
Inversion constructions like (13) display three salient properties: (a) the antifocus marker -ra-must delete; (b) the inverted subject (Subj) receives a contrastive focus reading; and (c) the verb agrees with the fronted object (Obj), not with the logical Subj.
The TEC and inversion constructions share three important properties: (a) the mandatory deletion of the antifocus marker -ra- on the verb, (b) the contrastive focus reading on the logical Subj, and (c) the lack of verb agreement with the logical Subj. In the TEC (14), the verb agrees with a there-type pro expletive Subj. This agreement is spelled out on the verb by the locative agreement marker ha- prefixed to T.11
A unified analysis of OVS and the TEC is proposed in Ndayiragije 1999, which derives the two constructions from a shared property: Ā-movement of the logical Subj from Spec,vP to the (rightward) Spec of a Focus projection (FocP) located between TP and vP, as depicted in (15).
The head Foc filled by the antifocus marker -ra- in sentences with nonfocused constituents like (12a) is associated with an EPP feature that attracts the closest goal, Subj. This accounts for the focus reading of Subj in both OVS order and the TEC. After Ā-movement of Subj to Spec,FocP, T probes a matching goal to value its uninterpretable features: either Obj raises to Spec,TP, yielding OVS order, or a pro expletive is externally merged in Spec,TP, yielding the TEC.12
Rightward Spec,FocP in (15) was stipulated in Ndayiragije 1999 only to account for the rightmost position of Subj in OVS order and the TEC. Yet nothing in the present article hinges on linearization of Subj. Incidentally, alternative derivations with leftward movement may be worked out, including the one in (16) where Subj moves leftward to Spec,FocP, then the remnant vP moves to the Spec of a functional projection above FocP (call it Polarity Phrase, PolP). This remnant-vP movement might follow from some interpretive rule that forces nonfocused phrases to vacate the Foc domain, along the lines of Holmberg’s (1999) Generalization for Icelandic object shift and TECs.13
I will not delve further into word order issues here. I retain just two salient features of the analysis in (15) that will be relevant to the discussion of OC in OVS order and the TEC: (a) that the logical Subj in both OVS order and the TEC is in an Ā-position, Spec,FocP, located below TP; and (b) that the fronted Obj in OVS order lands in an A-position, Spec,TP. The following facts support (b).
First, the raised Obj triggers subject agreement on the verb (17a). Second, the raised Obj may undergo pro-drop (17b) and right-dislocation (17c). Third, OVS order selects the negation marker nti- (17d) that occurs in matrix clauses, but not in constructions involving Ā-movement to the left periphery.14 Fourth, the raised Obj in OVS order triggers multiple agreement in compound tense constructions (17e), a property of Kirundi A-chains but not Ā-chains. Finally, the fronted Obj in OVS order cannot reconstruct below Subj, as shown by the unavailability of the OVS reading of (17f ).15
8books 8sa-pst-read-asp John
[Lit.: the books read JOHN]
‘JOHN read the books.’
[Lit.: they read JOHN]
‘JOHN read them.’
[Lit.: they read JOHN, those books]
‘Those books, JOHN read them.’
8books neg-8sa-pst-read-asp John
‘JOHN didn’t read the books.’
8books 8sa-prs-be-on 8sa-prs-read-asp John
‘JOHN is reading the books.’
With this in mind, we can now tackle OC obviation in OVS order.
(19) reads as follows: the logical Subj Petero is externally merged in Spec,vP of the infinitival CP, then raises to matrix Spec,vP for θ-checking purposes, in accord with the MTC. We are interested in what happens next. Under the analysis of OVS order in (15) or (16), the logical Subj Petero now in the matrix Spec,vP undergoes Ā-movement to Spec,FocP to value Foc’s EPP feature; then the infinitival CP raises to Spec,TP. Note that the trace ti inside the raised CP is a copy of the intermediate trace t’i in the matrix Spec,vP, a nondeletable LF variable bound by the c-commanding operator Subj Petero Ā-moved to Spec,FocP. Backward control follows.
The challenge for the MTC comes from what follows: the trace ti inside the A-raised CP in (18b) may receive an NOC PRO reading (control obviation), as shown in (20a), with the corresponding derivation in (20b).18
Recall that for the MTC, OC PRO and NOC PRO (small pro in the MTC’s view) are in complementary distribution: NOC PRO occurs where OC PRO is not allowed, that is, where raising of the embedded Subj to a matrix θ-position is impossible. Under that view, the PROarb reading of (20a) (= (18b)) is unexpected. Indeed, the latter requires Merge of Subj Petero in the matrix Spec,vP of (18a) with an NOC PRO in the embedded Subj position. Yet this is not permitted, for two reasons. First, under the MTC, (18a), from which (20a) is derived, constitutes the canonical configuration for raising (i.e., OC PRO, not NOC PRO). Second, for the MTC to account for both OC PRO in (18b) and NOC PRO in (20a), Subj-raising out of the embedded CP in (18a) would have to be both possible and impossible, obviously a contradiction.
Therefore, OC obviation in Kirundi OVS (20a) defies Boeckx and Hornstein’s (2003) commendable attempt to derive the distribution of OC and NOC PROs. On the other hand, it is easily derived from the PRO-based account of OC, under the derivation in (20b): the embedded CP A-raised to Spec,TP need not reconstruct, making both the PROarb reading in (20a) and the controlled PRO reading in (18b) possible.
The Passivization Puzzle
*John was hoped to win the game.
John was persuaded to come to the party.
John was believed to have won the game.
Boeckx and Hornstein (2004) reply that the problem with (21a) resides not in control but in passivization, something a theory of passivization/Case should deal with, not a theory of control. They write, “Only verbs that take complements that can check Case (DPs and CPs) can be passivized” ( p. 436); the verb hope is not one of these verbs.19
What follows shows that the ungrammaticality of (21a) has nothing to do with Case. Instead, the problem resides in a θ-Criterion violation, unexpected under the MTC but predicted under the PRO-based account.
Under the PRO-based account, (22b) violates the θ-Criterion: the trace ti in (22a–b) is filled by PRO, externally merged in Spec,vP to satisfy the θ-Criterion within the embedded clause. The problem lies in the matrix clause: the external argument Petero raised to Spec,TP originates from the matrix Spec,vP, a position whose θ-role has been absorbed by the passive morphology. The result is an A-chain with valued Case but no θ-role. The derivation crashes as a θ-Criterion violation.
On the other hand, under the MTC, the θ-Criterion is met in (22b): the Subj Petero is θ-marked in the embedded Spec,vP prior to raising to the matrix Spec,TP for Case feature valuation. For the MTC, then, the ungrammaticality of (22b) has to do with something other than the θ-Criterion, presumably a restriction on passivization as suggested for English (21a). Yet what follows shows that passivization is not to blame here.
Under the MTC and the unified analysis of the TEC and OVS order proposed in (15), (23) is derived as follows: the controlling argument Petero originates from the embedded Spec,vP, then raises to the matrix Spec,vP to value its external θ-role, then to Spec,FocP to value Foc’s uninterpretable features. Finally, a there-type pro expletive is externally merged in Spec,TP to value T’s uninterpretable features. The OC reading follows.
Assuming that this derivation is correct, what happens if the matrix verb in (23) is passivized? Surprisingly, the derivation converges as shown in (24), with a PROarb reading of the embedded Subj and a focus reading on the embedded CP.
The well-formedness of (24) clearly rules out passivization/Case as the source of the ungrammaticality of (22b). Therefore, the passivization puzzle in both Kirundi (22b) and English (21a) remains unaccounted for under the MTC. On the other hand, it is nicely resolved under the PRO-based account as a θ-Criterion violation: the controlling argument Subj raised to Spec,TP in (22b) and (21a) originates from the matrix Spec,vP, a non-θ-marked position, thus creating an A-chain with valued Case feature but no θ-role. The derivation crashes as a θ-Criterion violation under the PRO-based account of OC.
What follows provides a final empirical argument that favors the PRO-based account over the MTC, as the latter may overgenerate nonconvergent derivations.
Under both the MTC and the PRO-based account, (25) crashes for Case reasons: the DP Petero occurs in a position where it cannot value its Case feature.20 Assuming that this is correct, consider (26), which is (25) with the DP Petero Ā-moved to Spec,FocP, following the analysis of the TEC and OVS order in (15).
The question for the MTC is, why is (26) ungrammatical? Obviously not for Case reasons since the Ā-moved DP Petero values its Case feature in Spec,FocP as it does in (27), a replica of (26) with an active matrix verb.
Under the MTC, (26) should converge as do (27) and (24), yet it does not. On the other hand, the PRO-based account correctly predicts the ungrammaticality of (26) if only PRO can value the Case and agreement features of control C-T. Under this view, the DP subject Petero merged in Spec,vP of the embedded CP in (26) fails to move out of CP because it is first probed by control C-T whose need to check null Case and agreement features spells it as a silent category PRO unable to raise further to value the Foc’s EPP feature.21 Accordingly, (27) must contain PRO in the embedded Spec,TP rather than a trace, and the same is true for all instances of Kirundi OC where the MTC would posit raising out of CP.
An anonymous reviewer wonders whether Boeckx and Hornstein (2003) could not account for (26) by assuming that Kirundi CPs need Case and that it is impossible to have both a DP and a CP in a passivized TEC like (26). Such a solution is weak for the following reason: it wrongly predicts that the passive sentence (26) could be rescued by raising the embedded CP to the matrix Spec,TP to value its Case feature, if no expletive is merged in the subject position, or by raising the DP subject Petero to the matrix Spec,TP and Ā-moving the embedded CP to Spec,FocP, to value its Case feature against Foc. Yet, as shown in (28a–b), both derivations crash. Therefore, the ungrammaticality of (26) has little to do with CP’s alleged failure to value its active Case feature.
The above facts clearly show that the MTC lacks appropriate tools to derive convergent passivized OC constructions in Kirundi, which however are readily derived under the PRO-based account.
Raising out of Control
Raising out of the complement clause of control verbs is permitted in Kirundi, as illustrated by (29b), an inversion construction in which the Obj of the embedded clause in (29a) has raised to the matrix clause. I refer to (29b) as long-distance inversion (LDI).
(29b) displays three defining features of the inversion construction discussed in section 3. First, the fronted Obj agrees with the matrix verb. Second, the antifocus marker -ra- on the matrix verb in (29a) must drop in (29b). Third, the logical Subj occurs at the rightmost edge of the clause and receives a contrastive focus reading. Importantly, the OC PRO reading of (29a) is and must be preserved in LDI (29b).
In the remainder of this section, I demonstrate that (29b) is a true instance of long-distance A-movement out of the infinitival CP complement of control verbs. To do so, I show that (a) the fronted Obj is located in an A-position, (b) the matrix verb is a true control verb with no mixed properties of control and raising verbs, and (c) the fronted Obj is not externally merged in its surface position.
The fronted Obj in LDI displays all the A-chain properties of inversion constructions discussed in section 3. First, it freely undergoes pro-drop (30a) and right-dislocation (30b). Second, it triggers multiple agreement in compound tense constructions (30c). Third, it does not reconstruct (30d). Fourth, LDI selects the negation marker nti- (30e) that only occurs in root clauses without Ā-movement to the left periphery.
As a further matter, LDI obeys the MLC. Thus, (31b) shows that the Ben argument of the matrix clause in (31a) blocks LDI of the embedded Obj. (31c) shows that Ben-cliticization does not make Obj-raising possible.
All these facts follow if the fronted Obj in LDI heads an A-chain.
Control versus Raising
The matrix verb in LDI is a full control verb with no properties of raising verbs. Thus, consider a verb like -ank-, which may be used as a control verb meaning ‘refuse’ or a raising one meaning ‘fail’, as shown in (33).
As a raising verb, -ank- displays standard properties of raising constructions, including the ability to take a [–animate] Subj, as in (35a–b) from (34a–b), and an it-type pro expletive, as in (35c), where the Subj agreement prefix bi- spells out Ф-features of the null pro expletive in Spec,TP.
Ico giti ki-á-ra-gu-ye.
7that 7tree 7sa-pst-af-fall-asp
‘That tree fell.’
[Lit.: the rain fell]
Control verbs that allow LDI lack these properties of standard raising verbs. Thus, the verb -emer- ‘agree’ imposes a selectional restriction on its external argument: the latter must be [+animate] (witness the ungrammaticality of (36a–b), where a [–animate] argument fills the Subj position, and (36c), where an it-type pro expletive Subj is merged instead and spelled out by the agreement marker bi- on V).
Furthermore, whereas raising verbs resist passivization, as shown in (37b), control verbs do passivize, as shown in (38b), an indication that the infinitival complement of control verbs is a CP rather than an IP, if Rizzi’s (1996) extractability test holds.
Control versus Restructuring
Wurmbrand (1998, 2004) argues that OC constructions are monoclausal structures with no PRO at all, a proposal reminiscent of previous restructuring accounts such as those of Rizzi (1982) and Haegeman and Van Riemsdijk (1986), among others. Under her analysis, OC constructions like (39a) are instances of lexical restructuring where the matrix verb c-selects a VP complement, as schematized in (39b).
Mary hopes to win.
[TP Mary hopes [VP to win]].
A lexical restructuring analysis is not suited to Kirundi OC constructions, as it fails to derive two important properties of inverse OC constructions already discussed: verb agreement and control obviation.22 First, consider verb agreement. In (40a), the matrix verb selects a tensed CP complement. The latter has raised to Spec,TP in the inversion construction (40b), triggering bi-agreement on the verb.
Furthermore, the inversion construction in (41b) displays backward control (OC PRO). This is predicted under the restructuring account. Yet (41b) also allows an NOC PRO reading (control obviation), as shown in (42).
External Merge of Obj
The fronted Obj in LDI is not externally merged in the matrix Spec,TP. In other words, a sentence like (43b) should not be assimilated to, say, tough-constructions with operator movement in the embedded Spec,CP, as depicted in (43c).25
Sentence (44) illustrates Kirundi tough-constructions.
Tough-constructions like (44) display three properties that are never found in LDI. First, they allow the antifocus marker -ra- on the matrix verb, as shown in (44), an option ruled out in LDI and other inversion constructions. Second, tough-constructions freely permit it-type expletives in the matrix Spec,TP, as in (45a). Inversion constructions do not; witness the ungrammaticality of (45b).
Third, tough-constructions may take a Ben argument licensed by the applicative head -ir- on the matrix verb, as in (46b), where Ben is an object clitic. LDI does not have that option; witness the ungrammaticality of (47b).
All these asymmetries follow nicely if the fronted Obj in LDI undergoes raising out of the embedded CP. We thus have compelling evidence here that the infinitival CP complement of control verbs is transparent to A-movement, but only objects can raise out of CP.
This article evaluated the explanatory power of the MTC and the PRO-based account of OC using novel data from Kirundi. The PRO-based account consistently won against the MTC. The latter failed to derive subject control in promise/permit-constructions, control obviation in inverse OC constructions, and passivization in transitive expletive OC constructions.
Furthermore, the discussion of long-distance inversion constructions revealed that raising out of control C-T is a licit operation of CHL (the computational system of human language); it is feature-checking requirements (null Case and agreement features) of control C-T that prevent Subj from raising out of the embedded Spec,TP. The occurrence of PRO but not overt DPs in that position receives a principled explanation.
Finally, beyond the OC debate, A-transparency of [+control] C-T leads to a consequence for Chomsky’s (2007, 2008) phase theory: the phasehood of CP must be relativized, as first suggested in Ndayiragije 2005. Indeed, if the Phase Impenetrability Condition holds, then [+control] C-T cannot be a phase. If it were, LDI would lead to improper movement from an A-position to an Ā-position (the embedded Spec,CP) in compliance with the Phase Impenetrability Condition, then to an A-position, Spec,TP of the matrix clause.
Various aspects of this research have been presented at Cornell University, MIT, Leiden University, the University of Missouri, and the University of Toronto. I thank the audiences for their comments. I also thank Gabriela Alboiu, Jonathan Bobaljik, Cedric Boeckx, and Norbert Hornstein. I am especially grateful to Idan Landau and two anonymous referees for valuable comments and suggestions that have strengthened the arguments and clarified the exposition. Remaining errors are my responsibility.
3 Whether the infinitival complement in (2) is a CP or a TP remains an open question. For the sake of simplicity, I will use the CP notation.
4 As in (i)–(ii) from Boeckx and Hornstein 2003:274.
(i) Johni seemed to Mary ti to be tired.
(ii) Johni struck [Pnull Mary] ti as tired.
5 The Pnull hypothesis is also challenged by Landau (2007) on English-internal grounds. Other researchers such as Martin (1996) and Takano (2010) have claimed that promise-type subject control does not involve OC, hence is not derived by movement of the controller. The latter is instead base-generated in the Subj of the matrix clause and the embedded Subj is an independent null pronominal. In section 3, I show that base generation of the controller in the matrix clause extends to any Kirundi OC verb, including verbs such as ‘permit’, ‘try’, ‘remember’, ‘love’, ‘hate’. The discussion focuses on the permit/promise-construction.
6 The following abbreviations are used in the glosses: numerals (1, 2, etc.) indicate Bantu noun class; af = antifocus marker; appl = applicative; asp = aspect; c = complementizer; dem = demonstrative; inf = infinitive; neg = negation; om = object marker; op = object pronoun; pass = passive; prs = present tense; pst = past tense; refl = reflexive; sa = subject agreement.
7 See also Pylkkänen 2008 for an extension to other applicative constructions in Bantu.
8 It is worth noting that the ‘promise’ reading of (11a) is more felicitous under a ‘shared reading’ interpretation, where the child and some implicit agent are going to read the book together, or if the implicit agent is overtly realized, as illustrated in (i). It remains to be explained why those are the only readings available in the Kirundi and English counterparts.
9 As correctly pointed out by an anonymous LI reviewer, allowing unconstrained coindexation of ∆ (= PRO) with Subj over Ben in (10) leaves unexplained cases like (ib) where PRO-indexing with Subj across an indirect object is barred. See Landau 2001 for an account of the locality of Agree in promise OC constructions and others that appeals to the Principle of Minimal Compliance.
John asked [PRO to leave].
Johni asked Maryj [PROj/*i to leave].
10 Note that Obj must precede Subj in the TEC. Furthermore, VP modifiers occur between Obj and Subj in both the TEC and OVS sentences, as in (i)–(ii), an indication that Obj and Subj in (14) are not in multiple Specs of one single head—say, vP.
11 (14) becomes ill-formed if the verb agrees with the logical Subj or with the nonfronted Obj, as shown in (i) and (ii), respectively.
12 Note that the Ā-moved Subj does not value its Case/Ф-features against T since it does not agree with T. In Ndayiragije 1999, I argue that Subj’s uninterpretable features are instead valued by Foc, a reasonable assumption on semantic and syntactic grounds given the well-known link between structural Case and specificity/focus as suggested by Watanabe (2003) and Miyagawa (2005) for Japanese. An anonymous LI reviewer wonders how the focus-driven movement in (15) accounts for the focus reading of Obj in SVO (12b). If the Foc has an EPP feature that attracts the closest DP, why doesn’t Subj in Spec,vP block focusing of Obj? The answer is straightforward: Subj in Spec,vP indeed blocks Obj-focusing: Subj must raise out of Spec,vP for Obj to move to Spec,FocP. Thus, in (12b), Subj has moved to Spec,TP, as evidenced by word order and subject agreement, allowing the Obj to move to Spec,FocP. In contrast, Subj is still in Spec,vP in the TEC (i), and Obj-focusing is blocked. Compare (i) with (14).
The same reviewer notes that my account seems to assume countercyclic movement: Obj moves to Spec,FocP only after a higher head (T) is merged and attracts the subject. Furthermore, the account does not explain why Subj has to move to Spec,TP in order for Obj to be focused, and why the trace of Subj does not block Obj from moving to Spec,FocP. In other words, why is sentence (i) ill-formed, with Subj in Spec,vP? Both issues may be resolved under the phonological condition associated with the Phase Impenetrability Condition in Chomsky 2001. According to the phonological condition, the subject must move out of the inner Spec of the vP phase for the object to raise to the outer Spec, in accord with the Phase Impenetrability Condition, then to a higher A- or Ā-position. Under that view, Subj-raising is a precondition for every Obj-movement out of vP. The ungrammaticality of (i) follows.
13 Other possible derivations are discussed in Ndayiragije 1999 that prove untenable on empirical grounds. One of them assumes that Subj remains in situ and Obj raises in the outer Spec,vP, under the multiple-specifier hypothesis. This solution fails to account for reflexive binding in (ii) and variable binding in (iv), where the binder follows the bindee.
‘John hit himself.’
14 In those constructions, the negation marker -ta- is used instead of nti-.
15 This contrasts with Ā-movement, which allows reconstruction of the fronted Obj, as shown by Obj-clefting in (i).
16 That the raised constituent in (18b) is a CP is confirmed by subject-verb agreement: the noun class 8 bi- is the agreement marker for a CP subject, be it a nonfinite CP (ia), a finite CP (ib), or an it-type pro expletive (ic).
‘Buying a book is possible.’
‘That Peter buys a book is possible.’
‘It is possible that Peter buys a book.’
Furthermore, it is sometimes claimed that Bantu infinitives like the one in (ia) are ambiguous between infinitival and gerundive/nominal structures. In Kirundi, the nominal counterpart to the bracketed CP in (ia) would require merger of the vowel prefix u- on the infinitive recently analyzed as a D head by Ndayiragije, Nikiema, and Bhatt (2010). Incidentally, as pointed out by an anonymous reviewer, accusative Case is consistently available in bracketed constituents like those in (1a) and (18a–b); hence, a nominal analysis is ruled out and a thematic, silent Subj must be assumed in such constructions.
17 For the sake of simplicity, here I omit the polarity projection and the remnant-vP movement shown in (16). Also, it is worth noting that an alternative derivation with rightward movement to Spec,FocP as in (15) yields the relevant chains in (18)—that is, an Ā-chain for the focused Subj and an A-chain for the raised CP.
20 Kirundi thus differs from a language like Romanian, which according to Alboiu (2003) allows overt DPs in the embedded Subj of control verbs, a subjunctive form on the CP complement, and agreement marking on the embedded T, three properties that can only occur in Kirundi tensed CPs, not in OC constructions.
21 The inability of PRO to value Foc’s EPP-feature follows from a general PF restriction imposed on focused phrases: only constituents that can bear accent can be focused. Strong pronouns and lexical NPs can be assigned accent, hence are potential goals to be probed by Foc. Weak/Clitic pronouns and empty categories such as PRO and pro are not, hence cannot be probed by Foc; witness the grammaticality contrast between (ic) and (id). (ic) involves a strong object pronoun, (id) a clitic one. Looking at the verb in (id), note the absence of the antifocus marker -ra-, which signals focus activation. Compare (id) with (ie).
22 Clitic climbing, a key argument in restructuring analyses, cannot be used as a test here, as it doesn’t exist in Kirundi.
23 The same reasoning holds true for functional restructuring accounts along the lines proposed by Cinque (2004).
24 As pointed out by an LI reviewer, proponents of restructuring accounts might claim not that restructuring would have to be both possible and impossible, as I assert, but that restructuring is optional, just as reconstruction is optional under my account of OC PRO and NOC PRO in (41b)/(42).
Yet, as shown by the passivization facts in (i)–(ii), genuine control verbs like the one in (41a) resist restructuring. Sentence (ia) illustrates optional restructuring for the matrix verb -gi-, which is ambiguous between a control reading ‘to go’ and a (restructuring) future tense reading, very much like its English counterpart. (ib) shows that passivization of the embedded verb in (ia) is allowed, but only under a restructuring reading of the matrix verb, not under a control reading. Sentence (iia) illustrates an OC verb in the matrix clause. Sentence (iib) shows that passivization of the embedded verb of (iia) is impossible, a strong indication that genuine control verbs are not subject to optional restructuring in Kirundi.
The ungrammaticality of (iib) arises from a θ-Criterion violation: the matrix verb requires a [+human] DP argument to value or discharge its external θ-role; the raised argument iyo nzu ‘that house’ does not meet the requirement.