Intermediate positions in Dinka: Evidence for feature-driven movement Keywords: successive cyclicity - intermediate movement - Dinka - agreement Summary: This talk presents novel evidence from Dinka (Nilotic; South Sudan) for the idea that intermediate movement steps are feature-driven (Chomsky 1995; McCloskey 2002; Preminger 2011). I first show that Dinka offers support for successive-cyclic movement through intermediate Spec-CP, accompanied by feature checking. Specifically, intermediate movement satisfies the V2 property of intermediate clauses and has a morphological reflex at each edge. I then provide further evidence for this approach from Dinka ϕ-agreement, which consistently favors targetting ¯ A-moved phrases. This bias, I argue, is the result of the fact that such phrases undergo feature checking at C. The data presented here supply an argument against analyses in which intermediate movement is triggered in a special way, after all other syntactic operations within a phase have applied (e.g. Chomsky 2000, 2008; Boˇskovi´c 2002, 2007). Dinka is V2: Dinka is V2 in all finite main and embedded clauses (Andersen 1991, 2002), which I will take to be established at the CP-level, following Van Urk and Richards (to appear): ` G`Oc al´eth. ` (1) C`an a-b´ı Bol (2) Al´eth aa-b´ıi C´an k´e G`Oc Bol. ¨ ¨ Can dcl.sg-fut Bol buy clothes clothes dcl.pl-fut.ns Can pl buy Bol ‘Can will buy Bol clothes.’ ‘Clothes, Can will buy Bol.’ ¨ ¨ As evident in (2a–b), the XP that moves to Spec-CP governs ϕ-agreement on what I call the tenseforce prefix (a(a)-), which attaches to the highest verb or auxiliary. (The plural marker ke at the vP edge is associated with movement out of vP, as discussed in Van Urk and Richards, to appear.) Spec-CP is an intermediate landing site: Dinka provides evidence that Spec-CP is an intermediate landing site (Chomsky 1977 et seq.). In long-distance extraction, every intermediate Spec-CP must appear empty, in apparent violation of V2 (Van Urk and Richards, to appear): (3) YeNu´ yukku´ lu´eel, e-cukku´ c´am? (4) *YeNu´ yukku´ lu´eel, w`O c´e c´am? what imp.1pl say pst-prf.1pl eat what imp.1pl say we prf eat ‘What are we saying we had eaten?’ ‘What are we saying we ate?’ This follows if this position is the target of intermediate movement, so that Spec-CP is occupied at the relevant stage of the derivation, satisfying the V2 requirement of the embedded clause. The Dinka tense-force prefix: The Dinka tense-force prefix provides evidence that intermediate and terminal wh-movement require feature checking, since its morphological shape depends on ¯ whether Spec-CP is the target of A-movement (cf. McCloskey 2002 on Irish): (5) K`eek aa-ke-nhi`ar Go` n-d´en. (6) Yey´ıN`a e-ke-nhi`ar Go` n-d´en? ¨ ¨ they dcl.pst-pl-love house-3pl who.pl pst-pl-love house-3pl ‘They loved their houses.’ ‘Who all loved their house?’ To be more precise, this prefix marks tense and ϕ-agreement and comes in both a declarative and wh-series, as is demonstrated by the paradigms in (7) and (8). sg pl sg pl present past (7) (8) decl a- aadecl e- aa-kewh ∅- ∅wh e- e-ke¯ Feature-driven movement: I adopt the idea that all intermediate and terminal A-movement operations are triggered by a probing feature, such as an unvalued wh-feature (Chomsky 1995; McCloskey 2002; Preminger 2011). Following Preminger (2011), such probes are taken to be present on every phase head, under the proposal that a failure of Agree does not induce ungrammaticality. As a result, a wh-probe on a declarative C or v may fail to find a target without inducing a crash. I then propose that the declarative series spells out the failure of wh-probing on C, adopting the view that a failure of Agree may be registered morphologically (Preminger 2011; Halpert 2012). In contrast, the wh-version of this prefix registers the valuation of C’s wh-probe. 1

Crucially, intermediate movement steps also trigger the wh-series of the prefix: ´ en ke g´am kit`ap]? (9) Yey´ıN`a e-ke-y´ıi ke t`aak, [e-ke-c´ıi Ay` ¨ book who.pl pst-pl-impf.2sg pl think pst-pl-prf.ns Ayen pl give ‘Who all were you thinking that Ayen had given a book to?’ This follows if intermediate movement, like terminal wh-movement, is triggered by a wh-probe on the intermediate C. The intermediate C in (9) registers wh-agreement with the wh-phrase. Intermediate movement feeds agreement: Evidence for this approach comes from Dinka agreement facts. The tense-force prefix always agrees with the moved XP. This is evident in topicalization, in (1) and (2), but also under wh-movement: ´ en ke g´am kit`ap? (10) Ay´en e-c´e k`eek g`am kit`ap. (11) Yey´ıN`a e-ke-c´ıi Ay` ¨ ¨ book Ayen pst-prf them give book who.pl pst-pl-prf.ns Ayen pl give ‘Ayen had given them a book.’ ‘Who all had Ayen given a book to?’ Strikingly, this is true both for intermediate and terminal movement, so that even intermediate movement steps trigger ϕ-agreement on the tense-force prefix of the intermediate clause: ´ en g´am kit`ap]? (12) a. YeN`a e-y´ıi t`aak, [e-c´ıi Ay` ¨ book who pst-impf.2sg think pst-prf.ns Ayen give ‘Who were you thinking that Ayen had given a book to?’ ´ en ke g´am kit`ap]? Ay` b. Yey´ıN`a e-ke-y´ıi ke t`aak, [e-ke-c´ıi ¨ book who.pl pst-pl-impf.2sg pl think pst-pl-prf.ns Ayen pl give ‘Who all were you thinking that Ayen had given a book to?’ ¯ In short, ϕ-agreement in Dinka consistently prioritizes phrases undergoing A-movement. Adjunct extraction: That ϕ-agreement and wh-movement go together in Dinka is observable also in adjunct extraction. When an adjunct PP undergoes wh-movement, it moves as a nominal (P is removed by means of independently available mechanisms), and is targeted for ϕ-agreement: (13) Ye b`EEi ko´ ny`ank´ai ke w`anm´ath tu`OOc]? e-ke-y´ıi ke t`aak, [e-ke-c´enn´e ¨ send ¨¨ q villages which pst-pl-impf.2sg pl think pst-pl-prf.obl sister ¨ pl brother ‘Which villages had my sister sent my brother to?’ Favoring wh-phrases: I propose that the ϕ-probe is hosted on C. In addition, I posit an economy condition, following Pesetsky and Torrego (2001) and Van Urk and Richards (to appear), which forces a probe to check all of its features against a single goal if possible: (14) Multitasking: At every step in a derivation, if a probe can trigger operations A and B, and the features checked by A are a superset of those checked by B, the grammar prefers A. Multitasking ensures that if a C head is faced with the choice of ϕ-agreement with the subject DP or both ϕ- and wh-agreement with a wh-phrase undergoing long-distance movement, it must choose to value all of its features against the wh-phrase, because this is the more economical derivation. Under this view, what makes Dinka special, aside from the fact that it allows ϕ¯ agreement to target the same phrase multiple times, is that a ϕ-probe and A-features are hosted ¯ on the same head, C, causing ϕ-agreement to favor targeting A-moving phrases. Last resort approaches: There is a family of approaches, which differ amongst themselves, in which intermediate movement is construed as a last resort operation, triggered in a special way, which happens after all other syntactic operations in a phase have taken place (e.g. Chomsky 2000, ¨ 2008; Boˇskovi´c 2002, 2007; Heck and Muller 2003). The Dinka facts argue for the opposite: intermediate movement is just like other movement operations, and can be established simultaneously with other probe-goal relations, in this case ϕ-agreement. In this way, Dinka offers evidence that all instances of movement, whether terminal or intermediate, are established in the same fashion: by means of a probe-goal relation between a head and the moving phrase. Selected references: Preminger, O. 2011. Agreement as a fallible operation. PhD Thesis, MIT. - van Urk, C., and N. Richards. To appear. Successive cyclicity in Dinka. LI. 2