This article presents novel data from the Nilotic language Dinka, in which the syntax of successive-cyclic movement is remarkably transparent. We show that Dinka provides strong support for the view that long-distance extraction proceeds through the edge of every verb phrase and every clause on the path of movement (Chomsky 1986, 2000, 2001, 2008). In addition, long-distance dependencies in Dinka offer evidence that extraction from a CP requires agreement between v and the CP that is extracted from (Rackowski and Richards 2005, Den Dikken 2009b, 2012a,b). The claim that both of these components constrain long-distance movement is important, as much contemporary work on extraction incorporates only one of them. To accommodate this conclusion, we propose a modification of Rackowski and Richards 2005, in which both intermediate movement and Agree relations between phase heads are necessary steps in establishing a long-distance dependency.
The idea that long-distance dependencies involve successive-cyclic movement through the edges of CP and vP/VP goes back to Chomsky 1986 and is one of the core assumptions of phase theory (Chomsky 2000, 2001, 2008). In this article, we show that the Nilotic language Dinka provides remarkably strong and straightforward evidence for this claim. Dinka’s phase edge positions, Spec,CP and Spec,vP, have the EPP property, so that ordinarily they must be occupied by some XP. Because these positions are also edge positions, long-distance movement can satisfy these EPP requirements on the way, allowing us to trace the path of movement straightforwardly. This pattern then provides evidence for the claim that long-distance dependencies involve a sequence of movement steps of the extracted XP from edge to edge, as assumed in phase theory.1
Dinka also provides insight into the limitations of successive-cyclic movement, however. We will show that, in cases of long-distance extraction through Spec,vP, it is not actually the wh-phrase that satisfies the EPP property of v; rather, it is the finite CP from which the wh-phrase is extracted. In particular, we will argue that not only does the wh-phrase move through the edge of every phase—a CP from which extraction takes place must move to Spec,vP as well. However, because CPs are always linearized to the right in Dinka, this mimics the effects of long-distance movement. In this way, Dinka offers evidence that, in order to attract an XP from an embedded CP, v must stand in a syntactic relation with that CP. This conclusion is also reached in work on Tagalog, Hungarian, and Zulu (Rackowski and Richards 2005, Den Dikken 2009b, 2012a,b, Halpert 2012).
That both of these components constrain successive-cyclic movement is an important conclusion, as many contemporary proposals only incorporate one of them. Much work takes phase impenetrability to be the primary constraint on long-distance dependencies (e.g., Bošković 2002, 2007, Chomsky 2008, 2013), leaving a minimal role for Agree and featural triggers. Conversely, Rackowski and Richards (2005) and Den Dikken (2009b, 2012a,b) argue that agreement between v and CP allows v to ignore the CP phase as a locality boundary, thereby doing away with the need for intermediate movement to Spec,CP.
To reconcile these conclusions, we propose a modification of Rackowski and Richards 2005, in which the need for a syntactic relation between v and the CP from which extraction takes place is independent of phase impenetrability. This allows us to preserve the insight behind Rackowski and Richards’s proposal without jettisoning the traditional view of successive cyclicity, for which Dinka offers such striking evidence.
The article is structured as follows. In section 2, we introduce the two positions in the Dinka clause that must be occupied by an XP in declaratives, which we identify as Spec,CP and Spec,vP. In section 3, we demonstrate that the EPP property of these positions can be satisfied by whmovement, providing strong evidence for the claim that long-distance dependencies involve intermediate movement steps. In section 4, we then document a restriction on Spec,vP, which we attribute to v’s role as a Case assigner; and we present an argument for our approach to Dinka based on how this affects extraction of DPs and PPs. In section 5, we look more closely at long-distance extraction through Spec,vP and argue that it is the finite CP, not the wh-phrase, that satisfies v’s EPP property. We propose that this is an instantiation of the idea that agreement between v and an embedded CP is required for extraction to take place from that CP. In section 6, we present a theory of long-distance extraction, based on Rackowski and Richards 2005, which attempts to reconcile this finding with the notion of phase impenetrability. In section 7, we consider the Dinka facts in a crosslinguistic perspective. Finally, in section 8 we offer brief conclusions.
2 Two EPP Positions in Dinka
Dinka is a Nilotic language, spoken in South Sudan by at least 3 million people. There are many different varieties of Dinka, and they can be divided into five distinct dialect groups. This article presents data from Dinka Nyarweng, which is part of the southeastern Bor dialect group. No detailed syntactic work has been done on any Dinka Bor dialect, though we draw on work by Torben Andersen (1991, 2002, 2007, 2012) on the syntax of the Agar dialect and a short grammar (Nebel 1948) that describes some of the properties of the Rek dialect. All examples presented here are drawn from our own fieldwork.
Most Dinka words are monosyllabic, and most grammatical distinctions are made by means of tone, vowel length, or vowel quality (see, e.g., Andersen 1987, 1993, Malou 1988, Remijsen and Gilley 2008). In addition, Dinka is a verb-second (V2) language with a fairly rigid word order (Andersen 1991, 2002). Two aspects of Dinka word order will be particularly important here. There are two positions, which we identify as Spec,CP and Spec,vP, which must be occupied. We will start by outlining the properties of these two positions in declarative clauses.
2.1 Spec,CP as an EPP Position
Let us first introduce the properties of the position we will take to be Spec,CP in Dinka. We will show that, in finite clauses, this is an EPP position, in that it must be occupied by some XP. We then present several arguments that the position in question is Spec,CP, although it exhibits ‘‘mixed’’ properties, behaving both like an A-position and like an Ā-position.
To begin with, Dinka is a V2 language (Andersen 1991, 2002). In a declarative clause, the highest verb or auxiliary raises to C and must be preceded by one and only one XP.2 As in other V2 languages, different types of XP.2 can appear in Spec,CP. In (1a–c), Spec,CP hosts a subject, an object, and a locative, respectively. (We have boxed the XP that appears in the Spec,CP position. Where relevant, we will box both Spec,CP and Spec,vP, to make it easy to see how these positions interact with extraction.)
Dinka is V2 in finite embedded clauses as well, a property that will be important throughout this article. In an embedded declarative, for example, the highest verb or auxiliary is again in second position and must be preceded by an XP from the same range of phrases.3
As in other V2 languages, Spec,CP is generally not allowed to be empty. In a matrix clause like (3a), the only grammatical interpretation that is possible if the verb is initial is that of a yes/no question. In an embedded clause, as in (3b), ungrammaticality results (embedded yes/no questions require an overt complementizer, as in English).
There are three arguments for analyzing this position as Spec,CP. First, it is the landing site of wh-movement. When wh-words are merged with the focus particle ye-, they obligatorily front to first position, to form matrix and embedded questions.4
Second, as we will discuss further in section 3.1, the first position is also the landing site of successive-cyclic movement; if any phrase is extracted from an embedded clause, the initial position of that clause may not be filled.
A third reason for thinking that V2 is established in the C domain in Dinka comes from the fact that the V2 position hosts a clause-type particle, a-. In particular, 3rd person XPs that occur in first position trigger the appearance of this morpheme on the highest verb or auxiliary (boldfaced in (6b)).
This particle only appears in declaratives, however. When a wh-phrase moves to Spec,CP, a- is absent.
(7) First position agreement disappears under wh-fronting
Yeŋó cíi Ból tîŋ?
what PRF.NS Bol.GENSEE
‘What has Bol seen?’
c[ˆεn cé gâai [yeŋó cíi Ból tîŋ].
I PRF wonder what PRF.NS Bol.GEN see
‘I have wondered what Bol has seen.’
The particle also goes missing in questions in which there is no wh-word before the verb, such as in-situ questions or yes/no questions.
(8) First position agreement is absent in questions without wh-fronting5
ŋ:´r cé ŋó ku[`εn?
Ngor PRF what read
‘What has Ngor read?’
Cé Bôl kúrà càk?
PRF Bol.GEN bowl make
‘Has Bol made a bowl?’
We can explain these facts by saying that the first position is a specifier of a C head. We can then say that there are two kinds of C heads, a declarative one (marked by the appearance of awith 3rd person agreement) and an interrogative one. In contrast, if the first position is lower (say, Spec,TP), we have no reason to expect it to be morphologically sensitive to clause type.6
We have now shown three reasons to think of the first position in the Dinka clause as Spec,CP. It participates in V2 and in wh-movement, both terminal and successive-cyclic, and these are all phenomena standardly associated with Spec,CP. Moreover, the putative Spec,CP position is marked by morphology that is sensitive to clause type, again showing sensitivity to a property expected to be marked in the CP domain.
The phrase CP domain will be important, since the Spec,CP position under discussion here can be preceded by overt complementizers in embedded clauses (see also the examples in (2a–c)).
In fact, embedded V2 interacts in different ways with different complementizers. Under ke, for example, V2 is obligatory, while it is optional under ye. Other complementizers, such as the interrogative complementizers mèn ‘whether’ and ná ‘if ’, only tolerate verb-initial order.
We will therefore think of Spec,CP as part of an extended left periphery (Rizzi 1997), including at least two projections, one for Spec,CP and another for preceding overt complementizers in examples like (9a–b).7
One of the reasons why Dinka offers such a diversity of evidence for successive cyclicity is that long-distance dependencies go hand in hand with changes in case and agreement. As a result, the position we will call Spec,CP has some mixed properties, behaving both like an Aposition and like anĀ -position. One such property has already been mentioned: the DP in Spec,CP, in declarative clauses, controls φ-feature agreement on a following morpheme.
Perhaps relatedly, the phrase in Spec,CP occurs in the unmarked case form.
Note that Ayén ‘Ayen’ is in the genitive case in (12b–c), but in the unmarked case when fronted, as in (12a).8 Similarly, r:´k ‘town’ is in the locative case in (12a–b), but fronting converts it to the unmarked case in (12c). In general, the genitive case form is reserved for noninitial subjects and possessors, and the locative case appears on postverbal locatives (Andersen 1991, 2002); the cases are typically signaled, as in these examples, by alternations in tone and vowel length.
Not only is Spec,CP associated with case and agreement, but it also appears to behave like an A-position with respect to conditions on binding. A fronted object, for example, may bind into the subject (13a–b), in addition to being able to contain a bound variable (13c).
(13) Movement to Spec,CP has no Weak Crossover effects
Th:´_k-dé à-cé dhùk ébέn kâac.
goat-SG.his 3SG-PRF boy every bite
‘Hisj/*i goat bit every boyi.’
Dhùk ébέn à-cíi th:´_k-dé kâac.
boy every 3SG-PRF.NS goat-SG.his bite
‘Every boyi, hisi goat bit.’
Kit[`[p-kè àa-nhi[´[r dhu`_k é _b[´_n ké .
books-PL.3SG 3PL-love.NS boy.GEN every PL
‘Hisi books, every boyi loves.’
The properties of Spec,CP, then, are apparently quite mixed; it participates in various movement operations like an Ā -position, hosting V2 phenomena and wh-movement, but it behaves for purposes of case, agreement, and binding theory like an A-position.
It appears, then, that the properties of the CP and TP domains are not segregated in Dinka as they are in English (a fact that we might try to account for in terms of Chomsky’s (2001, 2008) idea that the features responsible for making Spec,TP an A-position are in fact inherited from C).9 See Van Urk, in preparation, for extensive discussion of the mixed properties of Spec,CP in Dinka, and a proposal for how these case alternations come about that is fully compatible with our claims here.
2.2 Spec,vP as an EPP Position
There is a position in the verbal domain that is similar to Spec,CP, in that it also has the EPP property, creating a V2-like effect in the vP. We identify this position as Spec,vP. The Spec,vP position immediately precedes the Dinka verb cluster (i.e., where verbs appear if they are not the highest verb or auxiliary in the clause) and must be occupied by a nominal object.
We can examine this position in clauses that contain an auxiliary (if no auxiliary is present, the main verb raises to C and there is no verb cluster). For example, in (14a–b), the perfective auxiliary cé occupies the V2 position. As a result, the main verb tîŋ‘see’ does not raise to C. Observe now that the position before the main verb must be occupied by the direct object.
It is never possible for the Spec,vP position to be empty.11
There is a key difference between Spec,vP and Spec,CP, however, which will be particularly important later in the article. Unlike Spec,CP, Spec,vP does not host adjuncts. In intransitives, for example, PP adjuncts may not occur before the verb.12
We have shown, then, that there is a position in the verbal domain that also has an EPP property. We will take this to be Spec,vP, because it appears to be an object position and because, as we will shortly show, it functions as an edge position, both properties generally ascribed to Spec,vP (e.g., Chomsky 1995, 2000, 2001, Wurmbrand 2001). It is important to note, at this point, that there is one complication with identifying this position as Spec,vP. In particular, when there are multiple auxiliaries in a clause, every auxiliary but the highest one appears in the verb cluster, following the object. In (18), for example, the future auxiliary bé is not the highest auxiliary and so appears in the verb cluster, after the object.13
Auxiliary may precede Spec,vP
Yìn à-cîi mìir bé tîŋ.
you 3SG-NEG giraffe FUT see
‘You will not see a giraffe.’
To account for this, we propose that Dinka auxiliaries are restructuring verbs in the sense of Wurmbrand 2001, which combine with a VP, so Spec,vP is in all instances introduced by the highest predicate. This explains why a variety of auxiliaries can appear after objects, and it fits well with the fact that Dinka has a wealth of auxiliaries, including ones that express various adverbial meanings, with no apparent limit on how many can appear in one clausal domain (see Andersen 2007 for extensive discussion).14
To sum up: we have shown that Spec,CP and Spec,vP function as EPP positions in Dinka, in that they must be occupied if possible. We now turn to their behavior in the context of longdistance extraction. As we will show, Dinka provides abundant evidence that these positions are used as intermediate landing sites in long-distance extraction.
3 The Signature of Successive Cyclicity
With this much understanding of the structure of the Dinka clause in place, we can turn to the evidence that Dinka offers for the nature of successive-cyclic movement. We will show that extraction exhibits two kinds of interactions with the positions we identified as Spec,CP and Spec,vP. First, movement across Spec,CP and Spec,vP typically requires that these positions not be overtly filled. Our hypothesis will be that extraction must take place successive-cyclically, occupying the edge positions on its way up the tree. Second, extraction of plural wh-phrases leaves behind a plural clitic ke in Spec,vP, again suggesting that wh-phrases must land in this position. Note that, although we will restrict discussion to wh-questions, the same facts obtain for other kinds of movement, like topicalization and relativization.
3.1 EPP Effects and Long-Distance Extraction
In section 2, we demonstrated that Spec,CP and Spec,vP must be filled in Dinka whenever possible, a fact that we described as an EPP effect. We will now show that long-distance extraction across these positions makes the EPP effects appear on the surface to be absent, requiring positions to be empty that are typically obligatorily occupied.15 We propose that the EPP requirements are satisfied, in this case, by the moved phrases themselves, which occupy Spec,CP and Spec,vP as they move past.
As noted earlier, in ditransitives one object must occupy Spec,vP.
Now observe that, when either object is extracted, it is not possible for the other, noninterrogative object to appear in Spec,vP.16
Extraction must then apparently take place via Spec,vP, leaving it empty. This is in clear contrast with declaratives, in which leaving Spec,vP empty is ungrammatical (see (16a–b)).
A similar effect is found with Spec,CP. If an XP is extracted from an embedded clause, the Spec,CP position of that clause may not be filled (21a–d), even though, as shown in section 2, it must be filled in declaratives (e.g., (3a–b)).
This effect extends to every Spec,CP position on the path of movement. Movement of an XP across multiple clauses causes every Spec,CP along the way to appear empty. For example, when extraction crosses two clause boundaries, both the Spec,CP of the intermediate clause and the Spec,CP of the most deeply embedded clause must be empty at the surface (22a). If either position is occupied, the sentence is ungrammatical (22b–d).
Thus, Spec,CP and Spec,vP positions on the path of movement appear empty at the surface.
The picture gets more complicated when we turn to long-distance extraction through Spec,vP, but we defer a complete discussion of this until section 5. In section 5, we show that, although the wh-phrase does move through the edge of every Spec,vP on the path of movement, the phrase responsible for satisfying v’s EPP requirement in long-distance extraction is actually the finite CP from which extraction takes place. However, finite CPs undergo obligatory extraposition, by being linearized to the right, thereby making Spec,vP appear empty. This will form the basis for our argument that long-distance extraction requires a syntactic relation between v and the CP from which the wh-phrase is extracted (Rackowski and Richards 2005, Den Dikken 2009b, 2012a,b, Halpert 2012).
Setting this complication aside for the moment, we propose that the pattern described in this section obtains because, in addition to being EPP positions, Spec,CP and Spec,vP are on the edge of a locality domain (e.g., a phase), so that extraction must proceed through them. We then maintain the generalization that, at the relevant stage of the derivation, Spec,CP and Spec,vP in Dinka are always occupied. Apparent exceptions only appear because the XP that fulfils this function may subsequently undergo movement.17
There is a second way in which the effects of successive cyclicity surface in Dinka, by means of a process we will refer to as ke-stranding. This process provides support for the picture sketched above and will function as a useful diagnostic throughout the article.
In Dinka, plural DPs strand a plural morpheme ke in each Spec,vP along the path of movement.18 The paradigm in (23a–b) illustrates. When a singular object is extracted, as in all the examples we have presented so far, the Spec,vP position appears empty (23a). When a plural object is the target of movement, however, it strands a plural morpheme ke in Spec,vP (23b).
This process is obligatory. Omitting the ke morpheme is ungrammatical.
Note that ke-stranding never happens in Spec,CP. It only ever affects Spec,vP positions. As in cases of local extraction, ke-stranding is obligatory in long-distance extraction. Omitting ke in any Spec,vP results in ungrammaticality.
The process of ke-stranding offers additional evidence for successive cyclicity, at least in the vP domain, as features associated with the wh-phrase end up in Spec,vP positions along the path of movement.19 It thus provides support for the idea that long-distance extraction of an XP touches down in every Spec,vP on the path of movement.
4 PP-Extraction and a Case Restriction on Spec,vP
Our discussion thus far has concentrated on the extraction of DPs. We now turn to PP-extraction, which differs from DP-extraction in illuminating ways. In particular, we will demonstrate that the facts of PP-extraction offer independent support for the view of Dinka we have just developed. In addition, this discussion will allow us to set the stage for section 5, in which we argue that long-distance extraction in Dinka also requires a syntactic relation between v and the CP from which extraction takes place.
We start with the observation, made in section 2, that low PP adjuncts, such as locatives and instrumentals, cannot appear in Spec,vP, not even in an intransitive.
The same is true for PP arguments, such as the locative argument of tu:`:c ‘send’.
The facts in (28a–b) appear to rule out an approach in which the PPs in question are unable to move to Spec,vP because they are generated outside vP. It is hard to see how conditions on selection would allow locative arguments to be generated outside vP. We propose, then, that the EPP property of v is associated with its role as a Case assigner, so that only phrases that are assigned Case by v may satisfy its EPP requirement.20 PPs lack a Case requirement and therefore cannot move to the Spec,vP for EPP reasons.
Now observe that this restriction on the EPP property of v also shows up under extraction. Extraction of argument or adjunct PPs fails to empty Spec,vP.
The Spec,vP positions in these examples are obligatorily filled by objects, just as if extraction had not occurred.
Does the failure to satisfy v’s EPP requirement indicate that these wh-phrases do not exit vP via its edge? That would be surprising, given current understanding of phase impenetrability. In addition, this suggestion would be inconsistent with data from Dinka’s other diagnostic for successive-cyclic movement, ke-stranding. Although they do not cause Spec,vP to appear empty, extracted locatives and instrumentals, when plural, do strand the plural morpheme ke on the edge of vP, just as plural DPs do. This is illustrated for wh-movement in (30a–b). The same point is made by the topicalization example in (30c), taken from a Dinka Christmas hymn.
Thus, we have evidence that these XPs do exit via the edge of vP; their movement simply fails to empty the vP edge. We formalize this observation by positing two movement-driving features on v: one specifically associated with Case and another that triggers wh-movement. As we will discuss more extensively later, we take movement to always be feature-driven (e.g., Chomsky 1995, McCloskey 2002).23 In examples like (30a), these two movement-driving features must be satisfied by different phrases; ye b[`_εi kô ‘to which villages’ is a wh-phrase but is inactive for Case, while wánmáth ‘brother’ is active for Case but is not a wh-phrase (in (31) we have omitted the subject, for ease of exposition).
Thus, in (31), movement of the locative wh-phrase ye b[`_εi koˆ ‘to which villages’ is driven only by the wh-feature of v, while movement of wá_nmá_th ‘brother’ is driven only by the Case feature of v. Further wh-movement of the PP will strand ke in the position the PP currently occupies, yielding the word order in (30a).24
At every step in a derivation, if a probe can trigger two operations A and B, and the features checked by A are a superset of those checked by B, the grammar prefers A.
Multitasking rules out (32b), since this example involves checking the Case and wh-features of v with separate phrases, when only one would do, as in (32a).26 This makes (32b) uneconomical, as it requires an extra movement step not necessary in (32a). The representation in (31), by contrast, does not run afoul of Multitasking, since the two movement operations to the edge of vP are triggered by different features, and neither phrase can satisfy both features.
We have shown, then, that the EPP property of v has the same nature in extraction and nonextraction contexts: a PP cannot satisfy the EPP property of v, regardless of whether or not the PP is a wh-phrase.
PP-extraction through Spec,CP provides support for this view. We have already shown that Spec,CP is less discriminating than the edge of vP. As is standard in V2 languages, Spec,CP in Dinka may seemingly be occupied by any phrase at all.
Since PPs can occupy Spec,CP, PP-extraction from CP then should empty the edge of CP, just as DP-extraction does. And indeed, extraction of a PP from an embedded clause blocks movement to Spec,CP of that clause.
Thus, the distribution of EPP satisfaction under wh-movement is the same as in declaratives: DPs, but not PPs, may satisfy the EPP property of v, and any phrase may satisfy the EPP property of C. These conditions hold both for wh-phrases and for non-wh-phrases, providing additional support for our proposal that the EPP property of Dinka Spec,CP and Spec,vP may be satisfied by successive-cyclic movement.
We are now ready to turn to the profile of long-distance extraction through Spec,vP. We will show that, in addition to providing strong evidence for successive-cyclic movement, Dinka offers support for the idea that extraction must be accompanied by an Agree relation between v and the CP from which extraction takes place, as proposed by Rackowski and Richards (2005), Den Dikken (2009b, 2012a,b), and Halpert (2012).
5 The Role of Complement Clauses in Extraction
In section 3, we showed that wh-movement satisfies the EPP property of intervening Spec,CP and Spec,vP positions. We also showed that this is true for monoclausal extraction through Spec,vP and for all Spec,CP positions that lie on the path of movement.
At first glance, this pattern appears to extend to long-distance extraction through Spec,vP. Extraction from a finite clause embedded by a ditransitive verb requires that the ditransitive verb not be preceded by an XP in Spec,vP.
Example (37a) demonstrates that Spec,vP can be filled by the nominal object of ditransitives like lέk ‘tell’, as we would expect. In (37b–c), this position is obligatorily not occupied in longdistance extraction.
Although (37a–c) seem to accord with what is known about Spec,CP, these facts are surprising from the perspective of the PP-extraction data just discussed. We proposed that only phrases that bear an active Case feature can satisfy the EPP property of v; this was our way of distinguishing between DP arguments, which can appear in Spec,vP, and PPs, which cannot. Examples like (37b) are problematic for this proposal, however. The wh-phrase yeŋà ‘who’ presumably lacks an active Case feature by the time it has been extracted from the embedded clause into the matrix clause, and it otherwise cannot undergo movement to that position. Why should it be able to satisfy the EPP requirement on the v of the higher clause?
We will argue that, in fact, it does not. Instead, we propose that in (37b), it is actually the complement CP itself that satisfies the EPP requirement on v. However, when this CP moves to Spec,vP, the CP is extraposed (by being linearized to the right), making it seem as if Spec,vP is empty. Importantly, this means that we will posit that CPs in Dinka check Case, or at least have the ability to do so.
Evidence for this proposal comes from independently observable properties of finite CPs in Dinka, which demonstrate that CPs can occupy all the same positions that nominal objects can and that, when they do, they must concomitantly undergo extraposition. In particular, finite CPs generally have the option of moving to Spec,CP and Spec,vP in Dinka, as long as they appear to the right. We can illustrate this with ditransitives. In a clause containing a ditransitive verb that takes two DP objects, one of the objects must occupy Spec,vP, and Spec,CP must, as usual, be filled.
However, when one of these objects is a finite CP, the edge positions in the matrix clause can be empty.
To account for this pattern, we propose that complement CPs in Dinka, like nominal objects, may check Case in Spec,vP and move to Spec,CP. In this way, we capture the observation that CPs pattern with DPs and not with PPs.27 However, unlike objects, Dinka complement CPs are subject to an independent constraint, which forces them to appear sentence-finally. To reflect this, we posit a Dinka-specific rule that linearizes CPs to the right (41).
Dinka extraposition rule
When a CP has undergone Merge with a head H, the CP is linearized to the right of H.
In other words, CPs move to rightward specifiers. As a result, when a complement CP moves to an EPP position, it appears to the right, and not to the left. This is why Spec,CP and Spec,vP can be empty when a complement clause appears in the clause. The CP can move to these positions, but it is linearized to the right.28
Support for this way of viewing the facts in (39a–c) and (40a–b) comes from an implicational relationship between Spec,CP and Spec,vP in such constructions. If Spec,CP is empty, then Spec,vP must also be empty (42).
This follows if the empty edge positions in clauses containing CP complements are the result of movement and rightward linearization of the CP. For the CP to reach the edge of the matrix CP, it must pass through the edge of the matrix vP, and examples like (42) are therefore impossible.
Matrix v in (43) might, in this case, actually have its specifier overtly occupied, but by a CP that is still linearized to the right. Under this proposal, an example like (43) really involves the structure in (44).
On this account, then, Spec,vP in (43) is not emptied by the moved wh-phrase, but occupied by the embedded CP, which linearizes to the right.
However, when long-distance extraction exits the CP, the CP must occupy Spec,vP.
We thus arrive at the descriptive generalization that a clause from which extraction takes place must move to Spec,vP if it can. On the assumption that movement operations are always parasitic on Agree relations, we could restate this generalization as a requirement that if v is to Agree with a wh-phrase in an embedded clause, it must also Agree with the embedded clause itself.
We will return to this generalization shortly, but first we offer two pieces of evidence in support of our way of viewing long-distance extraction through Spec,vP. Our first argument involves the behavior of nonfinite clauses in Dinka. As we illustrate in (47) with an object control verb, lɔ̤̂ɔ̤ŋ ‘encourage’, which takes a DP and a nonfinite clause as its objects, such clauses cannot move to Spec,vP or Spec,CP.
We propose, then, that nonfinite clauses, unlike finite CPs, cannot check Case in Dinka (we will return to the reason for this). With this in mind, observe that, in clear contrast to extraction from finite clauses, in extraction from nonfinite clauses an overt DP must appear before the verb cluster, in Spec,vP.29
Our second argument involves long-distance extraction of PPs. We have shown that local extraction of PPs does not empty Spec,vP.
However, long-distance extraction of PPs does appear to empty Spec,vP in the matrix clause, just when extraction takes place from a finite clause.
These facts demonstrate that whether the Spec,vP position of the matrix clause in a longdistance question appears empty or not is determined, not by properties of the moved wh-phrase, but by properties of the embedded clause. The Spec,vP position in question is empty just if the embedded clause is of the type that is capable of emptying it (i.e., if the clause is finite), regardless of the properties of the moved wh-phrase. In other words, in long-distance extraction through Spec,vP, it is irrelevant whether the wh-phrase is a DP or not, in strong contrast to the behavior of monoclausal wh-extraction through the same position. We take these facts as evidence that it is the embedded clause itself that empties Spec,vP.
Now that we have outlined this proposal, let us turn briefly to the obligatory aspect of it. Recall that when extraction takes place from a finite clause, the finite clause must move to Spec,vP (51a–b), even though this is optional when no extraction takes place.
We interpret this obligatoriness as evidence that, in the context of a long-distance dependency, there must be a syntactic relation between v and the finite CP that contains the phrase that is to be extracted.
This accords with work by Rackowski and Richards (2005) and Den Dikken (2009b, 2012a,b), who argue that extraction from a CP in Tagalog and Hungarian requires an Agree relation between v and CP. In Tagalog and Hungarian, this Agree relation is signaled via agreement morphology. In Dinka, the Agree relation has syntactic consequences, as we have shown.
Dinka thus provides evidence that the syntax of long-distance extraction involves at least two components: intermediate movement steps and an Agree relation between the embedding v and the CP to be extracted from. This is an important conclusion, as many models of successive cyclicity involve one but not the other. Rackowski and Richards (2005) and Den Dikken (2009b, 2012a,b) argue that agreement circumvents phase boundaries, obviating the need for intermediate movement either in some or in all cases. Similarly, a number of proposals take phase impenetrability to be the main constraint on long-distance dependencies (e.g., Bošković 2002, 2007, Chomsky 2008). The facts of Dinka Ā-movement demonstrate that an approach to long-distance movement is needed that links these two types of models. In the next section, we develop a theory of locality and extraction that captures the Dinka facts.
6 Phases as Interveners for Wh-Probing
We have shown that Dinka leads to the conclusion that the syntax of long-distance extraction involves two components: intermediate successive-cyclic movement and agreement between phase heads. In this section, we develop a modified version of Rackowski and Richards’s (2005) proposal, which incorporates both of these components.
The starting point for Rackowski and Richards’s (2005) proposal is the idea that if a CP (for example) contains a wh-phrase, then both the wh-phrase and the CP itself are possible goals for a higher probe that seeks wh-phrases. In fact, it will be important to develop a theory of locality in which the dominating CP is actually a closer potential goal for higher probes; this will be the key to forcing higher probes to Agree with the dominating CP before they can Agree with the wh-phrase.
We might formalize the idea that a CP dominating a wh-phrase is a potential goal for whprobes in a number of ways. For present purposes, we will assume, as we also did in footnote 23, that a phase head with a wh-phrase in its specifier must have a feature that is responsible for driving this wh-movement, even if the CP is declarative (Chomsky 1995, McCloskey 2002, Preminger 2011). On the proposal that all movement is feature-driven, these features must be present to allow the wh-phrase to escape the effects of phase impenetrability. On this view, a declarative CP from which wh-extraction takes place successive-cyclically has, at one point in the derivation, the partial structure in (52).
The CP node, because it is a projection of C, has the same featural content as the C head. Because C must carry a wh-probe (otherwise, the wh-phrase cannot escape its Spell-Out domain), this means that CP is necessarily a node with a wh-feature as well. As a result, CP and the whphrase are both potential goals for higher probes seeking wh-features. We can now ensure that CP will be the closer of the two goals for higher probes by defining the notion of closeness along the following lines:30
Definition of closeness
A goal A is a closer goal for an Agree relation with a probe P than a goal B if A is distinct from B and every node that dominates A also dominates B.
Assuming that Agree must target the closest goal, the condition in (53) guarantees that a goal A is more accessible to higher probes than a distinct goal B if A either c-commands or dominates B. By this definition, the CP in (52) is a more accessible goal for higher wh-probes than the whphrase itself, since the CP dominates the wh-phrase.
The conceptual basis behind (53) can be traced back to Chomsky’s (1964) A-over-A Principle. The main difference is that the mechanism of probing is featurally sensitive, so that we express the intuition behind the A-over-A Principle more precisely and restrict its application to cases in which the goals involved carry the same feature (thus avoiding problems of the sort noted by Ross (1967)). See also Takano 1994, Müller 1996, 1998, and Kitahara 1997 for arguments from constraints on remnant movement that this is a desirable move.
A note is in order about the objects that compete for ‘‘closeness’’ in the sense of (53). In principle, one might have expected the relevant objects to be the features that participate in Agree, on the theory that it is the ‘‘length’’ of Agree relations that is to be made as short as possible. In a tree like (52), however, this cannot be right; the computation must consider the ‘‘closeness’’ of the CP and whP nodes, if CP is to be a closer goal than whP. CP and whP should not be the positions in which wh-features are initially merged, on standard assumptions: these features should appear on heads dominated by these phrases.
We are driven to a similar conclusion for classic Superiority contexts like (54).
The wh-features dominated by whP1 and whP2 are not in a c-command relation. As in (52), then, closeness must apparently be sensitive to relations, not between features themselves, but between certain phrases that dominate the features. The question of how to determine which phrases are relevant for closeness is essentially the question of how to account for pied-piping, and we will not address it here. What will be relevant for us, however, is that locality conditions constrain relations, not between the features that participate in Agree, but between certain syntactic nodes that dominate those features.
In the tree in (52), then, CP is a closer goal than whP. We must now ask how the wh-phrase is to be extracted. Here we follow much work on multiple wh-constructions (e.g., Richards 1998, Hiraiwa 2001) in claiming that a probe may Agree with multiple distinct goals, as long as it does not ‘‘skip’’ the highest potential goal in favor of more deeply embedded goals. For example, a probe may Agree with two goals, one structurally higher than the other, as long as they are the highest goals in the search space; the fact that one of the goals is higher than the other does not disqualify the lower goal from being Agreed with. Rackowski and Richards (2005) present this idea via a version of Richards’s (1998) Principle of Minimal Compliance.
Principle of Minimal Compliance
Once a probe P Agrees with a goal G, P can ignore G for the rest of the derivation.
We revisit the precise formulation of (55) below.
The locality condition in (53), paired with the condition on multiple probing in (55), will guarantee that in a configuration like the one in (52), higher probes seeking to extract the whphrase must Agree both with the wh-phrase and with the CP dominating it. As we showed in the last section, the facts of Dinka support this conclusion; extraction from a tensed CP requires that v Agree with the CP, moving it to Spec,vP (after which the clause will extrapose, leaving the specifier position empty).
Let us illustrate this system with an explicit derivation along these lines. We focus on the example in (56).
The lower C is endowed with a wh-feature, which attracts the wh-phrase to Spec,CP (57).
The matrix v is also endowed with a wh-feature. It probes and tries to access the wh-phrase in the lower Spec,CP. However, because the embedded CP also carries a wh-feature, it acts as an intervener for the Agree relation between v and the wh-phrase by the logic in (53). As a result, an independent Agree relation between v and the CP is necessary for the derivation to converge.31 As mentioned previously, we posit that this Agree relation involves the same Case-checking operation that attracts DPs to Spec,vP. This is accompanied by movement of the CP to Spec,vP (58).
The CP moves to a rightward specifier—a movement we assume is driven by independent constraints on the linearization of CPs in Dinka, as previously discussed.
Because this Agree relation between v and CP allows v to ignore CP as a goal, v can now access the wh-phrase in Spec,CP. The wh-phrase can then be attracted to Spec,vP, from which it later undergoes movement to the matrix Spec,CP (59).32
As a result, no XP appears before the verb cluster in the matrix clause. Neither XP that satisfies v’s feature appears there: the CP because Dinka requires CPs to linearize to the right, and the wh-phrase because it must move to the matrix Spec,CP to be interpreted.
In the preceding discussion, it has been crucial that the condition in (55), which allows goals in an Agree relation to be ignored, applies not to features in a well-formed probe-goal relation but to heads dominating those features. The account just outlined involves the wh-feature on a v head being able to agree into a CP, because the same v head houses a Case feature that needs to Agree with CP on independent grounds. As Rackowski and Richards (2005) note, the account might also provide an explanation of Huang’s (1982) Condition on Extraction Domain (CED), which effectively states that the phrases that are transparent for extraction are just those that should be in an Agree relation with v (i.e., complements of the verb); extraction from phrases that do not Agree with v (subjects and adjuncts) is blocked.
An imaginable alternative version of the account would avoid reference to heads, restricting the condition in (55) so that it only applies to the particular features that participate in a Multiple Agree relation, rather than to all the probing features on a particular head. On this account, we would have to say that both the complement CP and the wh-phrase are goals of the wh-feature on v, rather than being goals of two separate probes on v. The alternative would lose the account of the CED just sketched; it also makes the derivation just described the only Dinka case in which multiple instances of movement are triggered by a single wh-feature (Dinka lacks, for example, multiple overt wh-movement). On the other hand, the idea of referring only to the probes that participate in Agree, rather than to the heads dominating those probes, has a certain conceptual appeal. We will have to leave the matter unresolved for now.
One final remaining question is why long-distance extraction from nonfinite clauses does not trigger movement of the clause to Spec,vP. We propose that such clauses lack a phasal C head, so that extraction may proceed from it without Agree. We tie the idea that nonfinite clauses cannot check Case to this property also. As extensively discussed by Hartman (2012), one of the ways in which a clause may come to act like a nominal is by the addition of D-like material to a full CP. We posit that, in Dinka, this material only attaches to phasal C heads, thus deriving the distribution of finite and nonfinite clauses.33
7 Beyond Dinka
In developing our account of the syntax of long-distance extraction in Dinka, we have made three claims that need to be evaluated in a larger linguistic context. First, we have argued that extraction from a CP requires an Agree relation between the embedding v and the CP. In section 7.1, we show that Dinka is part of a growing set of languages in which such effects have been documented (Rackowski and Richards 2005, Den Dikken 2009b, 2012a,b, Halpert 2012). Second, our derivation of long-distance dependencies in Dinka involves movement from an extraposed clause. In section 7.2, we provide evidence that this is possible and that extraposed CPs are not barriers to extraction. Third, our theory predicts that, alongside an Agree relation between v and CP, we should see effects of an Agree relation between vP and the next phase head above it. We discuss this prediction in section 7.3.
7.1 The Role of Agreement across Languages
There are other languages besides Dinka for which it has been argued that agreement plays a role in facilitating extraction (Rackowski and Richards 2005, Den Dikken 2009b, 2012a,b, Halpert 2012). In this section, we review some f these patterns and demonstrate that Dinka is part of a typologically diverse set of languages.
In Tagalog, extraction affects agreement morphology on the verb (Rackowski and Richards 2005). The Tagalog verb can enter into agreement with various arguments in the clause (Rackowski and Richards analyze this as case agreement, but nothing hinges on this for our purposes). In (60a), for example, the verb agrees with the subject; in (60b), it agrees with the clause.
(60) Verb agrees for case in Tagalog
M-agsa-sabi ang kalabaw [na masarap ang bulaklak].
NOM-ASP-say ANG water.buffalo that delicious ANG flower
‘The water buffalo will say that the flower is delicious.’
Sa-sabih-in ng kalabaw [na masarap ang bulaklak].
ASP-say-[cf9]ACC[cf1] CS water.buffalo that delicious ANG flower
‘The water buffalo will say that the flower is delicious.’
Now, the important observation is that the verb must agree with an embedded clause if an XP is extracted from within that clause.
(61) Extraction requires agreement with embedded clause in Tagalog
Kailan sa-sabih-in ng sundalo [na u-uwi ang pangulo]?
when ASP-say-ACCCS soldier that NOM.ASP-go.home ANG president
‘When will the soldier say [that the president will go home ]?’
*Kailan m-agsa-sabi ang sundalo [na u-uwi ang pangulo]?
when NOM-ASP-say ANG soldier that NOM.ASP-go.home ANG president
‘When will the soldier say [that the president will go home ]?’
A similar pattern exists in Chamorro (Chung 1998, Den Dikken 2009a). In this way, Tagalog and Chamorro, just like Dinka, require a syntactic relation between the embedding v and the CP from which extraction takes place. In Tagalog and Chamorro, the reflex of this relation is agreement morphology on the verb. In Dinka, this relation manifests itself as movement to Spec,vP.
Welsh also shows agreement under extraction that appears to instantiate this logic. As described by Willis (2008), Welsh has an object agreement clitic that optionally appears when an aspectual particle is present.34
(62) Object agreement optionally appears with aspectual auxiliaries in Welsh
Mae Megan wedi ei ddwyn e.
be.PRES.3SG Megan PERF3M.SG steal.INF it.M.SG
‘Megan has stolen it.’
Mae Megan wedi dwyn e.
be.PRES.3SG Megan PERF steal.INF it.M.SG
‘Megan has stolen it.’
Object agreement may appear on intermediate verbs in Welsh
Beth wyt ti ’n ei feddwl [bod hyn yn ei olygu]?
what be.PRES.2SG you PROG3M.SG think.INF be.INF this PROG3M.SG mean.INF
‘What do you think this means?’
As Willis observes, however, this intermediate agreement does not reflect the φ-features of the wh-phrase. Rather, it always surfaces as 3rd person masculine singular (64a). In addition, it may appear even with extraction of adjuncts (64b), which otherwise never trigger object agreement.
(64) Intermediate object agreement does not reference wh-phrase in Welsh
Pa lyfrau wyt ti ’n ei feddwl [oedd Megan yn
which books be.PRES.2SG you PROG3M.SG think.INF be.IMPF.3SG Megan PROG
‘Which books do you think Megan was reading?’
Ble wyt ti ’n ei feddwl [mae e ’n mynd]?
where be.PRES.2SG you PROG 3M.SG think.INF be.PRES.3SG he PROG go.INF
‘Where do you think he’s going?’
We propose that this object agreement clitic reflects an Agree relation between the verb and the clause that it is extracted from. This explains why it is invariantly 3rd person singular, as clauses do not vary in φ-features.
Den Dikken (2009b, 2012a) argues for the existence of another pattern along these lines for Hungarian. Finally, in the realm of A-movement, Halpert (2012) argues that raising out of a finite clause in Zulu is made possible by agreement between the raising verb and the complement CP (see also Thoms 2013 for a proposal that agreement plays a crucial role in patterns of Areconstruction).36
In all of these cases, then, extraction is accompanied by an Agree relation between the verb and the embedded CP, though this may manifest itself syntactically in a variety of ways. As a result, evidence for the role of Agree in facilitating long-distance extraction manifests itself in a typologically diverse set of languages, including Chamorro, Dinka, Hungarian, Tagalog, Welsh, and Zulu.
We have argued, however, that this aspect of the syntax of long-distance extraction must exist side by side with the notion of phase impenetrability and successive-cyclic movement, contra Rackowski and Richards 2005 and Den Dikken 2009b, 2012a,b, since Dinka offers such striking evidence for the successive-cyclic nature of long-distance movement.
7.2 On Extraction from an Extraposed CP
In the derivation we proposed for Dinka long-distance extraction in sections 5 and 6, extraction takes place from a CP that undergoes movement. At first glance, this may seem at odds with cases in which movement of CP seems to create freezing effects. This is evident, for example, in English pairs like (65a–b).
(65) Movement of CP blocks extraction in English
What is it unlikely [that John said t]?
*What is [that John said t] unlikely?
The same effect is found with CP objects. For example, Vicente (2005) shows that movement of CP renders extraction impossible in Basque. CPs can surface both to the right and to the left of the verb.
(66) CPs occur on both sides of verb in Basque
Jonek ez du uste [Mirenek liburua irakurri duenik].
Jon no AUX think Miren book read AUX.C
‘Jon doesn’t think that Miren read a book.’
Jonek ez du [Mirenek liburua irakurri duenik] uste.
Jon no AUX Miren book read AUX.C think
‘Jon doesn’t think that Miren read a book.’
However, when a CP surfaces to the left of the verb (in what Vicente argues is an A-position), extraction of a wh-phrase from within that CP is blocked.
(67) Extraction from moved CP is impossible in Basque
Zer ez du Jonek uste [Mirenek t irakurri duenik]?
what no AUX Jon think Miren read AUX.C
‘What doesn’t Jon think Miren has read?’
*Zer ez du Jonek [Mirenek t irakurri duenik] uste?
what no AUX Jon Miren read AUX.C think
‘What doesn’t Jon think Miren has read?’
(68) CPs may undergo scrambling in Dutch
Ik had [dat hij dat zou zeggen] niet verwacht.
I had that he that would say.INF not expected
‘I had not expected that he would say that.’
Ik had niet verwacht [dat hij dat zou zeggen].
I had not expected that he that would say.INF
‘I had not expected that he would say that.’
Extraction from within the scrambled CP is completely blocked.
(69) Extraction from scrambled CP is impossible in Dutch
Wat had je niet verwacht [dat hij t zou zeggen]?
what had you not expected that he would say.INF
‘What had you not expected that he would say?’
*Wat had je [dat hij t zou zeggen] niet verwacht?
what had you that he would say.INF not expected
‘What had you not expected that he would say?’
Such facts seem to suggest that movement of a CP is not compatible with extraction, which may be problematic for the current account.
However, extraposed clauses do not appear to pattern like other moved CPs. This is already evident in the Dutch examples given above. DP objects ordinarily appear preverbally in Dutch. We could then take the peripheral position of CPs to be the result of extraposition. As (69a) shows, extraction from a final CP is grammatical.
Similarly, in English, extraposition of CP does not block wh-movement.
Extraposition of CP does not block extraction in English
What did you say yesterday [that she wants to do t]?
(71) Relative clause extraposition feeds extraction in Norwegian
*Her er en bok som [ingen [CP som leser]] blir lykkelig.
here is a book that nobody that reads becomes happy
‘Here is a book that nobody who reads it becomes happy.’
Her er en bok som [ingen] blir lykkelig [CP som leser].
here is a book that nobody becomes happy that reads
‘Here is a book that nobody becomes happy who reads it.’
Per slipper jeg ikke inn [noen [CP som liker]].
Per let I not in anybody that likes
‘Peter, I didn’t let in anybody who likes him.’
Per slipper jeg ikke [noen] inn [CP som liker].
Per let I not anybody in that likes
‘Peter, I didn’t let anybody in who likes him.’
*Per slipper jeg ikke [noen [CP som liker]] inn.
Per let I not anybody that likes in
‘Peter, I didn’t let anybody who likes him in.’
Although we will not offer a theory of why extraposition differs from other types of movement in terms of freezing effects, it should be clear that the Dinka facts accord with those in other languages. Extraposition of CP is compatible with wh-movement of a phrase within it.
7.3 On Movement from Spec,vP
Another aspect of our proposal that deserves further discussion is the implication that reflexes of wh-agreement may also be expected between phase heads with regard to movement from Spec,vP. In particular, wh-movement from Spec,vP should also require an Agree relation between vP and the next phase head up.
In Dinka, this is not a problem, because, in the syntax we propose, the highest verb or auxiliary moves to C in every clause. If head movement is mediated by Agree, then, by transitivity, there is plausibly already an Agree relation between vP and every head up to C, as long as v is part of the complex head that moves to C. But we expect to find languages in which long-distance extraction requires a syntactic relation between heads in the same extended verbal projection. This could be C or T, and a vP from which extraction takes place.
Inversion in the context of wh-movement—in the Romance languages, for example (Torrego 1984, Rizzi 1991)—may be an effect of this type. There appear to be a number of languages in which the verb undergoes additional movement in the context of extraction. In the Romance languages, for example, many researchers have observed that a fronted wh-argument must be followed by the verb, so that the subject can no longer appear before the verb. This is true in Spanish (72a–b) and Italian (72c–d), for example.
(72) Verb raises to C in extraction context in Spanish and Italian
*Qué María lee siempre?
what Maria reads always
‘What does Maria always read?’
Qué lee María siempre?
what reads Maria always
‘What does Maria always read?’
*Che cosa Maria ha detto?
what Maria has said
‘What has Maria said?’
Che cosa ha detto Maria?
what has said Maria
‘What has Maria said?’
As Torrego (1984) shows for Spanish, the same effects are seen in embedded clauses when they are on the path of movement. In these clauses also, the subject must follow the verb.
(73) Verb also moves in embedded clauses in Spanish and Italian
Juan pensaba [que Pedro le había dicho [que la revista había publicado
Juan thought that Pedro him had told that the journal had published
ya el artículo]].
already the article
‘Juan thought that Pedro had told him that the journal had already published the
Qué pensaba Juan [que le había dicho Pedro [que había publicado
what thought Juan that him had told Pedro that had published
la revista t]]?
‘What did Juan think that Pedro had told him that the journal had published?’
On the assumption that Agree is a prerequisite for head movement, such patterns may reflect an Agree relation between vP and a higher phase head, such as T or C—that is, the parallel of the Agree relation between v and CP that we posited for Dinka.39
Alternatively, we may imagine that all heads in an extended projection share features, like a categorial feature or inflectional features. If so, we would not expect to see similar effects between C and vP, as they would already stand in the requisite Agree relation. We leave this as a question for future research.
In this article, we have shown that the Nilotic language Dinka offers particularly compelling evidence for the idea that long-distance dependencies involve intermediate movement steps through the edge of every verb phrase and every clause (Chomsky 1986, 2000, 2001, 2008). In addition, Dinka provides support for the idea that long-distance dependencies require agreement between all phase heads on the path of movement (Rackowski and Richards 2005, Den Dikken 2009b, 2012a,b, Halpert 2012). The syntax of long-distance extraction then appears to involve both of these effects: phase impenetrability and successive Agree relations between phase heads. On the basis of this conclusion, we have proposed a modification of Rackowski and Richards’s (2005) proposal, in which both of these components constrain long-distance dependencies and the role of agreement is not to void phasehood, but to allow a probe to access the phase edge.
We thank Noam Chomsky, Marcel den Dikken, David Pesetsky, and Omer Preminger for comments and discussion. We are indebted to Abiar Mako̤o̤r Gu:t and Mangok Bol for sharing their language with us, as well as to the Sudanese Education Fund and the Dinka community at Grace Episcopal Church in Everett, MA. Our thanks finally to audiences at talks at GLOW 36, ACAL 44, MIT, NELS 43, and Utrecht University, and to everyone in the Fall 2012 24.942 class at MIT.
This article uses a modified version of Dinka orthography. In particular, we mark tone (high = á, low = à, falling = â) and voice (creaky/modal = a, breathy = a̤). For typographical reasons, we omit the diacritic for breathy voice on the second vowel of long or extralong i. In addition, we use the following abbreviations: 1/2/3 = 1st/2nd/3rd person, ACC = accusative, ANG = ang morpheme, ASP = aspectual marker, AUX = auxiliary, C = complementizer, COP = copula, CS=case marker, DTR=ditransitive, FOC=focus particle, FUT=future, GEN =genitive case, HAB=habitual, IMPF = imperfective, INF = infinitive, LOC = locative case, M = masculine, NEG = negation, NOM = nominative, NS = nonsubject voice, OBL = oblique voice, PERF = perfective, PL = plural, PRES = present tense, PRF = perfect, PROG = progressive, Q = question particle, SG = singular, TR = transitive.
2 There are systematic exceptions to this with adjuncts, which may precede Spec,CP, creating a V3 structure. In this, Dinka contrasts with other V2 languages, like Dutch and German, in which this is not possible. We will not try to account for this variation here.
3 These examples illustrate another interesting property of Spec,CP. As is evident in (2a–c), a verb-initial order is exceptionally possible in the matrix clause if a complement clause is present. This is discussed at length in section 5, where we will propose that this happens because complement CPs may move to Spec,CP and extrapose.
4 In the examples in (4), we have decomposed the wh-phrase into its two component morphemes, the wh-phrase and the particle ye-. We will follow Dinka orthography otherwise in how we transcribe ye-.
The particle ye- appears to derive from the inverted form of the copula, from a cleft construction. Note, however, that wh-questions with the ye- particle are not synchronically clefts. Clefts have a different syntax. They make use of an overt pronoun that occupies Spec,CP of the embedded clause and agrees with the declarative particle, as (ia–b) illustrate.
In addition to this, questions with ye- do not have the semantics of clefts. In particular, such questions do not induce a uniqueness presupposition, so that ‘‘mention-some’’ questions are felicitous. For example, the question-answer pair in (iia–b) does not imply that only sweet potatoes are tasty with meat.
5 In-situ questions productively alternate with wh-questions and are fine even in out-of-the-blue contexts (see Cable 2012 for similar facts in Dholuo, a language in the same family). The fronting wh-words are formed by adding the particle ye- to them.
6 The picture is actually not quite this simple. Topic agreement also disappears in CPs headed by the complementizer ke. It is still clear, however, that the distribution of this agreement is conditioned by properties of the C domain.
7 Another option is to treat those complementizers that cooccur with V2 as adjoined particles of some sort. The difference between these complementizers and the ones that block V2, such as the interrogative ones, could be captured by saying that only those complementizers that block V2 are true C heads.
8 We call this the unmarked case because it is the citation form and because it is the most general in its distribution. On morphological grounds, it is hard to determine which case form is the least marked, as they are mostly distinguished by differences in tone and vowel quality.
9 As an anonymous reviewer points out, there is an alternative analysis, according to which V2 is established at Spec,TP in Dinka and all the C-properties it seems to have, including phasehood, are inherited from the left periphery (e.g., by means of selection or feature inheritance). We think this analysis is significantly less straightforward, but it is worth noting that adopting it would not affect the overall conclusions of the article, though it would have consequences for phase theory. See Diesing 1990 for a related proposal about the V2 position in Yiddish.
10 We will ascribe this freedom to a v probe that is in an Agree relation (hereafter, Agrees) with both objects, rendering it capable of triggering movement of either one to its specifier. It will thus be important for our proposal that locality conditions that require probes to Agree with the closest possible goal do not similarly constrain the choice of possible Move operations. A class of imaginable alternative proposals might posit derivations with different starting points for the two options in (19a–b); for example, we might suppose that the two objects may be base-generated in either order and that it is always the higher of the two objects that moves. We will not pursue this kind of proposal, in part because we cannot find support for it; we have been unable to discover in Dinka, for example, any semantic distinctions like those discussed by Oehrle (1976) between double object and prepositional-dative constructions. Such a proposal would also be inconsistent with our theory, developed below, of how wh-movement interacts with movement to the edge of vP. See footnote 26 for further discussion.
11 It is also not possible for both objects to occur before the verb cluster.
(i) Only one object can occur before verb cluster
*ɣ[ε̂n cé̤ kìtáp Ayé n yi[ε̤̂n.
I PRF book Ayen give
‘I gave Ayen a book.’
*ɣ[εn cé̤ Ayén kìtáp yi[ε̤̂n.
I PRF Ayen book give
‘I gave a book to Ayen.’
12 This means that there are instances in which Spec,vP can remain unoccupied, unlike Spec,CP. We will explain this by saying that the EPP property of v is tied up with its function as a Case assigner (to be more precise, we treat the EPP as a subfeature of the Case feature on v), as we discuss in section 4. As a result, assuming that intransitive v heads do not assign Case, the intransitive v, like the one in (17a–b), will end up not having the EPP property.
13 It may be relevant for this particular problem that bé̤ may actually not be best analyzed as marking future tense; it also functions as a marker of nonfinite clauses, for example.
14 There is an imaginable alternative view according to which Spec,vP is actually a specifier of a higher projection, like an aspectual or modal projection. A problem for this view is that it is unclear why the position before the verb cluster does not seem to be sensitive to the presence or absence of the subject. If we are dealing with the specifier of a projection above vP, the subject could in principle move there and should intervene for attraction of the object. In contrast, under our restructuring proposal, the fact that the position before the verb cluster cannot be occupied by the subject follows from the same factor that prevents v from assigning accusative Case to the subject in languages like English: when v is probing for a DP to satisfy its EPP feature, it c-commands only the object, not the subject.
15 Part of this observation goes back to Andersen (1991), who noted that local relativization requires Spec,CP to be empty.
16 See Cognola 2014 for discussion of somewhat similar facts in the Tyrolean dialect Mòcheno, a language that also allows both VO and OV word orders. The Mòcheno facts differ from the Dinka ones in a number of potentially illuminating ways; for one thing, Mòcheno word order appears to be sensitive to information structure in a way that Dinka word order is not.
17 One interesting aspect of these facts is that we could potentially derive from them a parsimony argument for models of long-distance dependencies that use movement, as EPP satisfaction is uniquely a property of movement. As a result, a movement approach can straightforwardly encode the generalization that underlies this pattern: Spec,CP and Spec,vP must contain an XP. In contrast, in theories that model long-distance dependencies without movement (e.g., Gazdar et al. 1984, Steedman 1987, Kaplan and Zaenen 1989, Sag and Fodor 1994, Bouma, Malouf, and Sag 2001, Goldberg 2006), one would have to resort to a more abstract analysis (e.g., a chain must terminate or cross C). It seems to us that a movement approach provides a simpler explanation in this case.
18Ke-stranding is thus reminiscent of the floating quantifier facts in West Ulster English discussed by McCloskey (2000).
19 Although we call this phenomenon ke-stranding, it is not like familiar instances of stranding, in that it is obligatory and can happen multiple times. In addition, ke is never pied-piped along with the wh-phrase or found on an in-situ DP. The latter fact we attribute to the facts that the spell-out of plural is highly idiosyncratic within the Dinka DP (as Andersen (2002:4) notes, ‘‘the number of a given noun cannot be determined from its phonological form’’) and that ke is not typically used to mark plurality on nouns.
Instead, we take ke to be an elsewhere plural morpheme that surfaces when a Number head is not spelled out inside a nominal. This proposal is supported by the fact that ke is otherwise used to mark plural in a wide variety of constructions (e.g., plural subject clitics, topic agreement in the past tense, plural possessum agreement). We posit that Dinka has a rule that cliticizes a plural Number head at the vP edge onto the vP, leading to the ke-stranding effect. Although this is an unusual proposal, alternative analyses of ke-stranding are implausible. Ke-stranding is not a type of wh-agreement, as the ke morpheme can stand on its own (see (30a–b)). Similarly, ke-stranding is not a type of resumption, as the distribution of resumptive pronouns is very different (resumptive pronouns are not limited to plural and are not allowed in Spec,vP).
21 Mass nouns like piú ‘water’ are formally plural in Dinka.
22Yécṳ̀ ‘Jesus’ is still in an object position, because there is no promotion to subject in a Dinka passive. An object of a passive cannot become genitive and can only appear where the object of an active sentence may appear. If we topicalize the object of a passive, for example, it must empty Spec,vP (as can be demonstrated with passives of ditransitives).
23 This means that we are adopting a featural view of intermediate movement. To be precise, we adopt the model of wh-movement suggested by Preminger (2011). A crucial component of this approach is the idea that a failure of Agree does not induce ungrammaticality. As a result, an unvalued wh-feature can be present on all phase heads. If a wh-phrase is present, this results in intermediate movement to the phase edge. If a wh-phrase is absent, then probing fails, but, since failures of Agree do not induce a crash, this is harmless. Similarly, we could assume that v always carries a Case feature, but that probing fails when a suitable object is not present, such as in an intransitive.
24 The relative positions of the two specifiers appear to be fixed. The ke morpheme cannot appear after wá̤nmá̤th ‘brother’ in (30a–b). Perhaps this ordering restriction reflects conditions on the placement of the ke clitic. We could imagine, for example, that ke must be initial within the vP phase.
25 The reference set for Multitasking is the set of possible operations one head can trigger, so that it is evaluated locally, without the need to compare derivations. Multitasking could also be stated as a transderivational constraint, however. This would make equivalent predictions for the data we discuss here, as far as we can tell.
26 It is important for our account, then, that (32a–b) are fed by the same underlying structure. We then posit that Dinka v assigns Case to both DPs in a ditransitive, by means of a Multiple Agree relation, and may, in principle, attract either one to satisfy the EPP property associated with the Case feature. This move ensures that (32a) and (32b) count as being in competition with each other for Multitasking. It also explains why both objects show up in the unmarked case.
27 This observation means that there needs to be a feature on independent grounds that picks out Dinka DPs and CPs, but not PPs. We take this feature to be Case here, but, if its identity were different, we could substitute it for Case throughout without affecting the overall proposal.
28 Note that there are several other ways of modeling CP-extraposition that would do equally well for our purposes. For example, we could propose that leftward movement to Spec,vP is followed by an independent step of rightward movement. Another option, which avoids movement of the CP altogether, is to posit a covert clausal expletive, which moves on the CP’s behalf. We will stick with the rule in (41) for simplicity here, though nothing really hinges on this choice.
29 Note that the ke-stranding example in (48a) verifies that the Spec,vP position of the object control verb lɔ̤̂ɔ̤ŋ ‘encourage’ is nonetheless used as an intermediate landing site.
30Rackowski and Richards (2005) offer a more complex definition of locality, which is intended not only to force Agree between phase heads but also to derive phase impenetrability. Their definition has the consequence that successivecyclic movement must move through the edges of intervening vPs, but may freely skip intervening CPs. The Dinka facts seem to demonstrate that this is untenable, at least for Dinka; successive-cyclic movement does indeed stop at the edge of every phase. We will therefore make use of the locality condition in (53), assuming that phase impenetrability is a separately enforced condition.
31 It is crucial here that the CP is a defective intervener, so that, although it acts as an intervener for wh-probing, it would not be a suitable goal. This could be ensured by stipulating that wh-probing is category-discriminating, in that a wh-probe will only initiate a successful Agree relation with a goal if the goal is of a specific syntactic category, such as DP or PP. This is analogous to Preminger’s (2011) proposal that φ-probing is case-discriminating (causing defective intervention with obliques in some languages, for example). For Dinka, we would then say that wh-probing can only successfully result in an Agree relation if the goal is a DP or a PP. We could allow for variation in this restriction, just as Preminger does for the property of case discrimination, since there are languages in which clausal pied-piping can be used in long-distance dependencies, such as Basque (e.g., Arregi 2003).
32 Note that this derivation requires that v be able to probe into its specifier, to access the wh-phrase. There are two alternatives that would also do the job. One would be to say that v can access material in the lower copy of the CP. Another would be to treat our constraints on intervention as representational (e.g., by saying that Agree relations must end up creating a contiguous span of probes and goals, so that a potential intervener just has to be agreed with at some point). This change of perspective would allow the wh-phrase to exit before the CP moves as long as movement of CP subsequently takes place.
33Hartman (2012) specifically proposes that CPs that move to argument positions are contained in a DP shell. We could adopt this proposal here and say that, in Dinka, this D head comes with a restriction that it only combines with finite CPs. This would be compatible with our account, regardless of whether we take this DP shell to be phasal or not.
34 As is evident in (62a–b), the clitic also triggers a soft mutation on the following infinitival verb.
36 An anonymous reviewer points out that long-distance agreement in Hindi might also be a relevant case (Bhatt 2005, Keine 2013). In Hindi, long-distance agreement into a nonfinite clause is obligatory when A-scrambling takes place out of that clause. This may be another instance in which agreement enables extraction. However, Keine (2013) argues that, when scrambling takes place, the result is actually a restructuring context, so that there are no phasal boundaries that would block agreement. On this account, Agree would not play a special role in facilitating extraction in Hindi.
37 Our thanks to an anonymous NELS conference reviewer for pointing out the relevance of these facts, which formed the impetus for this part of the article.
38 Norwegian appears not to obey wh-islands, the presence of which presumably blocks such derivations in English.
39 Another type of inversion that could possibly be analyzed this way is T-to-C movement, as in English, if T were taken to be a phase head also ( perhaps assuming, following Pesetsky and Torrego (2001), that the complementizer that is the spell-out of T-to-C movement in embedded clauses).