1 Introduction

Igbo (Benue-Kwa, Nigeria, Blench 1989) exhibits a subject (SU)/non-subject (non-SU) extraction asymmetry in constituent questions with a wh-constituent in the clause-initial position, illustrated in (1b–d) for a transitive predicate and in (1e) for a ditransitive predicate. (1a) provides the baseline declarative sentence. In (1c,d,e) we can see that the clause-initial question word representing a non-SU, viz. an (in)direct object (IO, DO) or an adjunct (ADJ), must be followed by the morpheme kà, glossed as a focus marker (to be motivated below). However, if the subject is questioned, the morpheme kà must be absent.¹

Hence, Igbo displays a SU vs. non-SU asymmetry. The same asymmetry is found in focus constructions (expressing new information or contrastive focus) with the focus XP in clause-initial position (see Amaechi & Georgi to appear on focus marking strategies in Igbo). Since the conditions on the presence or absence of the morpheme kà in sentences with a focused XP are identical to those in constituent questions, we will only consider interrogative clauses in this paper. The basic facts, i.e. the (apparently) grammatical function-driven distribution of kà in constituent questions, have been outlined in the mostly descriptive literature on Igbo before, see Goldsmith (1981); Ikekeonwu (1987); Uwalaka (1991); Ogbulogo (1995); Mmaduagwu (2012); Nwankwegu (2015). However, there is no study on the source of the asymmetry. The present paper aims to fill this gap and thereby to contribute to our understanding of extraction asymmetries in general.

On the surface, the asymmetry in Igbo manifests itself as a morphological phenomenon, viz. the presence vs. absence of the morpheme kà. In this paper, we argue that this morphological asymmetry in fact reflects a deeper syntactic asymmetry between SU and non-SU questions: While wh-non-SUs can undergo Ā-movement to a position in the left periphery (to be identified as SpecFOC), wh-SUs cannot move to this position and must stay in their canonical position SpecT. The morphological asymmetry arises because kà realizes the head (FOC⁰) of the left peripheral projection only if an overt XP occupies its specifier. Since wh-SUs do not reach this position, kà is absent in SU-questions. The result is similar to the that-trace effect in that a head of the (extended) C-system must be absent in SU-Ā-dependencies. Indeed, Igbo also exhibits the classic that-trace effect under long-distance SU-extraction. Furthermore, we provide evidence that the possibility for extraction of wh-SUs depends on the structural position of SUs and not on their features like case, θ-role etc. We develop an optimality-theoretic analysis that is based on the concept of Spec-to-Spec anti-locality (Erlewine 2016) and the variable size of clauses (the projection of the left-peripheral heads FOC and FORCE is subject to optimization, cf. Grimshaw 1997). In a nutshell, wh-SUs in SpecT do not undergo movement to the specifier of the immediately dominating projection because this movement step is too short and the constraint that militates against too short movement is high-ranked. In matrix SU-questions, this leads to the absence of the FOC-projection (which usually hosts wh-XPs) and consequently also to the absence of the focus marker. In long SU-questions, the anti-locality constraint leads to the absence of the FORCE-projection (that would usually provide an intermediate landing site for long wh-movement) and, as a result, to the absence of the embedding complementizer (that-trace effect). Wh-non-SU movement does not have any of these effects since movement of wh-non-SUs is not too short as they start from a lower structural position than SUs; FOCP and FORCEP can thus be projected as they do not interfere with anti-locality in non-SU questions. Apart from novel insights into the extraction asymmetry in Igbo, our study will also shed light on the nature of apparent wh-in-situ in the language, which behaves like overt syntactic movement in simple questions.

The paper is structured as follows. Section 2 provides some background on the language and the construction under discussion. In Section 3 we conduct various tests to explore the syntactic structure of non-SU questions and the nature of the morpheme kà. In Section 4 we investigate SU-questions and argue that wh–SUs do not undergo local Ā-movement; furthermore, we provide evidence for a structure-dependent approach to the SU/non-SU asymmetry and introduce the that-trace effect in Igbo. Section 5 explores the nature of wh-in-situ. These insights are incorporated into an anti-locality-based analysis in Section 6. Section 7 concludes.

2 Background

Igbo is spoken in Southern Nigeria by about 30 million people (grammars: Green & Igwe 1963; Carrel 1970; Manfredi 1991; Mbah 2006; Emenanjo 2015). Igbo is a tone language with three distinctive tones – low (à), high (á) and a downstep (ā) that indicate lexical and grammatical distinctions (Nwachukwu 1995). The language distinguishes between [+ATR] vowels (i, u, o, e) and [–ATR] vowels (ị, u, ọ, a); within a phonological word, ATR-harmony applies. Igbo has rich verbal morphology indicating tense and aspect as well as derivational affixes (see Uwalaka 1988 for an overview). There is, however, no verb–argument agreement in the language. Case distinctions (nominative-accusative alignment) can only be detected in the personal pronoun paradigm for 2nd and 3rd person singular (SG), see (2).²

The basic word order in an information-structurally neutral sentence is subject (SU) – verb (V) – indirect object (IO) – direct object (DO) – adjuncts (ADJ), see (3). This word order is strict, there is no scrambling-like operation within TP and adjuncts are confined to the clause-final position. The order can only be changed to express information-structural categories.

Following the syntactic literature on Igbo, we assume the clause structure in (4) for a declarative sentence with a transitive predicate (leaving adjuncts aside); traces are represented in< >:

Verbal projections are head-initial with specifiers linearized to the left of the selecting predicate. The external argument DP_ext obligatorily undergoes EPP-movement to SpecT; this is motivated by the observations that (a) the subject in Igbo precedes aspectual/temporal auxiliaries like gà in (5b), and (b) that there are no subjectless clauses in the language. SpecT must be occupied by an expletive if no XP moves there. We assume that the structurally highest verb (an aspectual auxiliary or, in its absence, the main verb) cyclically moves to T since it precedes sentential negation, see (6).

With this background on Igbo syntax, we come back to the kà–marking asymmetry. The asymmetry is found in constituent questions with simple wh-pronouns, as illustrated in (1), but also in sentences with D-linked wh-phrases introduced by the element òléē (though the appearance of kà there is subject to dialectal variation). Thus, kà can also follow non-SU wh-phrases:

kà-marking is only possible with clause-initial (ex-situ) wh-non-SUs. Wh-fronting in Igbo is optional, wh-XPs can also stay in-situ (on the surface, see Section 5 for details). But if they are in-situ, they cannot combine with kà:

In what follows, we will illustrate the patterns with the help of sentences that include simple wh-SU and wh-DO pronouns. Furthermore, we will use only matrix questions because Igbo does not have embedded questions; syntactically, this content is expressed by a relative clause and we never find kà-marking in relative clauses in the first place.⁴

3 The syntax of wh-constructions in Igbo

In this section we investigate the morphosyntactic properties of the wh-construction that exhibits the kà-marking asymmetry. We address the question whether these dependencies involve movement or base-generation and what kind of element the morpheme kà is.

4 The position of wh-subjects

Now that we know how non-SU questions are constructed, we can explore the structural position of wh-SUs. Due to Igbo’s strict SVO word order, SU wh-movement to SpecFOC would be string-vacuous. We thus need to apply syntactic tests to explore whether wh-SUs actually undergo Ā-movement or not. The tests suggest that wh-SUs cannot undergo (local) Ā-movement to SpecFOC but have to stay in SpecT. This also explains the absence of the focus marker in SU-questions. Evidence for the claim comes from a tonal reflex of Ā-movement and from ATB-movement. Further evidence against a purely morphological account of the absence of kà in SU-questions is provided in Section 4.2 based on long-distance SU-movement and island repair by resumption. In Subsection 4.3 we argue that the immobility of subjects is due to their high structural position and not to their features; we also highlight similarities between kà-marking and the that-trace effect.

5 On the nature of wh-in-situ

Before we present an analysis of the Igbo SU/non-SU extraction asymmetry, we need to clarify the nature of wh-in-situ. As we have shown in Section 2, non-SU wh-elements in Igbo can be in-situ or ex-situ. In this section we argue that wh-in-situ in simple questions is only apparent; real wh-in-situ only occurs in multiple questions. The status of wh-in-situ differs between languages (see Bayer & Cheng 2017 for an overview). Two basic types can be distinguished: In some languages, the in-situ wh-element does not undergo (overt) movement at all, while in others it moves in the syntax but the tail of the movement chain is pronounced. Igbo wh-in-situ seems to belong to the second type of languages. There are three reasons for this conclusion: Wh-in-situ licenses parasitic gaps (pgs); wh-in-situ does not give rise to Beck intervention effects; and wh-in-situ is island-sensitive. (36) illustrates that wh-in-situ (with matrix question interpretation) can license pgs just like wh-ex-situ (see also Ogbulogo 1995: Chapter 5.2 on pgs in Igbo); note that ‘to price’ is expressed by an inherent complement verb construction:

Since pgs can only be licensed by syntactic movement but not by LF-movement (Engdahl 1983), the in-situ wh-word in (36b) must have moved to SpecFOC (see also Branan & Sulemana 2018 on pg-licensing by wh-in-situ in Bùlì). Next, consider Beck intervention effects (Beck 1996; 2006; Kotek 2017): This effect arises when elements like e.g. negation intervene between the base position of the wh-element and its putative landing site in the left periphery at LF; movement across the intervener in the syntax obviates the intervention effect. As (37) shows, Igbo does not exhibit intervention effects when on the surface an in-situ wh-element is in the c-command domain of negation (or a focus sensitive element like ‘only’). Both the ex-situ and the in-situ question in (37) are grammatical, hence, the wh-element in (37b) must have undergone syntactic movement across the intervener.

Finally, (38) illustrates the island-sensitivity of wh-in-situ: the non-SU in-situ wh-word cannot occur inside a relative clause or an adjunct island:

The reader may object that we have provided grammatical examples from DP-coordination islands and PP-islands in the previous sections, repeated below, in which wh-words do occur in-situ:

How can this split between islands be explained? We propose the following: In those cases where wh-in-situ inside an island is grammatical (as in (40)), the whole island, including the wh-word, undergoes movement to SpecFOC; then the lower copy of the moved phrase is pronounced, leading to wh-in-situ on the surface. In the cases where wh-in-situ inside an island is ungrammatical (see (39)), pied-piping is impossible; and since the island blocks movement of the wh-element alone, ungrammaticality results. Independent evidence for this claim comes from the observation that overt pied-piping is possible with DP-coordination and PP-islands, but not with complex NP and adjunct islands, see (41c-d):

Since wh-in-situ inside an island that cannot be pied-piped is ungrammatical, we can conclude that one non-SU wh-element must undergo movement to SpecFOC in Igbo. It can, however, optionally be pronounced either in its landing site (leading to wh-ex-situ) or in its base position (resulting in apparent wh-in-situ), see Pesetsky (2000) for this kind of pronounciation rules for wh-XPs.²¹ Hence, in simple questions with a single in-situ wh-non-SU (‘Ada saw who?’), the wh-element has in fact moved to SpecFOC in syntax. Wh-SUs, on the other hand, do not have the option to move to SpecFOC, as we have argued. In multiple questions, however, there must be real wh-in-situ elements: We know that FOC can host only a single specifier (see Section 3.2), so all wh-elements but one indeed have to stay in-situ in syntaxn. We thus predict that a “real” in-situ wh-word in a multiple question should not be able to license a pg, for example. This prediction is borne out, see (42) where we try to license one pg by the ex-situ wh-DO (in the sole SpecFOC) and another one by the in-situ wh-IO, which is impossible:

6 An anti-locality approach to the asymmetry

The aim of this section is to provide a formal implementation of the morphosyntax of Igbo wh-movement. To summarize what we want to derive: In an interrogative sentence, exactly one wh-phrase has to move to the left periphery (SpecFOC); other wh-phrases, if present, need to stay in-situ. Since Igbo does not exhibit superiority effects, each of the wh-phrases in a multiple question can in principle undergo movement to SpecFOC. The wh-phrase that has moved to SpecFOC can be pronounced either in its landing site or in its base position. Wh-movement is possible for non-SUs, while local wh-movement of SUs is blocked. Morphologically, the head FOC⁰ is realized as kà if a phonologically overt XP occupies SpecFOC, otherwise it remains silent. Furthermore, we need to account for two facts: (a) Local Ā-movement of the SU is possible after all if it is an instance of relativization (movement of a relative operator), and (b) the absence of the embedded complementizer nà under long SU-movement (that-trace effect).

In what follows, we provide an analysis of these facts; however, we will not offer an explanation for the absence of superiority effects in Igbo (but see Bošković 1999 for relevant discussion) since this is orthogonal to the main questions. Our analysis is based on three core ingredients: (a) An anti-locality constraint on movement, (b) the violability of constraints, and (c) the variable size of clauses (not all left-peripheral heads are always projected).²² Anti-locality is a concept that has been used to derive a number of phenomena in which subject Ā-movement is restricted (see a.o. McCloskey 1990; Ouhalla 1993; Cheng 2006; Schneider-Zioga 2007; Brillman & Hirsch 2015; Bošković 2016; Erlewine 2016; Douglas 2017; Erlewine 2017; Sheehan, Douglas & Ranero 2018; Deal to appear). The core idea of anti-locality is that there is not only an upper bound on movement, but also a lower bound: Movement steps must not be too short, i.e. they need to cross a certain number of projections. We will adopt the definition of anti-locality from Erlewine (2016) in (43) according to which movement of an XP to the specifier of the immediately dominating projection is too short (for other definitions see Abels 2003; Grohmann 2003):

This constraint rules out local SU-wh-movement since SUs in Igbo occupy SpecT and would have to move to SpecFOC to check the probe [*FOC*]↑ on FOC⁰. However, since FOC⁰ and TP are sisters (see (22)), this movement step qualifies as too short by (43) and is thus blocked.

Our second central assumption is that syntactic constraints are violable. In Igbo, neither SSAL nor a constraint we will introduce below, viz. the Focus Criterion, are fulfilled in every clause. SSAL is violated in relative clauses where local SU-movement of the relative operator is possible even though it is too short (see also Erlewine 2016 on the violability of SSAL); and the Focus Criterion, which demands [Foc]-XPs to occupy SpecFOC (Rizzi 2006) is violated in multiple questions in Igbo where only one of the wh-phrases can move to this position. Still, these sentences are grammatical, so constraints must be violable without immediately causing the crash of the derivation. A framework in which constraints are violable is Optimality Theory (OT, Prince & Smolensky 1993), which we will adopt in what follows. In particular, we will assume a cyclic optimization procedure (see Heck & Müller 2000 et seq. on this concept in syntax) in which every clause is subject to optimization, from the most deeply embedded clause upwards. Technically, we assume that the numeration N that hosts all the elements that can be used in the derivation is structured into subnumerations SN, one SN containing the material of a clause. In a sentence with one level of embedding, the numeration thus contains two SNs, one for the matrix clause and one for the embedded clause: [_N { _SN1 … }, { _SN2 … }].

The third ingredient of the analysis is the variable size of clauses. We adopt the idea from Grimshaw (1997) that functional heads, especially those that constitute the C-domain (FOC⁰, FORCE⁰), are not necessarily projected; whether they are present in the structure or not is subject to optimization. Diverging somewhat from Grimshaw, we assume that all the heads of the split-C domain are – in addition to the other material used in the clause – present in the subnumeration for that clause, and the evaluation component will tell us whether they are projected or not. There is one restriction on the presence of heads in subnumerations in Igbo, though: Recall that Igbo does not have embedded questions; thus, in general, the FOC-projection cannot be present in such clauses, there should not even be a FOC-head in the subnumeration. We can model this by saying that a [–root] FORCE-head in Igbo cannot select FOCP, just TP; as a consequence, the generator cannot produce a candidate that includes FOCP. Alternatively, we could postulate a negative coocurrence restriction between [–root] FORCE⁰ and FOC⁰ in the same subnumeration. We adopt the view advocated in Grimshaw (1997) that the input consists of predicate-argument-structures plus tense/aspect specifications, while higher functional heads (here, those that make up the split C-domain) can be added freely by the generator (apart from the FOC-head in a SN that also contains [–root] FORCE⁰, see above). To summarize, the candidates in our competitions will, among other things, differ in whether FORCE⁰ and FOC⁰ are projected. Clauses can thus be bare TPs or FORCEPs (that either contain or do not contain FOCP). In the OT-system we develop below, a matrix SU-question will not contain FOCP, while a non-SU-question and multiple questions will include FOCP. This also explains their diverging properties concerning focus marking.

We assume that the candidates that can compete must be based on the same subnumeration, i.e. on an identical set of lexical and functional elements. However, there is no requirement that all the (functional) elements of the subnumeration must be used in the derivation, this is subject to optimization. Some left-peripheral heads may remain unused in the subnumeration. We need the following constraint for the analysis, see (45); most of them are general constraints on movement from the OT- and the non-OT-literature.

FOC.C requires every [Foc]-XP to be in the criterial focus position SpecFOC and thus movement of [Foc]-XPs; indirectly, it thereby favors the projection of FOCP; C.TYPE requires the encoding of clause-type, and since this feature is hosted by FORCE⁰, it also favors the projection of FORCEP. The other constraints concern movement: Stay prohibits movement (antagonist of FOC.C), LR prohibits movement that does not result in feature discharge, and SSAL militates against too local movement steps as defined in (43). FCH is a constraints that enforces the discharge of operation-inducing features such as the probe feature [*Foc*]↑ on FOC⁰, which needs to be c-commanded by a [Foc]-XP. The ranking in (46) produces the Igbo wh-movement pattern:

In short, the ranking favors the projection of FORCEP and FOCP in a sentence with a wh-XP as well as movement of wh-elements (which bear the feature [Foc]) to SpecFOC to minimize violation of FOC.C. However, this movement is blocked if it involves a too local or an untriggered movement step. Let us see how the system works in the relevant contexts. We start with a simple matrix SU-question (the wh-XP is represented as DP_[Foc] in what follows). The input contains material for the TP (with the subject in SpecT due to Igbo’s EPP-property) as well as the heads FOC_[*FOC*]↑ and FORCE. The question is whether these two heads are projected or not.²³ Table 1 illustrates the competition (for reasons of space, FORCE is abbreviated as F in the tableaux; “t” is a mnemonic device for lower copies of movement; a discharged probe is indicated by a strike-through; head movement to T is not represented in the candidates and the input): C1 is the candidate that neither projects FOCP nor FORCEP and C2 projects only FORCEP; C3 contains only FOCP but does not move the wh-SU to SpecFOC; C4 is like C3 but projects FORCEP in addition; in C5 (without FORCEP) and C6 (with FORCEP) the wh-SU does move to SpecFOC. All the candidates that do not project FORCEP incur a violation of C.TYPE and are thus harmonically bound by the candidates that differ only in the presence of FORCEP. If FOCP is not projected, the wh-SU violates FOC.C since it cannot be in SpecFOC in the output. To avoid a violation of FOC.C, FOCP must be projected. If it is projected but the wh-SU does not move to SpecFOC (C3, C4), the probe feature on FOC⁰ cannot be checked, which leads to a violation of FCH. This is avoided if the wh-XP moves to SpecFOC (C5, C6) where it can discharge the probe feature on FOC⁰. However, in the case of a matrix wh-SU, movement of the wh-XP to SpecFOC is blocked because it is too local and thus fatally violates SSAL. As a consequence, the best option is not to project FOCP at all and to leave the wh-SU in SpecT, even though this violates the low-ranked FOC.C. Thus, local wh-movement to SpecFOC is excluded in this system, wh-SUs stay in SpecT – and this is exactly what we have argued for in this paper. Furthermore, since FOCP is absent in C2, there is also no host FOC⁰ for the focus marker kà, which thus cannot surface, see the VIs in (23).

Table 1

Competition for a simple matrix SU question.

F, FOC[*Foc*]↑, [TP DP[FOC] [T’ T [ASPP Asp [VP tDP[FOC] [v′ v [VP V DP ]]]]]]		SSAL	C-Type	FCH	LR	FOC-C	STAY
	C1: [TP DP[FOC] [T′ T … [VP V DP ]]]		*!			*
	C2: [FP F [TP DP[FOC] [T′ T … [VP V DP ]]]]					*
	C3: [FOCP FOC[*FOC*]↑ [TP DP[FOC] [T′ T … [VP V DP ]]]]		*!	*		*
	C4: [FP F [FOCP FOC[*FOC*]↑ [TP DP[FOC] [T′ T … [VP V DP ]]]]]			*!		*
	C5: [FOCP DP[FOC] [FOC′ FOC[*FOC*]↑ [TP t′DP[FOC] [T′ T … [VP V DP ]]]]]	*!	*				*
	C6: [FP F [FOCP DP[FOC] [FOC′ FOC[*FOC*]↑ [TP t′DP[FOC] [T′ T … ]]]]]	*!					*

Next consider a simple matrix non-SU-question, exemplified by a wh-DO in Table 2. The candidates are by and large the same as before. But we leave out those that do not project FORCEP (since they are harmonically bound); we can thus also ignore C.TYPE in the tableaux. Either FOCP is not projected or, if it is, the wh-XP moves to SpecFOC or stays in-situ. The optimal candidate projects FORCEP and FOCP and moves the wh-DO to SpecFOC.

Table 2

Competition for a simple matrix non-SU (DO) question.

F, FOC[*FOC*]↑, [TP DP [T′ T [ASPP Asp [VP tDP [v′ v [VP V DP[FOC] ]]]]]]		SSAL	FCH	LR	FOC-C	STAY
	C1: [FP F [TP DP [T′ T … [VP V DP[FOC] ]]]]				*!
	C2: [FP F [FOCP FOC[*FOC*]↑ [TP DP [T′ T … [VP V DP[FOC] ]]]]]		*!		*
	C3: [FP F [FOCP DP[FOC] [FOC′ FOC[*FOC*]↑ [TP DP [T′ T … tDP[FOC] ]]]]]					*

The crucial difference to the competition with a wh-SU is that the movement step of the wh-DO, which avoids a violation of FOC.C, is not too local as the wh-DO starts out from a lower position than SpecT. Since FOCP is projected, FOC⁰ can be realized by the focus marker kà. Given the VIs in (23), kà surfaces if the wh-XP is pronounced in its landing site SpecFOC (wh-ex-situ); if the lowest copy is pronounced (wh-in-situ), FOC⁰ remains silent.

In multiple questions, represented here with a wh-SU and a wh-DO, projection of FOCP (and FORCEP) plus movement of the wh-non-SU to SpecFOC is optimal, see Table 3. C1 does not project FOCP and hence violates FOC.C twice, once per wh-pronoun. Projecting FOCP without moving a wh-XP there (C2) additionally incurs a violation of FCH for the unchecked probe feature of FOC⁰. To avoid violations of FOC.C, the wh-XPs should move to SpecFOC. In C3 the wh-non-SU moves, while in C4 the wh-SU moves. This reduces the violations of FOC.C by one in each case, the movement step discharges the probe feature on FOC (no violation of LR or FCH). However, movement of the wh-SU is too local and fatally violates SSAL, as in a simple SU-question. This also holds for C5 in which both wh-XPs undergo movement. Hence, the best candidate is the one that moves a non-SU-wh-XP to SpecFOC and leaves the other wh-XP (here, wh-SU) in-situ. In a multiple question without a wh-SU, any of the wh-non-SUs can move to SpecFOC due to the absence of superiority in Igbo; movement of more than one wh-non-SU to SpecFOC is still excluded even though none of the movements violates SSAL: What is fatal here is that all but the first instance of wh-movement violate LR since they do not lead to feature-discharge: The probe feature of FOC⁰ is discharged by the first wh-XP moved to SpecFOC.

Table 3

Competition for a multiple matrix question (wh-SU, wh-DO).

F, FOC[*FOC*]↑, [TP DP1[FOC] [T′ … [VP tDP1[FOC] [v′ v [VP V DP2[FOC] ]]]]]]		SSAL	FCH	LR	FOC-C	STAY
	C1: [FP F [TP DP1[FOC] [T′ T … DP2[FOC] ]]]				**!
	C2: [FP F [FOCP FOC[*FOC*]↑ [TP DP1[FOC] [T′ T … DP2[FOC] ]]]]		*!		**
	C3: [FP F [FOCP DP2[FOC] [FOC′ FOC[*FOC*]↑ [TP DP1[FOC] [T′ T … tDP2[FOC] ]]]]]				*	*
	C4: [FP F [FOCP DP1[FOC] [FOC′ FOC[*FOC*]↑ [TP tDP1[FOC] [T′ T … DP2[FOC] ]]]]]	*!			*	*
	C5: [FP F [FOCP DP2[FOC] [FOC′ DP1[FOC] [FOC′ FOC[*FOC*]↑ [TP tDP1 … tDP2 ]]]]]	*!		*		**

Since FOCP is projected in a multiple question, the FOC-head will be realized by kà if the non-SU-XP in SpecFOC is pronounced in this position, see the VIs in (23). Note that FORCE⁰ which is always projected in matrix clauses (to fulfill C.TYPE) is not pronounced; it surfaces as nà only in embedded declarative clauses, but never in main clauses. We can model this by saying that nà spells out declarative FORCE_[–root].

So far we have implemented the fact that in local SU-questions wh-movement is impossible and the focus marker must be absent. However, recall that local SU-movement is possible after all in relative clauses as indicated by the tonal reflex of movement there, see (26a). We also need to clarify why kà is absent in relative clauses in general in Igbo. Let us first consider the structure of relative clauses: Research on relativization in the split-CP system has come to the conclusion that relative operators (RelOPs, XP_[OP]) target SpecFORCE and not SpecFOC, relativization is not an instance of focus movement, see a.o. Rizzi (1997; 2006); Douglas (2017). In fact, we said at the beginning of the section that FOC is absent in embedded clauses in Igbo (FORCE_[–root] does not cooccur with FOC⁰ in the same subnumeration); hence, the focus marker cannot surface in RCs. The structure of a (SU) relative clause thus looks as in (47):²⁴

If RelOPs target SpecFORCE, this movement must be triggered by a feature. To achieve this, we postulate that FORCE⁰ in relative clauses bears a probe feature [*OP*]↑ that is discharged when it is c-commanded by a RelOP, i.e. by a XP with the feature [OP] in its specifier (akin to the probe-feature [*FOC*]↑) on FOC⁰). RelOP-movement to SpecFORCE takes place in order to discharge this feature and avoid a violation of FCH. However, given the subranking SSAL ≻ FCH, this would still preclude movement of a SU-RelOP from SpecT to SpecFORCE as too local given the structure in (47), causing a violation of (43). Hence, we need a further constraint that enforces RelOP-movement and outranks SSAL. In parallel to FOC.C, we postulate a RelOP Criterion (REL.C) that demands RelOPs to occupy SpecFORCE, see(48).

Given this ranking, the projection of FORCEP and movement of RelOP to SpecFORCE is enforced (to allow for the satisfaction of the high-ranked constraint REL.C) even if it is too local. One may say that this analysis of local SU-RelOP-movement is just a technical implementation of the facts (RelOP-movement is obligatory in Igbo). However, we would like to point out that cross-linguistic research on Ā-dependencies (de Vries 2005; Dryer 2013; Šimík 2018) has revealed that RelOP-movement (in wh-relatives) is obligatory, i.e. there do not seem to be clear cases of an (overt) RelOP in-situ, while wh-movement is not obligatory (in the sense that many languages use wh-in-situ as the main or at least as one strategy for forming constituent questions; and even in those with wh-movement, wh-in-situ is an option used in certain contexts, e.g. in multiple questions). Put differently, wh-movement does not seem to be necessary to derive a question interpretation (see Hagstrom 2003 for an overview of non-movement-based interpretation algorithms for questions) and probably happens for independent reasons (see Šimík 2018: 6 for a list). In contrast, RelOP-movement to the left periphery seems to be necessary for interpretation: Relative clauses are properties derived by lambda-abstraction that is brought about by syntactic movement. Thus, one could also say that the RelOP-Criterion is not just a very high ranked violable constraint, but probably an inviolable constraint of the grammar. In any case, the behavior of Igbo RelOPs is not special at all from a cross-linguistic perspective.²⁵

Up to this point, we have considered only matrix questions. We now turn to long-distance wh-movement. Recall that in the embedded clause the focus marker kà must be absent regardless of the grammatical function of the wh-XP in this clause; in the matrix clause the opposite holds: The marker kà must be present, also regardless of the function of the wh-XP in the embedded clause (i.e. the SU/non-SU split is neutralized under long movement). In addition, the declarative complementizer nà that spells out the embedded FORCE-head must be absent if the embedded subject undergoes long-distance movement (= that-trace effect). Before we can look at the derivation of this pattern, we need to consider the structure of the clauses in a long wh-dependency as well as the nature of long movement. Let us start with the structural aspects. As mentioned at the beginning of this section, the FOC-projection is not part of the subnumeration for embedded clauses (Igbo does not have embedded questions, Force_[–root] cannot select FOCP). This explains why we never see a focus marker in the embedded clause: If the FOC-head cannot be present, it cannot have an exponent. What is optimized in the embedded clause is thus mainly whether FORCEP is projected. In the matrix clause subnumeration both FOC⁰ and FORCE⁰ are present and can potentially be projected (subject to optimization).

The second issue we need to clarify concerns the nature of long wh-dependencies. A lot of evidence has been accumulated for the thesis that long-movement applies successive-cyclically in smaller steps (see a.o. van Urk 2015 for a recent overview). Technically, intermediate movement steps can be enforced by the Phase Impenetrability Condition (PIC, Chomsky 2000 et seq.) that requires material from the complement domain of a phase head to move to the edge (the specifier) of the phase to remain accessible for operations outside of the phase. CP is usually considered to be a phase, but what is the phase head in a split CP-system with a variable clause size? We follow suggestions in Bošković (2014) in assuming that the highest functional projection of the verbal domain is the phase. To enforce a wh-XP to be at the edge of the phase we adopt the OT-constraint PHASE.BALANCE by Heck & Müller (2000; 2003):

In the context of our discussion of long wh-movement in Igbo, this constraint basically says the following: For every FOC-head with the probe feature [*Foc*]↑ in the numeration (viz. in any of the subnumerations), there must be a [Foc]-XP at the edge of the phase in the phrase marker built from the current subnumeration, in order to keep this [Foc]-XP accessible for the discharge of the probe feature on FOC⁰ (which is contained in a different subnumeration).²⁶ This requirement holds unless there is another [FOC]-XP somewhere in the numeration, a case we will not consider here. Note that PB enforces movement steps that do not result in discharge of a probe feature on the phase head; rather Heck and Müller take intermediate movement steps to be a repair operation that violates LR. PB is the highest ranked among our constraints, see (50) (we ignore REL.C here since it is irrelevant for wh-movement):

We can now go through the derivations. We start with the optimization of the first cycle, i.e. the embedded clause, see Table 4. In the input there is material that constitutes the TP as well as the FORCE-head, but no FOC-head; FOC.C is thus necessarily violated by any of the candidates. C1 does not project FORCEP. As a consequence, C.TYPE is violated, but crucially, PB is fulfilled: There is a FOC-head with a probe feature in the numeration (the subnumeration for the matrix CP) and hence, the sole available [FOC]-XP, viz. the wh-SU of the embedded clause, is required to occupy the edge (specifier) of the phase; the phase is the highest projection of the clause, here the TP. Since a SU is in SpecTP anyway (due to the EPP-property of Igbo), a wh-SU is also at the edge of the TP-phase without undergoing any additional movement steps. A way to avoid the violation of C.TYPE is to project FORCEP, see C2 and C3. However, if this is done, the wh-SU must move to SpecFORCE in order to circumvent a violation of the highest ranked constraint PB, because now FORCEP is the phase and the wh-SU needs to occupy SpecFORCE (this movement does not lead to discharge of a feature of FORCE⁰ and hence violates LR). But crucially, movement of the wh-SU to SpecFORCE violates SSAL as it targets the specifier of the next higher projection. Since PB and SSAL outrank C.TYPE, the best option in the embedded clause of a long SU-question is not to project FORCEP at all. In fact, in Grimshaw’s (1997) OT-system (with a different set of constraints), the embedded clause in a long SU-dependency also turns out to be just a TP, it lacks higher functional structure that could host the complementizer. Hence, this clause is just a TP. This also explains why the embedded complementizer nà must be absent in this context (that-trace effect): The head it realizes (FORCE⁰) is not present in the winning candidate.

Table 4

Competition for the embedded clause, long SU-question.

I:	F, [TP DP[FOC] [T′ T [ASPP ASP [VP tDP[FOC] [v′ v [VP V DP ]]]]]]	PB	SSAL	C-TYPE	FCH	LR	FOC-C	STAY
	C1: [TP DP[FOC] [T′ T … [VP V DP ]]]			*			*
	C2: [FP DP[FOC] [F′ F [TP tDP[FOC] [T′ T … [VP V DP ]]]]		*!			*	*	*
	C3: [FP F [TP DP[FOC] [T′ T … [VP V DP ]]]]	*!				*	*

If a wh-non-SU, say the DO of the embedded clause, is to undergo long wh-movement, the best option is to project FORCEP and to move the wh-DO to its specifier, see Table 5. The crucial difference to long wh-SU-movement is that the wh-non-SU is not automatically at the edge of the phase when FORCEP is not projected, since it starts moving from within the vP. Thus, not projecting FORCEP and not moving the wh-DO leads to a fatal violation of PB, see C1 in Table 5. The latter can be avoided by moving the wh-DO to an outer specifier of T, see C2; however, since FORCEP is still missing, C.TYPE is violated. Both violations are avoided if FORCEP is projected and the wh-non-SU moves to SpecFORCE, see C3. Since FORCEP is projected and we are dealing with an embedded declarative clause, FORCE⁰ is realized by nà, i.e. there is a complementizer in the embedded clause in cases of long non-SU wh-movement.²⁷

Table 5

Competition for the embedded clause, long non-SU-question.

I:	F, [TP DP [T′ T [ASPP Asp [VP tDP [v′ v [VP V DP[FOC] ]]]]]]	PB	SSAL	C-TYPE	FCH	LR	FOC-C	STAY
	C1: [TP DP [T′ T … [VP V DP[FOC] ]]]	*!		*			*
	C2: [TP DP[FOC] [T′ DP [T′ T … [VP V tDP[FOC] ]]]]			*!		*	*	*
	C3: [FP DP[FOC] [F′ F [TP DP [T′ T … [VP V tDP[FOC] ]]]]					*	*	*
	C4: [FP F [TP DP [T′ T … [VP V DP[FOC] ]]]]	*!					*

Finally, we turn to the derivation of the matrix clause, see Table 6. Recall that in the matrix clause subnumeration there is both a FOC- and FORCE-head. The competition is basically the same as for matrix non-SU-questions, compare Table 2. FOC.C requires movement of the wh-XP to SpecFOC and thus the projection of FOCP in the matrix clause. This movement step can always take place; it is never too short since the wh-XP starts moving from the edge of the embedded clause (SpecFORCE or SpecT), but not from matrix SpecT (which is occupied by the matrix clause subject). Thus, no matter whether the wh-XP represents the SU or a non-SU of the embedded clause, it can move to SpecFOC and is forced to do so by the constraint ranking. The projection of FORCEP is enforced by C.TYPE, but this constraint is left out for reasons of space in Table 6; all candidates fulfill it (the ones that violate it are harmonically bound by those in the tableau). PB is irrelevant in the matrix clause (and hence also left out) since there is no further FOC-head with a probe feature in the numeration. If the wh-XP in matrix SpecFOC is pronounced in its terminal landing site, the FOC-head will be realized as kà in accordance with (23). Thus, we have correctly derived the third generalization about long wh-movement in Igbo: There must always be a focus marker in the matrix clause, regardless of the grammatical function that the wh-XP has in the embedded clause.

Table 6

Competition for the matrix clause in a long question.

I:	F, FOC[*FOC*]↑ [TP DP [T′ T … [VP V [FP/TP DP[FOC] [FORCE′/T′ … ]]]]]]]]	SSAL	FCH	LR	FOC-C	STAY
	C1: [FP F [TP DP [T′ T … [VP V [FP/TP DP[FOC] [F′/T′ … ]]]]]]]]]				*!
	C2: [FP F [FOCP FOC[*FOC*]↑ [TP DP [T′ T … V [FP/TP DP[FOC] [F′/T′ … ]]]]]]]]]		*!		*
	C3: [FP F [FOCP DP[FOC] [FOC′ FOC[*FOC*]↑ [TP DP [T′ … V [FP/TP tDP[FOC] [F′/T′ … ]]]]]]]					*

One issue that has not been addressed so far is the source of the tonal reflex of movement in Igbo. Recall that this reflex is triggered in subject relative clauses and in the embedded clause of a long SU-question, but not in a matrix SU-question and in any kind of non-SU-Ā-dependency. What does the reflex realize? What these contexts have in common is that there is a gap (viz. an unpronounced copy of the local subject) in SpecT because the SU of the clause has undergone movement (to SpecFORCE in relative clauses and to the matrix clause in long SU-questions, respectively). Therefore, we propose that the tonal reflex spells out an (unpronounced) copy in SpecT; note that this also explains the presence of the reflex in the that-trace repair configuration with the special complementizer sí in (35a). The tone is floating and attaches to the closest element to its right at PF, viz. the T-head that hosts the finite verb in Igbo (see Amaechi to appear for further details on the attachment site, the underlying shape and the effects of this floating tone on the inherent tones of the element in T). In other contexts such as matrix non-SU questions, multiple questions, non-SU relative clauses, the embedded clause of a long non-SU question, SpecT is occupied by the overt subject of the minimal clause and hence, the tonal reflex is absent.

The very last context that we want to address in this section is the derivation of the coordination examples in (30c) and (31c), where one of the conjuncts is a resumptive pronoun (RP) that resumes a wh-element in clause-initial position, which must be followed by the focus marker. Crucially, using a RP is not the only way to express the intended content; the wh-element can also stay inside the &P, see (30a) and (31a). Optionality between a sentence without a RP and one with a RP in a single language has been taken as evidence that the candidates that represent these clauses do not compete (see Salzmann 2017). In fact, our definition of reference set at the beginning of this section makes their competition impossible because the sentences are not based on the same subnumeration (the same lexical and functional elements): (30c) and (31c) contain the RP, which is absent from the subnumeration that (30a) and (31a) are based on. How are the two types of examples derived? The derivation of the wh-in-situ examples (30a) and (31a) is straightforward. The input looks as in the previous tableaux: There is TP-material as well as FOC⁰ and FORCE⁰ in the numeration. C.TYPE enforces the projection of FORCEP. The wh-XP needs to fulfill FOC.C, hence FOCP is projected, but movement of the wh-element to SpecFOC is blocked by the &P-island, which we take to be due to an inviolable constraint of the generator, viz. an island violating candidate cannot be generated in the first place. If the wh-element is contained in a non-SU-&P, the only way to fulfill FOC.C is to pied-pipe the whole &P to SpecFOC (and to percolate the [Foc]-feature to &P), so that the probe feature on the FOC-head can be discharged (the competition thus basically corresponds to the one for wh-non-SU-movement in Table 2). This presupposes that any constraint that may militates against pied-piping (see Heck 2008) must be ranked lower than FOC.C in Igbo. The &P that contains the wh-element can then be pronounced in SpecFOC (in which case it must be followed by the focus marker, as predicted) or in-situ (and the focus head remains silent). If the wh-element is contained in a SU-&P, movement of the entire &P is blocked by SSAL, compare the tableau for local wh-SUs in Table 1. Thus, a SU-&P containing a wh-element as one of its conjuncts stays in SpecT; the best option is then to not project FOCP in the first place.

The derivation for the cases that involve a RP, viz. (30c) and (31c), is very different. In this case, the numeration contains TP-material (including the RP), FOC⁰, FORCE⁰ as well as a wh-XP with the feature [Foc]. The TP is generated with the resumptive as one of the conjuncts inside &P. If FOCP is projected, a wh-XP needs to be in its specifier to discharge the probe feature on FOC⁰. There is no wh-XP inside the TP, but still one in the numeration. So this wh-XP is externally merged in SpecFOC and thereby satisfies FCH (it c-commands the probe feature on FOC⁰). As usual, C.TYPE enforces the projection of FORCEP. In addition, the wh-XP in SpecFOC semantically binds the RP inside the &P. FOC⁰ is realized by kà because an overt XP (the externally merged wh-XP) occupies its specifier – and this holds independently of whether the pronoun that resumes this wh-element is contained inside a SU- or a non-SU-island, hence the absence of the SU/non-SU split in these examples. Since there is no wh-movement in this derivation, neither the &P-island nor any other constraints on movement such as SSAL are violated. In essence, base-generation of a wh-XP plus having a theta-marked RP inside the TP is another way in Igbo to circumvent violations of constraints on movement, in particular island constraints and SSAL.²⁸ An interesting question is when a RP can be part of the numeration in the first place. As mentioned in fn. 18, the possibility of repair by resumption is highly restricted in Igbo. It is not always available, not even for all strong islands. More research on resumption in Igbo is needed to provide an answer; we leave this for future research.

7 Conclusion

We have investigated a subject/non-subject extraction asymmetry in wh-constructions in Igbo. While (ex-situ) wh-non-SUs need to be followed by the morpheme kà, wh-SUs must not co-occur with kà. Based on novel data we have argued that despite surface appearance, this asymmetry is not just a morphological phenomenon where exponence is sensitive to grammatical functions; the presence or absence of kà does not simply reflect any inherent properties of the wh-element. Rather, the overt split reflects a deeper syntactic asymmetry: The morpheme kà is a focus marker that realizes the left-peripheral head FOC⁰ whenever an overt XP occupies its sole specifier. kà surfaces with ex-situ wh-non-SUs because they move to SpecFOC⁰, while wh-SUs cannot undergo local Ā-movement to this position, they have to stay in the canonical position SpecT. However, wh-SUs are not per se immobile in Igbo; they do undergo wh-movement to SpecFOC in long-distance questions and in relative clauses. Based on the subject-object inversion construction, we have argued that it is the structurally highest XPs within the TP that is frozen for Ā-movement to the minimal SpecFOC. We attribute the SU/non-SU asymmetry to an anti-locality constraint on movement which prohibits too local movement steps. Local wh-SU-movement would qualifies as too local and is hence blocked; local wh-non-SU elements start from a lower position and do not violate this condition and hence, they can move freely. We use an OT-system to model the Igbo facts because anti-locality as well as the requirement to fill SpecFOC by a [FOC]-bearing XP are violable in Igbo. The analysis also captures local SU-relativization as well as the distribution of the focus marker in long-distance wh-dependencies and the that-trace effect. Furthermore, we have argued that wh-in-situ in simple questions in Igbo is only apparent: In every sentence with at least one [FOC]-bearing element, exactly one of them must move to SpecFOC; the optionality between wh-in-situ and wh-ex-situ is the result of optionality in the pronounciation of the bottom or the head of the wh-movement chain at PF.

Abbreviations

1/2/3 = 1^st/2^nd/3^rd person, ACC = accusative, ADJ = adjunct, ATB = across-the-board movement, ATR = advanced tongue root, CNPC = complex noun phrase constraint, Conj = conjunct, DEM = demonstrative, DO = direct object, DFCF = doubly-filled Comp filter, FOC = focus marker, FUT = future, IO = indirect object, OT = Optimality Theory, N = numeration, NEG = negation, NOM = nominative, OP = relative operator, pg = parasitic gap, P = preposition, PFX = prefix, PL = plural, PROG = progressive, PST = past, RC = relative clause, RP = resumptive pronoun, SCO = strong cross-over, SG = singular, SN = subnumeration, SU = subject, WCO = weak cross-over, <X> = lower copy (trace) of X

Funding Information

This research was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – project number 317633480 – SFB 1287, Project C05 (Georgi). Publication of this research is supported by the Singapore Ministry of Education under WBS R-103-000-145-115 to guest editor Michael Yoshitaka Erlewine.

Competing Interests

The authors have no competing interests to declare.