Absence of Clausal Islands in Shupamem

Abstract

Decades-long research on islands has led to the conclusion that island constraints are candidates for language universals. A recent surge in research on islandhood in African languages has revealed some would-be island configurations that are transparent for $\overline{A}$- dependency formation. In this article, we show that in Shupamem, all clausal configurations expected to have the status of opaque island domains fail to block the formation of long-distance $\overline{A}$- dependencies involving object ex situ focus. In support of the claim that $\overline{A}$- movement has occurred in such cases, we rely on evidence from three wh- movement diagnostics (weak crossover effects, reconstruction phenomena and quantifier float). Furthermore, we show that non-movement dependencies across purported island boundaries in the language are also possible through the licensing of “island”-internal negative concord items by external non-local negators. We conclude that clausal island effects fail to materialize in Shupamem ex situ focus constructions and negative concord item-licensing domains. Based on an exploratory typological survey of islands in African languages, we indicate a trend toward varying degrees of island permeability in the area, concluding that while Shupamem is not an isolated example, it features one of the most permissive grammars known to date in this respect.

1. Introduction

Certain configurations that are expected to have the status of syntactic islands across languages seem to permit extraction1 of categorially diverse arguments and adjuncts in Shupamem. For example, sentential subject constructions (1a), conditional clauses (2a), temporal clauses (3a), and clausal complements of definite nouns (4a) admit the displacement of domain-internal DP direct objects (1b), non-PP adverbs (2b), locative PPs (3b) and manner PPs (4b) to a left-edge focus position.2

(3)	a.		má
			VIEWP3
	b.

The data are not only surprising from the standard Generativist perspective on islands, but they also suggest that island escape in Shupamem shows no sign of category-based selectivity of the type documented by Hein (2020a, 2020b, n.d.). We might entertain two broad analytical options with regard to the data in (1–4): either the prominent constituent (X) has undergone $\overline{A}$- movement out of the purported island or X is base-generated in its surface position and binds an empty category in the suspected island, as schematized in (5a) and (5b), respectively.

In this paper, based on several diagnostics, we argue for the analysis in (5a). Contrary to the expectation that the structures in (1–4) constitute syntactic islands, we conclude that they are transparent for the formation of wh- $\overline{A}$- movement dependencies and thus do not constitute syntactic islands in Shupamem. Explaining the absence of island effects in these domains and their implications for syntactic variation/parameterization (à la Kandybowicz 2009) is beyond the scope of this article and is left for future research.

The remainder of this paper is organized as follows: Section 2 provides relevant background information on the grammar of Shupamem, the morpho-syntax of focus-cleft $\overline{A}$- configurations in the language, and the diagnostics that we will use to argue for $\overline{A}$- movement out of suspected islands. In Section 3, we present the apparent movement of a variety of constituents out of six purported island configurations. This includes the displacement of subjects, objects and adverbs out of sentential subjects, two complex noun phrase constructions (definite relative clauses and clausal complements of definite nouns), and three adjunct clauses (temporal, reason and conditional clauses). We argue for movement out of these configurations in Section 4, based on three well-known diagnostics for $\overline{A}$- movement.4 In Section 5, we show that non-movement dependencies across purported island boundaries in the language are also possible by demonstrating the licensing of “island”-internal negative concord items by external non-local negators. Section 6 reviews other diagnostics that for independent reasons yield non-decisive results when applied to Shupamem and thus do not distinguish movement from in situ analyses in the language. Section 7 concludes.

2. Empirical and Analytical Background

Section 2.1 briefly reviews certain grammatical facts that will be relevant for the forthcoming discussion, namely, Shupamem word order, the complementizer system and pronominal resumption. Section 2.2 introduces the $\overline{A}$- configuration that we will consider with respect to “island” extraction in the language and presents the diagnostics that will be used to argue for $\overline{A}$- movement out of the relevant clausal domains in this paper.

2.1. Background on Shupamem

Shupamem (ISO 639-3: bax) is an Eastern Grassfields Bantu language whose speech community numbers approximately 420,000 speakers in the Western Province of central Cameroon (Eberhard et al. 2021). Figure 1 below situates the Shupamem speech community within Cameroon/Africa.

This section outlines a number of grammatical facts that will be relevant for the forthcoming discussion. The basic word order of the language is subject–verb–object–x (6a), where x may be an indirect object (6b) among other syntactic functions (6c).5^,6

Much of the data in the following sections involve relative and complement clauses. Complement clauses in Shupamem are introduced by a single invariable non-inflecting complementizer (7a), whereas relative clauses (RCs) feature one complementizer that follows the RC head and agrees with it in number and noun class (7b–7e) (e.g., ) alongside the RC-final and morphologically invariable relative particle .

Pronominal resumption is implicated in a number of constructions considered in this article. It varies primarily based on syntactic position and animacy, being obligatory in the focus of fronted subjects (8a), human/animate-denoting direct (8b, 8c), indirect (8d) and oblique (8e) objects. (Overt) resumption is unavailable when inanimate-denoting direct objects are focused (8f).

These resumption patterns vary in the context of RC configurations. Definite RCs lack resumption in direct object positions (9a), but obligatorily resume subjects (9b, 9c), indirect objects (9d), and oblique constituents such as instruments (9e).

An anonymous reviewer suggests that resumption may be pervasive and pronouns obligatorily present in syntax even in phonetically null positions like direct object positions (e.g., 8f, 9a), as Saah (1992, 1994) has argued for in Akan. This hypothesis receives preliminary support in Shupamem from resumption patterns in “topic drop” dialogues. In this context, an object pronoun must be pronounced (10B–B${}^{\prime }$) in a clause immediately following the mention of a human-denoting antecedent (e.g., Rájé in (10A)). In contrast, a null position is licit (11B) if it resumes an inanimate-denoting antecedent in the immediate context (e.g., , ‘bananas’, in 11A), suggesting the presence of a null pronoun in (11B). Together with the pattern of obligatory resumption in focus-cleft constructions (8), obligatory resumption in such dialogues, which precludes a movement analysis, is seemingly consistent with an Akan-like analysis of pervasive resumption in Shupamem.7

Data from epithet licensing, however, suggest that null positions in focus-cleft constructions represent traces in the language. Recent advances in the debate over the categorial status of epithets seem to rule in favor of a structure in which a pronoun serves as an anchor to the epithetic expressive material (Patel-Grosz 2012). Moreover, it has been shown that an epithet can only anchor to a pronoun, not a trace (Demirdache and Percus 2011). This makes epithets an ideal tool to determine whether null positions in certain contexts represent null resumptive pronouns or traces. When R-expression direct objects are focus-clefted in Shupamem, they may be resumed using a resumptive pronoun in the form of an epithet. If the object is human-denoting, a weak resumptive pronoun must be used for both expressions to corefer (12a). In the absence of a pronominal anchor, coreference is unavailable (12b).

Faced with inanimate-denoting extractees, which lack overt resumption morphology in the language (8f), we find that expressive material in situ may not anchor to the linearly adjacent null position to serve as an expressive descriptor of the ex situ content (13b).

We conclude that in such focus-cleft configurations, the null object position reflects the presence of a trace, as a (null) resumptive pronoun would be expected to yield a well-formed epithetic structure in the language.8

2.2. $\overline{A}$- Movement in Shupamem

Building on the base sentence in (14a), the focus-clefting of a prominent constituent in indicative (14b) and interrogative mood (14c) is a productive $\overline{A}$- configuration in Shupamem.

Focus–cleft constructions are best described in terms of relativization, where the prominent constituent, preceded by an expletive subject a, heads a relative clause (RC). The prominent constituent is followed by the present-tense copula pǎ in positive declarative clauses (14b), but in casual speech it may form a portmanteau with the expletive subject, resulting in the form áǎ. The copula is absent altogether in interrogative clauses (see Nchare 2012, p. 452).9 More often than not, the focus-clefting of direct objects entails a change in the tonal melody of the transitive predicate selecting that direct object (as in (14b–c)). In ex situ question formation (i.e., interrogative focus-clefts) but not in affirmative indicative focus-clefts, this change is accompanied by an H melody on the ex situ constituent. A focus marker is absent in focus-cleft contructions and a left-edge relativizer must follow the focused constituent (14b) (

Table 1. Morphosyntactic properties of focus-cleft constructions in Shupamem.

	Expletive Subject	Copula	Focus Marker	Relativizer
Indicative Focus-cleft	✓	✓	✗	✓
Interrogative Focus-cleft	✓	✗	✗	✓

) in tandem with a right-edge relativizer particle.

If focus-cleft constructions in Shupamem are amenable to a head raising analysis of RCs à la Kayne 1994, it would follow that prominent constituents (i.e., focus-clefted XPs) are $\overline{A}$- extracted from their original external merge positions.10 To determine whether this is the case and ultimately argue in favor of a movement account of these $\overline{A}$- configurations in the language, we will employ several diagnostics of $\overline{A}$- extraction in this section, namely weak crossover effects, reconstruction for scope, and quantifier float.

Crossover phenomena concern binding relations between $\overline{A}$- moved elements and more deeply embedded pronouns. They come in several varieties. According to two widely known crossover generalizations, $\overline{A}$- moved elements cannot move across c-commanding pronouns that they end up binding (Strong Crossover, see Wasow 1979) (15a), nor can they move across non c-commanding pronouns that they bind (Weak Crossover, see Postal 1971) (15b).

(15)	a.	* Whoi did they inform himi that Joan would call ___? (Postal 1993, p. 543)
	b.	* Whoi did hisi sister call ___ a moron? (Postal 1993, p. 540)

When a wh- object crosses over a non c-commanding pronoun in subject position, weak crossover effects (16) are observed in Shupamem. This fact is consistent with an analysis in which the prominent/relativized constituent has undergone $\overline{A}$- movement.11

Reconstruction effects regard any configuration in which an $\overline{A}$- displaced constituent behaves as if it occupies a lower structural position with respect to interpretive considerations. Reconstruction phenomena are not all equally reliable as diagnostics for movement in $\overline{A}$- dependency formation. Reconstruction for Condition A of the binding theory, whereby an anaphor must be locally c-commanded by its antecedent (Chomsky and Lasnik 1993; Guéron 1979; Reinhart 1976) is controversial. The controversy is especially salient with respect to picture noun anaphoric expressions (e.g., Safir 2004, p. 116), which are seemingly exempt from Condition A (see Charnavel and Bryant 2023, for a recent analysis of the facts in English).12 Condition C of the binding theory, whereby R-expressions must be universally free, offers a more promising route, but it introduces a different set of challenges.13 Therefore, we capitalize on wh–quantifier interaction in reconstruction for scope to determine whether $\overline{A}$- dependencies that implicate movement are present in Shupamem. We assume that scope ambiguities in configurations that involve an ex situ wh- item and an in situ quantifier phrase (QP) are due to syntactic reconstruction of the wh- item in a position below the QP (Aguero Bautista 2001). In the absence of reconstruction, the wh- object in (17) admits a single, individual entity reading in Spanish, but reconstruction yields the availability of a pair-list reading, such that each witness in (17) refers to a different person as the direct object of pegar (‘hit’).

(17)	A	quién	dijo	cada	testigo	que	María	le-quería	pegar?
	to	whom	said	each	witness	that	Maria	him-wanted	to.hit
	‘Whom did each witness say that María wanted to hit?’ (Aguero Bautista 2001, p. 172)

The availability or not of pair-list readings in wh-quantifier interactions has been recently used to argue against wh- movement in the relevant contexts in Awing, another Grassfields Bantu language of Cameroon (Fominyam 2021).14 To ensure the reliability of this test in Shupamem, we must exclude the possibility that quantifier raising (QR) can derive a pair-list reading by raising to a position above the ex situ constituent in question at LF. If QR can derive pair-list readings in the language, such interpretations are expected to be available when an indefinite noun subject c-commands a universally quantified direct object, contrary to fact (18a). In such constructions, the indefinite expression takes scope over the universal quantifier but not vice versa (18a). In stark contrast, only the pair-list reading is available when the QP is focus-clefted to a position above the indefinite subject (18b). These facts suggest the absence of a QR operation in the language.

In critical response to this conclusion, one might argue that the subject position is special in its topical properties (a recurring proposal in Bantu linguistics, see Bresnan and Mchombo 1987; Downing and Hyman 2016; Givón 1976; Henderson 2006, among many others), forcing a singular, individual reading of the subject, which would force the unavailability of wide-scope (pair-list) readings in (18a). Not only does the lack of scope ambiguity (and the concomitant unavailability of an individual reading) in the ex situ variant in (18b) speak against this hypothesis, but the same effect is observed when the relevant DP occupies a direct object position alongside an indirect object QP, as in paradigm (19).

Since the universal quantifier cannot outscope the in situ indefinite object in (19a), we conclude that QR cannot derive pair-list readings in structures involving QPs that are c-commanded by other scope-taking material, making the interaction between wh- ex situ and more deeply embedded quantifiers a reliable diagnostic of reconstruction and thus $\overline{A}$- movement in the language. Having ruled out QR as the source of pair-list readings, we conclude that for structures that involve a universally quantified subject c-commanding a wh- pronoun (20a), scope ambiguity, and specifically the availability of a pair-list reading in their wh- ex situ counterparts, as in example (20b), must be the result of reconstruction of the wh- element to a more deeply embedded in situ position in which the quantifier outscopes it. If it were not for reconstruction into the in situ position of the wh- object, response B would be infelicitous as an answer to (20b). In fact, such pair-list answers are felicitous, making reconstruction for scope a reliable diagnostic test for $\overline{A}$- movement in the language.

Additional corroborative evidence for $\overline{A}$- extraction out of the domains in question comes from quantifier float data. Quantifier float (QF) refers to configurations in which a quantifier is construed together with its associate noun or wh- item despite a non-local relation between them. If quantified items are licensed syntactically, then QF configurations are expected to be unavailable when the associate vacates a strong island configuration, as in Irish English (McCloskey 2000) and French (Baunaz 2008). Root clauses with quantified wh- objects (21a) facilitate QF configurations in the language (21b), which yield the same interpretations as when the quantifier and its associate appear together in situ (21a) or in a fronted position (21c).

The reliability of the floating quantifiers diagnostic hinges on the possibility that we reject an alternative analysis under which floating quantifiers do not float after all, but rather attach to in situ null pronouns that corefer with inanimate ex situ wh- items. This alternative analysis is strengthened at first glance by the fact that clause-medial quantifiers may attach to null in situ anaphoric pronouns in “topic drop” contexts (22B) in which movement is excluded.15

Although this puts movement and base-generation analyses of purportedly floating quantifiers as in (21b) on equal footing at first, the demonstrated unavailability of stranded epithets in the focus-clefting of inanimate-denoting objects (13b) favors the view that left-edge prominent constituents are $\overline{A}$- extracted and leave behind traces in the relevant cross-clausal constructions in which we observe syntactic connectivity effects in Section 4. Because the quantifier in structures like (21b) is right-adjacent to a trace (and crucially not a null resumptive pronoun), it is truly floating. As will be demonstrated below, floating quantifiers are admissible in Shupamem when a quantified wh- object is focus-clefted out of suspected strong clausal island configurations.

Once we demonstrate the possibility of focus-clefting arguments and adjuncts internal to suspected island structures in Section 3, we adduce evidence for $\overline{A}$- extraction of the prominent constituent (Section 4) based on the three $\overline{A}$- diagnostics discussed in this section. Although subjects, objects and adjuncts may all serve as prominent constituents linked to positions in the relevant “islands” (Section 3), the diagnostics that we employ in this article are applied to direct objects. Subjects cannot be used as extractees in crossover scenarios that require $\overline{A}$- displaced elements to move over “island”-internal pronominal expressions that they end up binding. Therefore, only more deeply embedded constituents within a purported island may be focus-clefted to test for their coreference with a pronoun in a subject position. Reconstruction effects similarly require structures in which moved elements are c-commanded by structurally higher “island”-internal material at some level of analysis, thus precluding the application of the diagnostic to subjects.

3. Extraction out of Clausal Domains Is Pervasive

The following subsections illustrate the focus-clefting of subjects, objects and adjuncts internal to sentential subject constructions (Section 3.1), complex NP constructions (Section 3.2) and adjunct clauses (Section 3.3).

3.1. Sentential Subject Constructions

Sentential subject constructions (23a) admit the focus-clefting of subjects (23b), objects (23c) and adverbial adjuncts (23d).

3.2. Complex Noun Phrase Constructions

The CNPCs that we investigate in this paper are definite relative clauses (subject RCs (24a) and object RCs (24b)) and clausal complements of definite nouns (24c).

It is possible to focus-cleft subjects (25a), direct objects (25b) and temporal adjuncts (25c) internal to definite RCs. In the cases of argument movement considered below, a subject is extracted from an object RC (25a), whereas an object is extracted from a subject RC (25b).

Clausal complements of definite nouns attest to similar facts concerning focus-clefting (26).

3.3. Adjunct Clauses

In Shupamem, temporal clauses, reason clauses and conditional clauses, all of which constitute strong islands cross-linguistically, align with sentential subject constructions and complex NP configurations in the language with respect to their transparency for focus-clefting of core and non-core constituents.

3.3.1. Temporal Clauses

We use example (27a) as a base temporal clause. With a matrix verb in the present progressive form (, ‘be buying’) and a future-oriented temporal adverb (, ‘tomorrow’) inside the adjunct clause, we preempt matrix readings of adverbs when the adverb serves as the prominent constituent. Subjects (27b), objects (27c) and adjuncts (27d) internal to temporal adjunct clauses may all be focus-clefted.

3.3.2. Reason Clauses

Reason clauses are introduced in Shupamem by (‘because’, literally translated as ‘on the matter that…’), as in our base sentence (28a). Like temporal clauses, reason clauses also permit the focus-clefting (28) of material internal to them.

3.3.3. Conditional Clauses

Conditional clauses in the language are formed using the conditional morpheme , preceded for the most part by the logical subject (29a). As with the other domains investigated in this section, material internal to such clauses may be focus-clefted (29).

If the data presented in this section instantiate $\overline{A}$- movement, as we argue in Section 4, then all domains reviewed in Section 3 are unexpectedly transparent for $\overline{A}$- extraction of core and non-core constituents in Shupamem and therefore cannot be considered islands.

4. Arguments for $\overline{\mathit{A}}$- Movement out of Clausal Domains

In this section, we apply three diagnostics to test and ultimately argue for $\overline{A}$- extraction out of the clausal domains under investigation. In all configurations, focus-clefting of material inside the domain in question gives rise to both crossover and reconstruction effects. Additionally, quantifier float data furnish corroborating evidence that $\overline{A}$- extraction from the domains in question has taken place. In what follows, we apply the diagnostics exclusively to direct objects in each configuration for the reasons enumerated at the end of Section 2.2.

4.1. Sentential Subject Constructions

When material internal to sentential subjects is focus-clefted, weak crossover effects (30) are observed (i.e., the focus-clefted wh- object cannot bind a higher non c-commanding pronoun).

Reconstruction effects are observed when wh- objects inside sentential subjects (31a) are focus-clefted (31b).

The availability of a pair-list interpretation, as made evident by the fact that (31B) is a possible answer to both examples in paradigm (31), indicates that the displaced wh- item in this structure is base-generated in a low position inside the sentential subject construction where it is c-commanded by the domain-internal universally quantified subject.

Sentential subject constructions that contain a quantified wh- object yield the same interpretation when the quantifier and its associate are both in situ (32a) and when the wh- item is focus-clefted and the quantifier is stranded (32b).

If the relation between the quantifier and its associate is syntactic, and a trace underlies the null position adjacent to the quantifier, in line with the argument from epithet stranding (Section 2.1), then QF data provide further evidence that objects internal to sentential subject constructions can vacate their external merge position inside this clausal domain.

4.2. Complex Noun Phrase Constructions

Focus-clefting of material internal to definite relative clauses gives rise to weak crossover effects (33).16

In clausal complements of definite nouns as well, focus-clefting of clause-internal material gives rise to weak crossover effects, as illustrated in example (34).

In the same vein, reconstruction effects are observed when material that is internal to definite RCs (35a) and clausal complements of definite nouns (36a) is focus-clefted (35b, 36b). As before, using an ex situ wh- object in both complex NP structures yields a scope ambiguity in its interaction with a c-commanding domain-internal universally quantified subject. This is consistent with the hypothesis that $\overline{A}$- extraction is implicated in the focus-clefting of material internal to definite relative clauses and clausal complements of definite nouns.

Quantified wh- objects that are focus-clefted out of relative clauses (37a) and clausal complements of definite nouns (38a) are construed together with floating clause-internal quantifiers as if they both occupy a position inside the complex noun phrase construction (37b, 38b).

Our diagnostics are thus consistent with the conclusion that $\overline{A}$- movement derives the position of prominent constituents originating in a variety of complex NP configurations.

4.3. Adjunct Clauses

The adjunct clauses investigated in the following subsections (temporal, reason and conditional clauses) align with the transparency of sentential subject constructions and complex NPs with respect to the $\overline{A}$- movement diagnostics considered in this article.

4.3.1. Temporal Clauses

Focus-clefting of material inside adjunct temporal clauses gives rise to weak crossover effects (39).

Reconstruction effects are observed when wh- elements inside adjunct temporal clauses (40a) are focus-clefted (40b). Whether a universally quantified embedded subject c-commands the interrogative pronoun (40a) or not (40b), a pair-list answer to such questions is available alongside a unique entity answer. These data are consistent with an analysis in which the derivation of such examples implicates $\overline{A}$- extraction of the left-edge wh- item from a position internal to the temporal adjunct clause.

Facts from the domain of QF are also consistent with this analysis. Temporal clauses that host quantified wh- items (41a) facilitate QF. In example (41b), the focus-clefted wh- object is construed with its in situ floating quantifier.

4.3.2. Reason Clauses

Weak crossover effects (42) materialize when constituents internal to reason clauses undergo focus-clefting.

Reconstruction effects are observed when a wh- item that is c-commanded by a universal quantifier in reason clauses (43a) is displaced in a focus-cleft construction (43b).

Evidence from floating quantifier configurations provides further support for the status of fronted constituents as $\overline{A}$- extractees that originate inside reason clauses. Quantifiers that are stranded under focus-clefting of their wh- associate yield grammatical structures (44b) that are interpreted as if their wh- associates never moved (44a).

4.3.3. Conditional Clauses

Weak crossover effects arise in the focus-clefting of constituents that are thematically linked to a position in conditional clauses (45).

Furthermore, reconstruction effects are observed when wh- items inside conditional clauses (46a) are focus-clefted over c-commanding domain-internal QPs (46b).

Considerations involving quantifier float also support this analysis. QF configurations in which a conditional clause-internal floating quantifier is construed with its focus-clefted associate (47b) give rise to grammatical structures that are semantically equivalent to their in situ counterparts (47a).

4.4. Interim Summary

All clausal domains reviewed in Section 3, which are expected to be opaque for $\overline{A}$- extraction, give rise to weak crossover effects and manifest reconstruction effects within each domain when domain-internal direct objects are focused-clefted. In addition, QF data suggest connectivity between the prominent peripheral expression and the embedded clausal domain. Lastly, we found no new or exotic resumption patterns in the cases considered. That is, the resumption patterns attested in ex situ constructions that do not involve clausal embedding (8) are identical to those found in the instances involving clausal domains considered in this section, suggesting a unified derivational analysis. We conclude that analysis (5a) is an adequate account of the focus-clefting of constituents internal to these domains. The classic island configurations reviewed in Section 3 are thus transparent for $\overline{A}$- extraction of material internal to them and therefore do not constitute islands for wh- dependencies in Shupamem, at least as far as direct object extractions are concerned.

5. Additional Evidence for Absence of Clausal Islands in Shupamem

This section presents evidence from negative concord item (NCI) licensing as an additional argument for the non-island status of the clausal domains investigated in this article. We demonstrate that NCIs embedded within the relevant clausal domains are licensed by matrix negation. Thus, we argue that the permeability of clausal domains in the language is not limited to $\overline{A}$- extraction, but extends to probes “looking into” such configurations.

If the domains previously considered are indeed porous for $\overline{A}$- dependency formation, then we might expect constituents within those domains to be accessible to outside probes. In this section, we show that this prediction is borne out, relying on data from the licensing of N-words—a.k.a NCIs—which, in Shupamem, take the form of initial lexical items, as in (48a) (Nchare 2012, p. 404). NCIs are expressions that are licensed in the presence of root clause-level negation (48a) (they give rise to infelicitous NCI readings of N-words in the absence of negation in the language (48b)), yield single negation readings (48a) (Jespersen 1922), and may be used as fragment negative answers (49) (Giannakidou 2006).

NCI licensing is island-sensitive. This is motivated by considerations from various languages such as West Flemish (Haegeman and Zanuttini 1991) and Spanish, in which matrix negation cannot license an island-internal NCI (50) (Aranovich 1994, p. 209).

The clause-mate requirement on negation is also documented in Xhosa factive clauses by Carstens and Mletshe (2016). This is consistent with NCI licensing being a narrow syntactic phenomenon. Syntactic analyses of NCI licensing that are relevant for our discussion here involve feature agreement via the operation Agree (Chomsky 2000, 2001). See Zeijlstra 2008 for an analysis centered on agreement with [Negation] features and Carstens and Mletshe 2016 for an analysis involving [Focus] feature agreement.17

Further evidence that (some of) the Shupamem clausal domains considered in this article do not have island status comes from the fact that NCIs embedded in them are successfully licensed by domain-external negative morphemes. Consider first Complex NP Constructions. As discussed above, the RC domain constitutes a barrier to external NCI licensing in languages where RCs are strong islands. The data in (51) show that this domain is accessible to matrix negation in Shupamem, a fact that follows from the finding that RCs fail to have island status in the language. This finding generalizes to all RCs in the language. Kandybowicz and Nchare (2023) show that RC-internal NCIs in both restrictive and non-restrictive relative clauses in Shupamem are licensed by domain-external negative morphemes.

Other complex noun phrases align with RCs as far as the licensing of NCIs is concerned. Non-local negation may license NCI interpretations of N-words embedded inside clausal complements of nouns (52a) that are otherwise unavailable (52b). These facts are consistent with our analysis that clausal complements of nouns in Shupamem are not islands.

N-words embedded within reason clauses are similarly licensed under the scope of domain-external matrix negation (53a).

Thus, evidence from three suspected strong island configurations in the language indicates that they are porous for long-distance syntactic dependencies that do not involve movement. We are unable to bring forth evidence from the licensing of NCIs in the other clausal domains considered in this article (i.e., sentential subjects, temporal clauses, and conditional clauses) because of confounding factors. Properties specific to these three domains influence whether an N-word in Shupamem will be interpreted as an NCI or a Negative Polarity Item (NPI) (see Nchare (2012, p. 404) for more on the flexible bivalent NCI/NPI status of N-words in Shupamem). For example, in presupposition-dependent downward entailing environments that license NPIs (Condoravdi 2010; Von Fintel 1999) such as temporal clauses, conditional clauses, and sentential subjects under the scope of factive matrix predicates like ‘surprise’ (54a), N-words may be licensed in the absence of matrix negation. Because the addition of matrix negation has no effect on the licensing of N-words in these environments (54b), the NCI licensing diagnostic is not applicable in such domains.

6. Other (Indecisive) Diagnostics for $\overline{\mathit{A}}$- Movement out of Clausal Domains

While other tests may be used to detect $\overline{A}$- movement, at least four other possible diagnostics prove indecisive in the context of Shupamem clausal extractions: parasitic gap licensing (Section 6.1), superiority effects (Section 6.2), idiom formation (Section 6.3) and sluicing (Section 6.4). We discuss their applicability in Shupamem (a) in order to assure readers that a wide range of extraction diagnostics were considered and (b) to inspire confidence in skeptical readers that the diagnostics applied in Section 4 are truly the most decisive diagnostics we can appeal to.18

6.1. Parasitic Gap Licensing

Parasitic gap licensing (Engdahl 1983) is observed when the creation of a non c-commanding $\overline{A}$- gap (55a) licenses an otherwise illicit gap (55b). The other gap is thus “parasitic” on this $\overline{A}$- dependency and its licensing serves as evidence that $\overline{A}$- movement has occurred.

(55)	a.	Here is the influential professor that John sent his book to ___ in order to impress ___. (Engdahl 1983, p. 11)
	b.	* John sent his book to the influential professor in order to impress ___.

The unacceptability of a base sentence with a single illicit embedded gap is a prerequisite for the applicability of this diagnostic. If (55b) were acceptable, the grammaticality of (55a), with gaps in both positions, would not be decisive between movement and non-movement analyses.

In Shupamem, the ungrammaticality that results from there being an unpronounced position in the second occurrence of the object ndáp (‘house’) in sentence (56a), as in (56b), is remedied once the matrix object is focus-clefted (56c), indicating that the gap in the adjunct clause is parasitic on the creation of a non-c-commanding gap, in line with the $\overline{A}$- extraction of the focus-clefted constituent from its external merge position.

The data in (56b) and (56c) thus suggest that parasitic gap licensing might constitute another reliable $\overline{A}$- movement diagnostic in the language. Despite this promising result, parasitic gap licensing is not a stable test of $\overline{A}$- movement in Shupamem and thus it is unreliable at the current stage of this research project. During the first period of fieldwork on this study (December 2020–April 2021), the focus-clefting of inanimate-denoting material internal to suspected island configurations gave rise to parasitic gap licensing patterns like the one shown in (56b) vs. (56c). However, upon re-elicitation (May 2021–September 2021), base examples with single embedded gaps (i.e., structures like (56b)) were judged acceptable. This discrepancy in the data makes parasitic gap licensing a currently unreliable diagnostic when attempting to distinguish $\overline{A}$- movement from base-generation analyses in Shupamem. In the remainder of this subsection, we present the initial results of the parasitic gap licensing test as it was applied to the “island” domains under consideration in this paper. In these trials, otherwise illicit gaps were licensed once inanimate-denoting expressions were focus-clefted out of all suspected island domains, thus furthering the argument that the clausal domains in question lack island status based on parasitic gap licensing.

The focus-clefting of material internal to sentential subjects appears to license parasitic gaps inside subject CPs (57b) that are not licensed in the absence of focus cleft constructions (57a). Recall that in this and subsequent examples, the judgement of base sentences containing gaps has proved variable, which renders this diagnostic favorable to our movement approach, but presently unreliable until the data can be checked against judgements from more native speakers.

Similarly, when constituents internal to complex NPs are focus-clefted, otherwise illicit gaps (58a, 59a) appear to be licensed, as we illustrate below with definite relative clauses (58b) and clausal complements of definite nouns (59b).

The remaining strong “island” configurations that were investigated align with sentential subject and complex NP constructions with respect to parasitic gap licensing. Thus, otherwise illicit gaps inside reason clauses (60a) were licensed following the focus-clefting (60b) of material internal to them.

Similarly, the focus-clefting (61b) of material internal to conditional clauses appears to license otherwise illicit gaps inside these clauses (61a).

A similar pattern is observed in temporal clauses. When an illicit gap occurs in a temporal ‘after’ clause that is embedded in a structurally higher ‘before’ clause (62a), the focus-clefting of a co-referential object in the higher clause embedded below the matrix predicate renders grammatical the otherwise illicit gap (62b).

6.2. Superiority Effects

Superiority effects are observed in questions with multiple wh- elements when a structurally lower wh- item moves over a higher wh- item yielding ungrammatical outputs. Under the movement analysis sketched in (5a), otherwise licit focus-clefting of “island”-internal wh- items would be predicted to be blocked in the presence of a higher interrogative expression. The base-generation approach in (5b), however, would predict the absence of superiority effects in these cases, making the consideration of superiority effects a potentially decisive diagnostic for teasing apart movement from base-generation analyses in cases of purported island escape in the language.

Unfortunately, this diagnostic is not applicable in Shupamem due to the absence of superiority effects in the language (63–64), as in other West African languages such as Ikpana (Kandybowicz et al. 2023), Krachi (Torrence and Kandybowicz 2015), Akan (Saah 1994), and Yoruba (Adesola 2006).

The data in (63) and (64) show that in multiple wh- question constructions, any wh- expression may undergo focus-clefting. In the absence of focus-clefting asymmetries in multiple wh- source structures, we do not have recourse to the use of superiority effects as a diagnostic of $\overline{A}$- extraction in Shupamem.

6.3. Idiom Formation

Focus-clefting of “island”-internal idiom chunks would be predicted to yield idiomatic interpretations under the movement analysis sketched in (5a), on the assumption that all parts of the idiom must form a constituent at some stage of the derivation (as in English). Under the base-generation approach in (5b), only literal interpretations would be predicted to be available in these cases. In this way, idioms could offer a potentially decisive diagnostic between movement and base-generation analyses of purported cases of island extraction in the language.

Unlike idioms in English, idioms in Shupamem are a purely surface phenomenon. Only when all parts of the idiom appear linearly adjacent do idiomatic interpretations become available. Since movement of any sort, i.e., both $\overline{A}$- movement (65b, 66b, 67b) and A- movement (65d, 66d), precludes idiomatic interpretations, both movement and base-generation analyses correctly predict the absence of non-literal interpretations when “island”-internal idiom chunks are focus-clefted. These properties of Shupamem idioms are illustrated below for three distinct idiomatic expressions (‘X shockingly succeeded’ (65), ‘X is in deep trouble’ (66)), and ‘Who the hell is X?’ (67).19

Appealing to idioms, therefore, is not an effective diagnostic of $\overline{A}$- movement out of clausal domains in Shupamem.

6.4. Sluicing

Sluicing is a type of ellipsis where, in most cases, everything except for a wh- expression is elided (Merchant 2001; Ross 1969), as in the dialogue in (68).

Sluicing is island-sensitive in some languages, implicating movement in the derivation of the sluice (e.g., Nupe, see Mendes and Kandybowicz 2023). Given island sensitivity, the movement analysis (5a) would make the prediction that sluices containing surviving wh- expressions that originate in any of the so-called island structures under discussion in this paper should be unavailable, while the base-generation analysis (5b) would predict the possibility of such sluices. If this were true for Shupamem, then sluicing could serve as a decisive diagnostic between movement and base-generation analyses of purported long-distance $\overline{A}$- dependencies across “islands”.

Despite the promising nature of this test, sluicing is not a decisive diagnostic of overt $\overline{A}$- movement in Shupamem because sluicing in the language appears to have a wh- in situ source structure.20 Paradigm (69) below shows that sluicing of the second conjunct of an NP coordinate structure (69A), an island in the language (70), is possible (69B).

The acceptance of sluicing a second conjunct /wh/- despite the opacity of the second conjunct for extraction in NP coordinate structures (70) supports a move-and-delete derivation of such fragment answers in the language, as represented in (71). The source of sluice (69B) appears to be a wh- in situ structure.

Further evidence that the source of sluice (68B) is a wh- in situ structure comes from the tonal realization of (‘what’). In situ occurrences of surface with L tones (71a), while focus-clefted surfaces with H tones (64b).21 The L realization of ‘what’ in (69B), therefore, supports the wh- in situ source structure of Shupamem sluices. This entails that example (69B) does not necessarily involve actual wh- movement, but rather a wh- in situ + delete derivation. In further support of the in situ derivation of fragment answers in Shupamem, consider N-words that, by definition, can serve as fragment answers in the language (example (49) from Section 5 is repeated below as (72)).

Shupamem N-words cannot be fronted, as shown by the unacceptability of the focus-cleft variant (73b) of example (73a) with the N-word as the direct object to be fronted.

These data are consistent with an in situ derivation of the N-word fragment answer in (72B).22

Consequently, because of its wh- in situ source structure in the language, sluicing cannot be used as a decisive diagnostic to test whether movement out of the “islands” considered in this paper has occurred.

7. Conclusions

Focus-clefting direct objects out of the clausal domains discussed in this paper triggers weak crossover effects, gives rise to reconstruction effects based on wh-quantifier interaction, and feeds quantifier float in Shupamem. In addition, long-distance licensing of negative concord items across some of these domains suggests that these configurations are porous for (non-movement) dependencies that are otherwise unexpected across them. Furthermore, parasitic gap licensing within the structures in question, although presently not a stable judgement pattern, has been observed. The findings summarized in

Table 2. Arguments for $\overline{A}$- movement out of Shupamem clausal domains.

	Crossover Effects	Reconstruction Effects	Parasitic Gap Licensing	Quantifier Float
Sentential Subjects	✓	✓	(✓)	✓
Definite Relative Clauses	✓	✓	(✓)	✓
Cl. Complements of N	✓	✓	(✓)	✓
Temporal Clauses	✓	✓	(✓)	✓
Reason Clauses	✓	✓	(✓)	✓
Conditional Clauses	✓	✓	(✓)	✓

, capturing only those diagnostics for which we presented data from all relevant configurations, suggest an absence of clausal islands in the language.

This result is very surprising from the perspective of domain-specific Generativist theory inasmuch as the conceptual necessity for computationally efficient syntactic derivations entails the universality of strong islands (Chomsky 2008).23 The pervasive transparency of these domains in Shupamem also challenges domain-general accounts that center on the observation that long-distance $\overline{A}$- dependencies tax the human parser in different ways (Abrusán 2014; Kluender 1998, and references therein). Such accounts reduce island effects to considerations of online language processing. For example, given a relative notion of complexity in language processing that affects variation in acceptability across types of domains and types of fillers (see Hawkins 1999, for example), islands are not the universal product of syntax or its interfaces, but an emergent property of language that is expected to arise in different languages due to domain-general limitations on language processing. From this perspective, the across-the-board transparency of clausal domains in Shupamem is somewhat perplexing. This begs the question of the grammatical properties that facilitate their pervasive transparency above and beyond what the languages of Europe will have us hypothesize.24

From the perspective of crosslinguistic and areal variation in this domain as represented in the literature, the lack of clausal opacity in Shupamem is remarkable in its pervasiveness, but it cannot be dismissed as a unique quirk. Temporal and conditional adjunct clauses in Norwegian, unlike reason clauses, do not have the status of strong islands Bondevik et al. 2021; Faarlund 1992; Kush et al. 2018)25 and in Ancash Quechua, both arguments and adjuncts can $\overline{A}$- move out of wh- in situ islands so that they fail to have the status of weak islands (Cole and Hermon 1994). Similarly, recent papers on islands in the languages of Africa point to an areal trend whereby one or more suspected island configuration is transparent for the formation of long-distance $\overline{A}$- dependencies. Outside of Shupamem, we have identified seven such languages. All languages except for Swahili cluster areally in western Africa, of which four, including Shupamem, are Grassfields Bantu languages. The results of our survey of the literature at the time of writing are reported in

Table 3. Patterns of $\overline{A}$- dependency formation out of clausal domains in some languages of Africa.

	Akan ${}^{\mathit{a}}$	Avatime	Awing	Ikpana	Limbum	Medumba	Swahili
Sentential Subjects	✓	–	–	✗	–	N/A	–
Definite RCs	✓	✗	✓	✗	✓	✓	✓
Cl. Compls. of N	✓	✓	–	✗	✓	✓	–
Temporal Clauses	–	✗	✓	✓	–	✓	✓
Reason Clauses	✓	–	✓	✓	✓	–	✓
Conditional Cls.	–	–	–	✓	–	–	–

${}^a$ Here Akan relates specifically to Asante Twi.

, in which a “–” stands for currently unavailable data, “✓” represents permissible extraction and “✗” represents impossible extraction from the domain in question.

Shupamem remains the one language to date with the most clausal domains documented to be transparent for $\overline{A}$- extraction. The Asante Twi variety of Akan (Hein and Georgi 2021; Korsah and Murphy 2019) comes closest to it with at least four out of six transparent “island” domains. Regardless of the possible transparency of temporal and conditional clauses, though, Asante Twi is more restrictive than Shupamem since only DPs can $\overline{A}$- extract from “island” configurations in the former.26 In the Ghana-Togo Mountain language Ikpana, adjunct clauses of all varieties are transparent domains for $\overline{A}$- extraction, but other classic strong islands have strong island status (Kandybowicz et al. 2023). In Avatime, another Ghana-Togo Mountain language, clausal complements of definite nouns are $\overline{A}$- transparent inasmuch as a wh- item can be fronted from within (Devlin et al. 2021; Major and Torrence 2021), but relative clauses and temporal clauses are opaque domains (Devlin et al. 2021).27 Of the remaining languages in Table 3, three belong to the Grassfields Bantu group in Cameroon, like Shupamem, but do not have porous clausal domains to the same extent. In Awing, long-distance $\overline{A}$- dependencies can seemingly be formed between an ex situ wh- item and the position with which it is linked inside relative clauses, temporal clauses, and reason clauses (Fominyam 2021).28 In Limbum, clausal complements of Ns, definite RCs, and reason adjunct clauses are transparent for $\overline{A}$- extraction (Hein 2020a, n.d.). In Medumba, complex NPs and temporal clauses are escapable when it comes to $\overline{A}$- extraction (Keupdjio 2020). Finally, in Swahili, at least definite RCs (Gould and Scott 2019) and temporal and reason clauses (Scott 2021) allow for $\overline{A}$- extraction. Although this sample is admittedly limited for the purposes of making novel typological generalizations, it is clear that as previously argued for in other areas of grammar (see Bresnan 1990; Henderson 2011), an Africanist perspective is essential to refining syntactic theory in the domain of islands and locality.29

To conclude, in this paper we deployed several standard Generative diagnostics and argued for $\overline{A}$- extraction of direct objects out of a number of purported island domains in Shupamem. It is premature to conclude that Shupamem features transparent “island” configurations that undermine the universality of islands without also considering whether each of the domains in question truly has the syntactic structure requisite for it to be considered an island. For example, recent work by Sichel (2014, 2018) and Cinque (2020) argues that what appear to be surprising instances of successful extraction from relative clause “islands” in Mainland Scandinavian, Romance, and Hebrew do not actually involve movement out of complex NP structures (i.e., strong islands), but more accurately represent instances of extraction out of CPs of the sort that constitute weak islands in the languages. The clausal domains involved in such apparent violations of strong island constraints differ structurally from strong islands as traditionally construed in the Generative framework (Chomsky 1986, 2001). Therefore, they count among instances of “surface island variation” in the sense of Phillips (2013b) and do not undermine the validity of a universal constraint that derives the ungrammaticality of $\overline{A}$- dependencies across island configurations. Such surface island variation contrasts with “deep island variation” in Phillips’ typology of island violations, whereby no argument presents itself to support any structural differences between the extraction domains in question and strong island configurations, but $\overline{A}$- extraction from these domains is nonetheless grammatical. The next natural step is therefore to consider whether each clausal domain in Shupamem has the structure of a traditional island (from the Generativist perspective) and account for its $\overline{A}$- extraction transparency. We speculate that the transparency of all clausal “islands” in Shupamem is closely connected to the syntax of relativization, given that the syntax of relative clauses is implicated in all cases of $\overline{A}$- extraction discussed in this paper.