Category-Sensitive Escape from Islands in Limbum and Asante Twi

Abstract

While strong islands generally constitute domains from which extraction is not possible, it has been observed that under certain conditions, they may allow DP but not PP gaps. Based on the recent literature on Asante Twi (Kwa, Ghana) and on novel data from Limbum (Grassfields Bantu, Cameroon), this paper shows that strong island configurations in these two African languages are permeable to nominal extractees without restrictions, but the otherwise admissible movement of VPs and PPs is blocked. As DP-displacement from islands shows properties of $\overline{\mathrm{A}}$-movement; an explanation in terms of base-generation and binding of a covert resumptive pronoun, which is only available for nominal elements, is not feasible. Taking into account the overall distribution of overt and covert resumptive pronouns, for Asante Twi, an account of the selective island permeability in terms of repair by resumption, as suggested in previous literature, might be possible. For the Limbum pattern, however, this paper argues that such an approach seems implausible. It then goes on to develop an analysis of selective island permeability based on the distribution of $\varphi$-features and their interaction with complementizer agreement.

1. Introduction

This paper investigates island effects in two African languages: Asante Twi (Kwa, Ghana) and Limbum (Grassfields, Cameroon). Ever since the seminal dissertation by Ross (1967), islands have played a major role in syntactic theorizing. Essentially, islands are syntactic configurations that do not allow or at least considerably degrade ($\overline{\mathrm{A}}$-)movement dependencies between a gap that is located within the island domain and its filler, which is located outside of this domain. One commonly distinguishes strong islands, which do not allow any extraction, from weak islands, which allow the extraction of some phrases but not others. Subsequent decades of work on the topic have refined the view of islands and have led to a more nuanced picture of the empirical landscape. One consequence is that while island effects can be observed across a variety of different languages, they also show some degree of variation such that they are attenuated or even completely absent depending on properties of the island configuration itself (finiteness, type of island, and/or definiteness), the syntactic dependency (topicalization, wh-movement, and/or focus), or properties of the affected element (argument status, specificity, and/or category).

Of particular interest for this paper are two general observations. The first one is that strong islands may allow restricted extractions after all. Given the right conditions (e.g., with regard to tense, Chomsky 1986; Manzini 1992, and definiteness, Fiengo and Higginbotham 1981; Manzini 1992; Postal 1998), strong (or absolute) islands may allow a nominal gap. Weak (or selective) islands, on the other hand, allow nominal as well as PP gaps under the same conditions (Cinque 1990). The set of strong (absolute) islands typically comprises at least complex NPs (with a complement clause or a relative clause), adjuncts, subjects, and coordinate structures. Embedded wh-clauses constitute the classical weak island, but negatives and factives as well as extraposed constituents may also induce a weak island. Thus, while wh-fronting of a nominal is possible from an untensed adjunct (1a) as well as an embedded interrogative clause (2a), extraction of the PP is only accepted from the interrogative (2b) but not from the adjunct (1b).1

(1)	Extractions from strong adjunct island (Szabolcsi and Lohndal 2017, p. 4)
	a.	[DP Which topic]1 did you leave [adjunct without talking about――1 ]?
	b.	*[PP About which topic]1 did you leave [adjunct without talking――1 ]?
(2)	Extractions from weak wh-island (Szabolcsi and Lohndal 2017, p. 4)
	a.	[DP Which topic]1 did John ask [CP whether to talk about――1 ]?
	b.	%[PP About which topic]1 did John ask [CP whether to talk――1 ]?

One influential treatment of this dichotomy is due to Cinque (1990). He argues that the DP gap in strong islands is not a movement trace but rather a silent resumptive pronoun pro that is bound by the ‘displaced’ element, which is base-generated in its displaced position (see Borer 1984; Chomsky 1977; McCloskey 1990; Sells 1987 for base-generation plus binding approaches to overt resumptive pronouns). The escape from an island in (1a) is therefore only apparent because actual syntactic movement is not involved in the derivation of this structure. Since no empty resumptive element exists for PPs, such a derivation circumventing the island is not available in (1b). Under this view, the variability in extraction possibilities from strong islands (DP vs. PP) is not attributed to a variability of the island constraints themselves but rather to an independent property of the displaced element (existence of pro for DPs but not PPs). Example (1a) thus constitutes a ‘surface’ island violation in Phillips (2013a, 2013b) terminology. The variability in extraction possibilities from weak islands under this approach has a different source and may well be attributed to an underlying variability in the syntactic constraints themselves. That is, the extractions from weak islands as in (2) are ‘deep’ island violations Phillips (2013a, 2013b).

The second general observation relevant for the present paper is that overt resumptive pronouns in some languages alleviate an expected island violation (Aoun and Choueiri 2000; Borer 1984; Chomsky 1986; Keupdjio 2020; Koopman 1982, 1984, 2000; Korsah and Murphy 2020; McCloskey 1990; Scott 2021; Shlonsky 1992). For a subset of these languages, this effect has been argued to be due to the derivation of resumptive pronouns more generally: While a gap is created by syntactic movement of the displaced constituent, a resumptive pronoun involves base-generation and binding (Borer 1984; Chomsky 1977; McCloskey 1990, 2006; Rouveret 2011; Sells 1987; also see the discussion in Korsah and Murphy 2024). The amelioration of the island is therefore only apparent since a violation is never actually induced: that is, there is no repair because nothing is broken. This entails that resumption dependencies, in addition to being insensitive to islands, should also not exhibit other properties of movement. Nevertheless, in some languages, resumptive pronouns do behave just like movement gaps (Alexopoulou 2006; Aoun et al. 2001; Engdahl 1985; Keupdjio 2020; Koopman 1982, 1984, 2000; Korsah and Murphy 2020, 2024; Sichel 2014; Zaenen et al. 1981; see Salzmann 2017 for a recent overview).2 In cases where such resumptive pronouns still ameliorate island violations, this has been taken to be due to their phonological overtness as compared to a silent gap: that is, phonological overtness repairs (possibly as a Last Resort) an otherwise illicit output (Kandybowicz 2008; Perlmutter 1972; Pesetsky 1998; Shlonsky 1992). In turn, the island restriction then has to take the form of a PF-requirement (cf. Boeckx 2012; Griffiths and Lipták 2014; Hornstein et al. 2007; Lasnik 2001; Merchant 2001; Perlmutter 1972; Pesetsky 1998) rather than a constraint on syntactic operations/configurations per se (though see Boeckx 2003; Klein 2017; Müller 2014 for accounts of island circumvention by resumption in narrow syntax).

The two African languages discussed in this paper, Asante Twi and Limbum, exhibit a pattern of island sensitivity that is very similar to (1): that is, they allow $\overline{\mathrm{A}}$-extraction from strong island configurations for nominal extractees but not prepositional or verbal ones. In contrast to the English data above, however, the islands do not have to be tenseless. Both languages also show overt grammatical resumption in at least some contexts. In Asante Twi, NPs/DPs may freely undergo $\overline{\mathrm{A}}$-extraction from what are commonly held to be strong island configurations—complex NPs with a complement clause, complex NPs with a relative clause, adjuncts, and subjects—whereas PP- and also VP-extraction triggers an island effect (Hein 2017, 2020; Korsah 2017; Korsah and Murphy 2020, 2024; Saah 1994). Korsah and Murphy (2020, 2024) argue that all DP-extraction leaves a resumptive pronoun that terminates the $\overline{\mathrm{A}}$-movement dependency. The resumptive pronouns are overt for animate DPs. The gap that appears with the extraction of inanimate object DPs is explained as the result of a PF-deletion rule. The island-circumventing nature of resumptive pronouns is accounted for by treating islands as PF-constraints (Boeckx 2012; Griffiths and Lipták 2014; Hornstein et al. 2007; Lasnik 2001; Mendes 2020; Mendes and Kandybowicz 2023; Merchant 2001; Perlmutter 1972; Pesetsky 1998) that are evaluated derivationally prior to the PF-deletion of the inanimate resumptives (an instance of counterbleeding). PPs and VPs do not leave resumptive pronouns, and their extraction from an island therefore leads to ungrammaticality. However, undermining this approach, Hein and Georgi (2021) show that some non-referential nouns do not incur island violations either, despite not leaving a resumptive pronoun.3 As suggested by Hein and Georgi (2021), it therefore seems as though the islands are selectively porous for elements whose lexical head is of the category noun, while they are opaque for the extraction of prepositional or verbal constituents (though see Korsah and Murphy 2024 for potential issues with this and a tentative argument that these non-referential noun types do leave an obligatorily PF-deleted resumptive pronoun after all).

Based on novel, hitherto unpublished, data, I show that islands in Limbum, a Grassfields Bantu language, exhibit a strikingly similar category sensitivity: NP/DP-elements may freely undergo $\overline{\mathrm{A}}$-extraction from complex NPs with complement clauses, complex NPs with relative clauses, and adjunct clauses, while movement of PP- and VP-elements is blocked. Like in Asante Twi, an overt resumptive pronoun in the tail of an $\overline{\mathrm{A}}$-dependency is obligatory for subjects and possible for objects. However, the occurrence of the latter is restricted to displaced bare nouns if they are interpreted as specific. And even then, the object resumptive is merely an option in addition to a gap. An explanation of the selective island-sensitivity in terms of deletion of underlying resumptives parallel to that of Korsah and Murphy (2020, 2024) for Asante Twi therefore seems implausible since the necessary PF-deletion rule would have to target the unnatural class of all resumptives except those with a specific bare noun antecedent. Again, an account of the observed pattern of island-insensitivity as due to a local PF-requirement on the pronunciation of the tail of $\overline{\mathrm{A}}$-movement seems insufficient. Rather, it appears that $\overline{\mathrm{A}}$-extraction from islands is directly sensitive to the category of the affected element: blocking PP- and VP-movement but allowing NP/DP-movement.

The current paper therefore contributes to a growing body of work on islands in African languages that shows that (at least some) would-be island configurations are consistently permeable for $\overline{\mathrm{A}}$-extraction (of at least some types of elements) (cf. Schurr et al. 2024 on Shupamem, Smith 2024 on Mende, Korsah and Murphy 2020, 2024 on Akan, Keupdjio 2020 on Medumba, Gould and Scott 2019; Scott 2021 on Swahili, Georgi and Amaechi 2020 on Igbo). While the similarity of the pattern of island escape in Asante Twi and Limbum makes a unified approach seem desirable, no attempt at such an approach will be undertaken here. For Asante Twi, an account that treats islands as PF-constraints that can interact with PF-deletion of underlying resumptive pronouns (Korsah and Murphy 2020) cannot be excluded in light of the new arguments presented in Korsah and Murphy (2024). For Limbum, however, such an approach seems unreasonable. Building on the pattern of complementizer agreement in the language, I will therefore develop an account of the selective behaviour of islands that exploits the presence of $\varphi$-features on nominals and their absence on prepositional and verbal elements.4 Moreover, concerning the displacement of nominal elements, I will restrict myself to the displacement of objects.

The remainder of the paper is structured as follows. Section 2 is dedicated to escape from islands in Asante Twi. It summarizes Korsah and Murphy (2020), introduces the relevant construction in Section 2.1, and presents arguments for its $\overline{\mathrm{A}}$-nature in Section 2.2. The selective island-sensitivity and how resumption can repair island violations is discussed in Section 2.3. Section 2.4 presents Hein and Georgi’s (2021) work on the island-insensitivity of noun types that do not leave resumptive pronouns upon extraction as well as Korsah and Murphy’s (2024) recent treatment of it. In Section 3, I discuss the island situation in Limbum. Section 3.1 introduces the á-focus construction in Limbum, and Section 3.2 provides some arguments that it involves $\overline{\mathrm{A}}$-movement. The selective island-sensitivity of this movement is treated in Section 3.3, while I argue that repair by resumption is implausible in Section 3.4. A tentative proposal as to how the category-sensitivity of islands in Limbum can be accounted for is presented in Section 4. Section 5 concludes the paper.

2. Escaping Islands in Asante Twi

2.1. The Focus Construction in Twi

Asante Twi is a dialect of Akan, a Kwa language (Niger–Congo) spoken in Ghana (Dolphyne and Kropp Dakubu 1988; Kropp Dakubu 2009). The language’s basic word order in a neutral declarative clause can be identified as SVO. Adverbs generally have to appear in the clause-final position (3).

(3)	Asante Twi neutral declarative clause
	Kofí	á-si	dán	ɛnora.
	Kofi	prf-build	house	yesterday
	‘Kofi has built a house yesterday.’

The language has an ex situ focus strategy (encoding contrastive focus) in which a constituent appears in the left periphery of the clause followed by the focus marker na (4). This construction has received considerable attention in previous literature (see Ameka 1992; Amfo 2010; Boadi 1974; Ermisch 2006, 2007; Genzel and Kügler 2010; Hein 2017; Korsah and Murphy 2020; Ofori 2011; Pfeil et al. 2015; Saah 1988).

(4)	Asante Twi object focus (Korsah and Murphy 2020, p. 831)
	Bayérέ1	na	Kofi	kítá	――1.
	yam	foc	Kofi	hold
	‘It is yam that Kofi is holding.’

It is also used to form ex situ wh-questions like (5).

(5)	Asante Twi object question (Korsah and Murphy 2020, p. 845)
	Déέn1	na	Yaw pέ ――1?
	what	foc	Yaw like
	‘What does Yaw like?’

Generally, the focus of an animate object requires the presence of a resumptive pronoun in the base position (6a). Inanimate object focus seems to force a gap in the base position of the object, while a resumptive pronoun is ungrammatical (6b) (Saah 1992; Saah and Goodluck 1995; Saah 1994).

(6)	Animate vs. inanimate object focus (Korsah and Murphy 2020, p. 845)
	a.	Hwáń1	na	Yaw pέ	*(no1)?
		who	foc	Yaw like	3sg.o
		‘Who does Yaw like?’
	b.	Déέn1	na	Yaw pέ	(*no1)?
		what	foc	Yaw like	3sg.o
		‘What does Yaw like?’

However, Korsah (2017); Korsah and Murphy (2020, 2024) argue that the gap in sentences like (6b) is actually a silent resumptive pronoun. There is a general rule in Akan that forces inanimate object pronouns to be phonetically unrealized (Christaller 1964; Osam 1996; Riis 1854), as exemplified in (7).

(7)	Pro-drop of inanimate objects (Osam 1996, p. 160)
	a.	Kofi bɛ-tɔn	[DP	dua	no	].
		Kofi fut-sell		tree	def
		‘Kofi will sell the tree.’
	b.	Kofi bɛ-tɔn	{ ____ /	*no	}.
		Kofi fut-sell		3sg.o
		‘Kofi will sell it.’

Overt inanimate object pronouns, however, do exceptionally surface in three contexts: namely, in the presence of clause-final adverbs (8a) (Saah 1994), with change-of-state verbs (8b) (Boadi 1971; Osam 1996), and with secondary predicates (8c) (Korsah 2017).

(8)	Contexts for inanimate pronoun realization (Korsah and Murphy 2020, pp. 845–47)
	a.	Kofi bɛ-tɔn *(no)  ɔkyena.
		Kofi fut-sell  3sg.o tomorrow
		‘Kofi will sell it (e.g., the tree) tomorrow.’
	b.	Kofi bu-u   *(no).
		Kofi break-pst  3sg.o
		‘Kofi broke it (e.g., the chair).’
	c.	Kuukua té  [SC *(no)  mónó ].
		Kuukua pluck  3sg.o  fresh
		‘Kuukua plucks it (e.g., the flower) fresh.’

Korsah and Murphy observe that in the same three contexts, the expected gap in the base position of a na-focused object is replaced by a resumptive pronoun. Thus, while there is a gap in the absence of a clause-final adverb like anɔpá ‘in the morning’ (9a), a resumptive pronoun is used when it is present (9b).

(9)	Resumptive pronoun with clause-final adverb (Korsah and Murphy 2020, p. 846)
	a.	[Aduane	nó]1	na	Kofí pέ	(*no1).
		food	def	foc	Kofi like	3sg.o
		‘It’s the food that Kofi likes.’
	b.	[Aduane	nó]1	na	Kofí pέ	*(no1)	anɔpá.
		food	def	foc	Kofi like	3sg.o	morning
		‘It’s the food that Kofi likes in the morning.’

Similarly, there is a gap if the object of a regular predicate is focused (10a), but there is a resumptive pronoun if it is the object of a change-of-state verb (10b).

(10)	Resumptive pronoun with change-of-state verb (Korsah and Murphy 2020, p. 846)
	a.	[Akonwa	nó]1	na	Kofi kŕá-aέ	(*no1).
		chair	def	foc	Kofi import-pst	3sg.o
		‘It’s the chair that Kofi imported.’
	b.	[Akonwa	nó]1	na	Kofi bú-u	(*no1).
		chair	def	foc	Kofi break-pst	3sg.o
		‘It’s the chair that Kofi broke.’

Again, while we observe a gap in the absence of a secondary predicate (9a), its presence requires an overt resumptive (11).

(11)	Resumptive pronoun with secondary predicate (Korsah and Murphy 2020, p. 847)
	[Aduane	nó]1	na	Kofí pέ	[SC	*(no1)	hyehyééhyé ].
	food	def	foc	Kofi like		3sg.o	very.hot
	‘It’s the food that Kofi likes very hot.’

They therefore contend that the apparent gap is the result of the inanimate pro-drop rule applying to an underlying resumptive pronoun.

2.2. Evidence for Movement in Asante Twi

Nonetheless, Korsah and Murphy (2020) (and also Korsah and Murphy 2024) show that (overt and covert) resumptives behave like gaps for the purposes of several movement diagnostics. A first one is reconstruction of the ex situ phrase into the position of the resumptive for binding. As shown in (12a), in a non-movement configuration, Asante Twi shows Principle C effects, where an R-expression must not be bound by a c-commanding pronoun. These effects pertain even when the phrase containing the offending R-expression is displaced (12b). This indicates that the relevant phrase is reconstructed into the position of the resumptive pronoun.

(12)

Reconstruction for Principle C (Korsah and Murphy 2024, p. 15)

a.

*ɔi-pɛ

[DP

Kofíi

ḿfónírí

yí

].

3sg-like

Kofi

picture

this

‘Hei likes the picture of Kofii.’

b.

*[DP

Kofíi

ḿdónírí

yí

]1

na

Ám!má

níḿ

[CP

sɛ

ɔi-pέ

no1

paa

].

Kofi

picture

this

foc

Ama

think

that

3sg-like

3sg

really

‘It’s this picture of Kofii that Ama thinks hei really likes.’

A parallel case showing reconstruction for Principle A involving the anaphor hó (cf. Saah 1989) is provided in Korsah and Murphy (2020, p. 849).

In similar fashion, a phrase that contains a variable that is bound by a quantifier can be focused with a resumptive in its base position (13), even across a clause boundary (cf. Schneider-Zioga 2009). The grammaticality of such examples suggests that the ex situ object reconstructs into the position of the resumptive (cf. Aoun et al. 2001; Sichel 2014), which in turn indicates that even long-distance focus involves syntactic movement.

(13)

Reconstruction for variable binding (Korsah and Murphy 2020, p. 850)

[DP

Nei-máńfó$\acute{\mathsf{ɔ}}$

yíe-y$\acute{\mathsf{ɔ}}$

hó

]1

na

Kofí

níḿ

[CP

sɛ

abán

bíárái

dwéné

poss-people

well-be

self

foc

Kofi

know

that

government

every

think

no1

dáá

].

3sg.o

every.day

‘It’s the well-being of itsi people that Kofi knows that everyi government thinks

about every day.’

A further diagnostic employed by Korsah and Murphy (2020) is weak crossover effects (e.g., Koopman and Sportiche 1982; Lasnik and Stowell 1991; Postal 1971). They provide the example in (14), where the focused wh-element hwáń ‘who’ cannot be coreferential with the possessive embedded in the subject despite the fact that the resumptive in the former’s base position is not c-commanded by the latter.

(14)	Weak crossover in Asante Twi (Korsah and Murphy 2020, p. 851)
	??Hwáńi	na	néi-núá	tán	nói	(nó) ?
	who	foc	poss.3sg-brother	hate	3sg.o	cd
	‘Whoi does hisi brother hate?’

Since such effects are also observed in languages like Vata, where resumptives are island-sensitive and have therefore been argued to involve movement (Koopman and Sportiche 1982), but they are absent in languages like Irish, where resumptives are taken to be base-generated and $\overline{\mathrm{A}}$-bound (McCloskey 2011); Korsah and Murphy take (14) as further evidence of the resumptive being generated by movement. However, in their recent paper (Korsah and Murphy 2024, § 3.4), they revisit the grammaticality status of (14) and, following recent arguments in Hewett (2023), relativize the relevance of cross-over effects as a diagnostic for movement in Asante Twi.

Yet another phenomenon in which resumptives behave like movement gaps is reconstruction for scope. Thus, while the existential necessarily scopes over the universal in (15a), Korsah and Murphy take the availability of a pair-list answer to (15b) to indicate that the wh-element reconstructs into the base position of the resumptive (e.g., Agüero-Bautista 2001; Panitz 2014).

(15)

Reconstruction for scope in Twi (Korsah and Murphy 2020, p. 852)

a.

Obi

ka-a

[CP

sέ

abɔfrá

bíárá

dɔ

Kofí

].

someone

say-pst

that

child

every

love

Kofi

‘Someone said that every child loves Kofi.’

(*$\forall >\exists$, $\exists >\forall$)

b.

Hwáń1

na

Kofí

ká-a

[CP

sέ

abɔfrá

bíárá

dɔ

no1

]?

who

foc

Kofi

say-pst

that

child

every

love

3sg.o

‘Who did Kofi say that every child loves?’

($\forall >\mathit{wh}$, $\mathit{wh}>\forall$)

A final and pivotal observation of their paper is that there is a tonal reflex associated with successive-cyclic $\overline{\mathrm{A}}$-movement in which lexical low tones on all verbs crossed by the dependency are overwritten with high tones. This process of high tone overwriting is active in the na-focus construction independent of whether there is an apparent gap, i.e., covert resumptive (16), or an overt resumptive present in the base position (17).

(16)	Tonal overwriting in focus construction with gap (Korsah and Murphy 2020, p. 834)
	a.	Kofí	kaé	sɛ	Ám!má	kita	bayérέ.
		Kofi	remember	that	Ama	hold	yam
		‘Kofi remembers that Ama is holding a yam.’
	b.	Déέn1	na	Kofí	káé	sɛ	Ám!má	kítá ――1?
		what	foc	Kofi	remember	that	Ama	hold
		‘What does Kofi remember that Ama is holding?’
(17)	Tonal overwriting in focus construction with resumptive (Korsah and Murphy 2020, p. 834)
	a.	Kwame	nim	sέ	Ám!má	hu-u	Efua.
		Kwame	knows	that	Ama	see-pst	Efua
		‘Kwame knows that Ama saw Efua.’
	b.	Hwáń1	na	Kwame	níḿ	sɛ	Ám!má	hú-u	no1?
		who	foc	Kwame	knows	that	Ama	see-pst	3sg.o
		‘Who does Kwame know that Ama saw?’

As they point out, no such tonal effect is observed in the parallel topic construction, where the topic phrase appears clause-initially before a topic marker deɛ, while a resumptive occurs in the regular argument position (18b). Importantly, this construction also does not pass other movement diagnostics.5

(18)	No tonal reflex in topic construction (Marfo 2005, p. 110)
	a.	Á!má1	na	Kofi ré-bóá	nó1.
		Ama	foc	Kofi prog-help	3sg.o
		‘It is Ama who Kofi is helping.’
	b.	Á!má1	déέ	Kofi re-boá	/ *ré-bóá	nó1.
		Ama	top	Kofi prog-help	prog-help	3sg.o
		‘As for Ama, Kofi is helping her.’

Based on this evidence, Korsah and Murphy (2020) conclude, and Korsah and Murphy (2024) reaffirm, that resumptives in Asante Twi can appear at the bottom of an $\overline{\mathrm{A}}$-movement dependency, which directly explains their behaviour with respect to the abovementioned diagnostics (see Alexopoulou 2006; Aoun et al. 2001; Boeckx 2003; Borer 1984; Engdahl 1985; Georgi and Amaechi 2023; Klein 2017; Koopman 1984; Pesetsky 1998; Salzmann 2017; Scott 2021; Sichel 2014; Zaenen et al. 1981 and others on resumptives with $\overline{\mathrm{A}}$-movement).

2.3. Selective Island-Sensitivity and Repair by Resumption in Asante Twi

There is one obvious exception to this pattern: both overt and covert resumption seems to be insensitive to islands, as shown on the basis of the complex NP island (19a) and (20a), and the wh-island (19b) and (20b).

(19)

Island insensitivity with resumptives (Saah 1994, p. 172; Korsah 2017, p. 117)

a.

Hwáń1

na

wo-hú-u

[DP

onipa

ko

[CP

áa

ɔ-b$\acute{\mathsf{ɔ}}$-ɔ

nó1

nó ]]?

who

foc

2sg.s-see-pst

person

def

rel

3sg.s-hit-pst

3sg.o

cd

‘Who did you see the person who hit?’

Á!má1

na

Kofí

bísá-a

[CP

sέ

hwán2

na

ɛ2-d$\acute{\mathsf{ɔ}}$

nó1

nó ].

Ama

foc

Kofi ask-pst

that

who

foc

3sg.s-love

3sg.o

cd

‘It is Ama who Kofi asked who loves.’

(20)

Island insensitivity with gaps (Saah 1994, pp. 172, 197)

a.

Déέn

na

wo-níḿ

[DP

onipa

ko

[CP

áa

ɔ-t$\acute{\mathsf{ɔ}}$-ɔ-έ

――1

nó ]]?

what

foc

2sg.s-know

person

def

rel

3sg.s-buy-pst-ye

cd

‘What do you know the person that bought?’

b.

Deɛn1

na

Mary

bisa-a

[CP

sέ

hwán

na

ɔ-yɛ-e

――1

nó ].

what

foc

Mary

ask-pst

that

who

foc

3sg.s-make-pst

cd

‘What did Mary ask who made?’

This seems to contradict the view that the na-focus construction involves movement. However, Korsah and Murphy (2020) draw attention to the fact that island effects emerge when the extracted phrase has a different category other than DP or NP and therefore lacks a resumptive pronoun. Thus, while focus of a VP is generally possible in Asante Twi (21a), it is ungrammatical if the VP is focused from inside an island configuration (21b) (see also Hein 2017, 2020). Note that a resumptive is impossible in these sentences.

(21)

Island effects emerge with VP focus (Hein 2017, p. 38)

a.

[VP

Dán

sí]-é

na

Ámá

káa

sέ

Kofí

á-y$\acute{\mathsf{ɔ}}$

{――VP /

*nó}

house

build-nmlz

foc

Ama

say.pst

that

Kofi

pfv-do

3sg.o

anɔpá.

morning

‘Ama said that Kofi built a house in the morning (not bought a car).’

b.

*[VP

Dán

sí]-é

na

mé-n-tée

[DP

atétésέm

bíárá

[CP

sέ

house

build-nmlz

foc

1sg.s-neg-hear.pst

rumour.pl

any

that

Kofí

á-y$\acute{\mathsf{ɔ}}$

――VP ]].

Kofi

pfv-do

‘I didn’t hear any rumours that Kofi has built a house.’

Likewise, focalization of a PP, while possible in the language (22a), is ungrammatical from inside an island configuration (22b).

(22)

Island effect emerges with PP focus (Korsah and Murphy 2020, p. 847f)

a.

[PP

Akonwá

nó

mú ]

na

Kofí

dá

{——PP /

*hɔ}

anɔpá.

chair

the

in

foc

Kofi

lie

there

morning

‘Kofi is lying in the chair in the morning.’

b.

*[PP

Akonwá

nó

mú ]

na

Ama

níḿ

[DP

neá

ńtí

[CP

áa

Kofi dá

chair

the

in

foc

Ama

know

thing

because.of

rel

Kofi lie

――PP ]].

‘Ama knows the reason why Kofi lies in the chair.’

Moreover, nominal focus from inside an island behaves just like nominal focus from non-island configurations with respect to some of the movement diagnostics mentioned above (cf. the discussion of movement properties with resumptives in islands in Korsah and Murphy 2024). Thus, there is reconstruction for Principle C into a complex NP island as shown in (23) and reconstruction for variable binding into a complex NP island as shown in (24), which is the island version of (13).

(23)

Reconstruction for Principle C into complex NP island (Korsah and Murphy 2024, p. 15)

[DP

Ám!mái

adúané

nó

]1

na

m-á-té

[DP

atetésέm

bí

[CP

sέ

Ama

food

def

foc

1sg.s-perf-hear

rumour

indef

that

$\acute{\mathsf{ɔ}}$*i/j-dí

no1

dáá

]]

3sg.s-eat

3sg.o

every.day

‘It’s Ama’si food that I have heard a rumour that s/he*i/j eats every day.’

(24)

Reconstruction for variable binding into complex NP island (Korsah and Murphy 2020, p. 859)

[DP

Nei-máńfó$\acute{\mathsf{ɔ}}$

yíe-y$\acute{\mathsf{ɔ}}$

hó

]1

na

m-á-té

[DP

atésέḿ

bí

[CP

3sg.poss-people

well-be

self

foc

1sg.s-perf-hear

rumour

indef

sɛ

abán

bíárái

dwéné

no1

dáá

]].

that

government

every

think

3sg.o

every.day

‘It’s the well-being of itsi people that I have heard a rumour that everyi government thinks about every day.’

In addition, nominal focus from inside a complex NP island exhibits weak crossover effects, as shown in (25), which is the island version of (14). Though keep in mind that this diagnostic has its problems (cf. Korsah and Murphy 2024, § 3.4).

(25)	Weak crossover effects with focus from complex NP island (Korsah and Murphy 2020, p. 859)
	??Hwáńi	na	wo-á-té	[DP	atésέḿ	bí	[CP	sɛ	néi-núá	tán
	who	foc	2sg.s-perf-hear		rumour	indef		that	3sg.poss-sibling	hate
	nói	nó	]]?
	3sg.o	cd
	‘Whoi have you heard the rumour that hisi/heri sibling hates (himi/heri)?’

Last, as with focus from non-island configurations (17), high tone overwriting is observed also in nominal focus from inside an island. This is shown for the complex NP island in (26b).

(26)

Tonal movement reflex with focus from inside a complex NP island (Korsah and Murphy 2024, p. 13)

a.

Me-hu-u

[DP

onipa

ko

[CP

áa

ɔ-bɔ-ɔ

Kofí

nó

]].

1sg.s-see-pst

person

def

rel

3sg.s-hit-pst

Kofi

cd

‘I saw the person who hit Kofi.’

b.

Hwáń1

na

wo-hú-u

[DP

onípá

ko

[CP

áa

ɔ-b$\acute{\mathsf{ɔ}}$-ɔ

nó1

nó

]]?

who

foc

2sg-see-pst

person

def

rel

3sg.s-hit-pst

3sg.o

cd

‘Who did you see the person who hit?’

Taken together, as Korsah and Murphy (2020, 2024) point out, the evidence suggests that the na-focus construction in Asante Twi involves $\overline{\mathrm{A}}$-movement. An account in terms of base-generation (Saah 1994) seems unfit.

What this leaves to explain is why nominal focus can circumvent island violations while verbal and PP focus cannot. Recall that the gap that is observed when an inanimate nominal is focused is only superficial. It is the result of a general PF rule enforcing the zero spellout of inanimate object pronouns that applies to an underlying resumptive pronoun. This means that both animate and inanimate object focus constructions have a resumptive pronoun present in the base position of the object in the syntax. It has long been observed that resumptive pronouns may alleviate island violations (Ackerman et al. 2018; Aoun and Choueiri 2000; Borer 1984; Chomsky 1986; Keupdjio 2020; Kroch 1981; McCloskey 1990; McDaniel and Cowart 1999). As Korsah and Murphy (2020, 2024) argue, this is also the case in Asante Twi. It is the existence of a resumptive pronoun for nominal elements that allows the language to circumvent island violations when undergoing focalization. In turn, the absence of such a resumptive pronoun for non-nominal elements like V(P)s or PPs prevents them from being focused out of an island configuration.

In particular, Korsah and Murphy (2020) follow a strand of research that goes back to Perlmutter (1972) and takes the perspective that islands are representational constraints at PF that militate against certain linguistic representations that have a trace/gap in the base position (see Boeckx 2012; Griffiths and Lipták 2014; Hornstein et al. 2007; Lasnik 2001; Merchant 2001; Pesetsky 1998). These constraints can be satisfied by avoiding the offending representation. One way to achieve this is by replacing the trace/gap with a resumptive pronoun.6 They argue that this immediately explains the island insensitivity of animate object focus (with overt resumption) if this replacement takes place prior to the evaluation of island constraints. The island-obviating effect of the covert resumptives with inanimate objects is then accounted for by PF-ordering such that the obligatory PF pro-drop rule for inanimate pronouns only applies after the island constraints have been checked. The required order of application between the relevant operations is given in (27).

(27)	Order of application of PF-operations in Asante Twi
	Trace-to-pronoun conversion ≺ island evaluation ≺ inanimate pro-drop

At least the examples with a surface gap can then be classified as Phillips (2013a, 2013b) surface variation islands because while it seems on the surface that extraction has left a gap inside the island configuration, there actually is a resumptive pronoun whose presence at the time when the island constraint was checked voided a violation.

2.4. Island-Insensitivity without Resumption in Asante Twi

While the PF-deletion plus order-of-application account nicely explains the discussed island insensitivities, it does not straightforwardly extend to some recent observations that certain nominal elements can escape islands despite not leaving a resumptive pronoun. In a recent paper, Hein and Georgi (2021) note that almost all of the sentences used in the discussion of resumptive pronouns, gaps, and their interaction with island contexts involve extraction of wh-elements, proper names, definite noun phrases with an overt determiner, or bare nouns that are interpreted as specific indefinites. They show that several other noun types do not fit the pattern described in the previous section, where focused animate nouns leave an overt resumptive pronoun and inanimate ones result in a surface gap that arises by PF-deletion of an underlying resumptive pronoun. Rather, extraction of these noun types always results in a gap even if they are animate, in which case the deletion rule should be inapplicable, or inanimate and appearing in the deletion-suspending contexts.

First, nominal objects that are part of an idiom never show resumption. Recall that the PF rule that deletes inanimate (resumptive) pronouns in object positions can be suspended in the presence of a clause final adjunct (see (8) and (9)). In this case, the underlying resumptive pronoun surfaces. Therefore, we would expect an overt resumptive pronoun to appear in the object position of (28b). However, on the idiomatic reading only a gap is possible, whereas the presence of an overt resumptive pronoun forces the literal reading (28c).

(28)	No resumption with idiomatic nouns in object position (Korsah and Murphy 2020, p. 855; Hein and Georgi 2021, p. 226)
	a.	ɔ-gya-a	ne-nán	[PP	wɔ	dán	nó	mú	].
		3sg.s-leave-pst	3sg.poss-leg		loc	room	def	inside
		Idiomatic: ‘He defecated in the room.’
		Literal: ‘He left his leg in the room.’
	b.	Ne-nán1	na	ɔ-gyáɛ	――1 [PP	wɔ	dán	nó	mú	].
		his-leg	foc	3sg.s-leave.pst		loc	room	the	inside
		Idiomatic: ‘It’s defecating that he did in the room.’
		#Literal: ‘It’s his leg that he left in the room.’
	c.	Ne-nán1	na	ɔ-gyáɛ	no1	[PP	wɔ	dán	nó	mú	].
		his-leg	foc	3sg.s-leave.pst	3sg.o		loc	room	the	inside
		#Idiomatic: ‘It’s defecating that he did in the room.’
		Literal: ‘It’s his leg that he left in the room.’

A further example using the idiom bɔ n’ano twɛdeɛ ‘to eat’ (literally ‘to punch one’s own mouth’) is given in (29).

(29)	No resumption with idiomatic nouns in object position (Hein and Georgi 2021, p. 226f.)7
	Daabi.	[N’ano	twɛdeɛ]1	na	Kofi bɔ-ɔ	{――1 /	*no1 }	ɛnora.
	no	3sg.poss-mouth	own	foc	Kofi hit-pst		3sg.o	yesterday
	Idiomatic: ‘No. Kofi ate yesterday’ (He did not punch the dog.)
	#Literal: ‘No. He punched his own mouth yesterday.’

Hein and Georgi (2021) take this to show that idiomatic noun phrases do not leave a resumptive pronoun when they undergo focus movement. A further noun type that behaves like this is predicative nominals, such as tíkyani ‘teacher’ in the following example.

(30)	No resumption with nominals in predicative use (Hein and Georgi 2021, p. 227)
	Tíkyani1	na	Kofi bέ-yέ	{――1 /	*nó1}	afe	yí.
	teacher	foc	Kofi fut-be		3sg.o	year	this
	‘It is a teacher that Kofi will become this year.’

A resumptive pronoun in (30) is ungrammatical despite the fact that the extracted nominal is animate and there is a clause-final adverb. This strongly indicates that the gap is a true syntactic one.

A last type of noun that Hein and Georgi (2021) report to be incompatible with resumptive pronouns are non-specific indefinites. In the context provided for the elicitation of (31), the bare noun ķyerɛkyerɛni ‘teacher’ is referring to some random teacher at a school. Although the sentence contains a final adverb kane ‘first’ and the extractee is animate, a resumptive pronoun is strongly dispreferred.

(31)	No resumption with non-specific indefinites (Hein and Georgi 2021, p. 227)
	Daabi.	ɔkyerɛkyerɛni1	na	me-be-bisa	{――1 /	??no1 }	kane.
	no	teacher	foc	1sg.s-fut-ask		3sg.o	first
	‘No. I will ask a (random) teacher first.’ (one of the many teachers around)

These three noun types—idiom parts, predicative nouns, and non-specific indefinites—therefore arguably do not leave a resumptive pronoun in their base position upon extraction at any point in the derivation. Given the crucial role that syntactic resumptives play for Korsah and Murphy’s (2020) explanation of the island-insensitivity of nominal object focus, this leads to the prediction that focalization of these noun types should be sensitive to islands just like V(P) and PP focus is. However, this prediction is not borne out. Like other nominals, these three noun types can freely be extracted from island configurations even though they do not show any indication of resumption. This is illustrated in (32), where the nominal object in each case undergoes focus movement from inside a complex NP island, leaving a true gap. As was the case in the non-island extractions above, a resumptive pronoun is impossible even in the presence of a final adjunct.

(32)	Island-insensitivity of nominal extraction without resumption (Hein and Georgi 2021)
	a.	Ne-nán1	na	m-á-té	[DP	atésέm	bí	[CP	sέ	ɔ-gyáɛ	{――1 /
		his-leg	foc	1sg.s-pfv-hear		rumour	a		that	3sg.s-leave.pst
		*nó1}	wɔ	dán	nó	mú	]].
		3sg.o	loc	room	the	inside
		‘It’s defecating that I have heard a rumour that he did in the room.’
	b.	Tíkya1	na	m-á-té	[DP	atésέm	nó	[CP	sέ	Kofi bέ-yέ	{――1 /
		teacher	foc	1sg-perf-hear		rumour	the		that	Kofi fut-be
		*nó}	afe	yí	]].
		3sg.o	year	this
		‘It is a teacher that I have heard the rumour that Kofi will become this year.’
	c.	Nnípa1	na	wo-té-e	[DP	atésέm	nó	[CP	sέ	Kofi súró	{――1 /
		pl.person	foc	2sg.s-hear-pst		rumour	the		that	Kofi fear
		*nó1	/	*wɔn }	páa	]].
		3sg.o		3pl.o	really
		‘It’s people that I have heard the rumour that Kofi really fears.’

This shows that nominals that leave a resumptive upon extraction are only a subset of the nominals that can be extracted from inside an island configuration (33). Therefore, whether a given constituent can or cannot undergo $\overline{\mathrm{A}}$-movement from inside an island cannot depend upon whether or not it shows a resumptive in the base of the dependency (pace Korsah and Murphy 2020).

(33)	Distribution of gaps vs. RPs and island-sensitivity (Hein and Georgi 2021, p. 234)
		[+N]regular	[+N]special	VP/PP
	(c)overt RP	yes	no	no
	island-sensitive	no	no	yes

It rather seems to be the case that the category of the extractee plays a direct role in its extractability and that the latter is independent of the issue of whether the extractee can leave a resumptive pronoun or not.

Concerning resumption, Hein and Georgi (2021) propose a partial copy deletion account (Landau 2006; Pesetsky 1998; van Urk 2018) that deletes the maximal projection of the lexical core of a moved phrase’s lowest copy. This leads to full deletion and therefore a gap in case the moved phrase is a VP, PP, or—as Hein and Georgi (2021) argue is the underlying structure for the problematic non-referential noun types—an NP. For a DP, only its NP-complement is deleted, and the stranded D-head is realized as a resumptive pronoun (cf. Elbourne 2001; Jenks and Bi 2019; Postal 1969). The lack of a D-layer of the non-resumptive noun types thus is the reason for their showing no resumption upon extraction. When it comes to island-sensitivity, Hein and Georgi (2021) suggest that islands are selectively transparent for $\overline{\mathrm{A}}$-movement of elements with a ‘nominal’ core but are opaque for those without one, thereby deriving the split in island-sensitivity between DPs and NPs with such a core on the one side and VPs and PPs without such a core on the other.

In response to the empirical challenge posed by apparently non-resumptive noun types, (Korsah and Murphy 2024, pp. 27–28) tentatively argue that there is a more general ban on overt pronouns that are co-indexed with such nouns. Some evidence for this comes from the obligatory absence of overt pronouns in anaphoric contexts when they refer back to predicative (34a) and non-specific indefinite (34b) though unambiguously animate nouns.

(34)	No overt pronouns with predicate and non-specific indefinite nominals (Korsah and Murphy 2024, p. 27)
	a.	Na	Kwadwo	pɛ	sɛ	ɔ-yɛ	odusinii	nanso	ɔ-a-n-yɛ.
		pst	Kwadwo	want	comp	3sg.s-fut.be	herbalist	but	3sg.s-perf-neg-be
		?bii	/	*noi	/ ――i.
		indef		3sg.o
		‘Kwadwo wanted to become a herbalist, but he did not become (one).’
	b.	Nípai	na	Kofi súró ――i	[CP	ésánesέ	ɔ-fέré	{ ――i /	*nói	} ]
		person	foc	Kofi fear		because	3sg.s-be.shy.of		3sg.o
		‘It’s people that Kofi really fears because he is shy of (them).’

They suggest that this might be accounted for by a second obligatory pro-drop rule (in addition to the one for inanimate object pronouns) that applies to underlyingly present pronouns for these noun types and, unlike the latter, is not suspended in the presence of a clause final adverb. As this rule arguably extends to resumptive pronouns, Korsah and Murphy (2024) conclude that the pattern observed by Hein and Georgi (2021) can presumably be reconciled with their approach. The underlyingly present resumptive pronoun for a non-referential noun is only deleted after the evaluation of island constraints at PF, thereby enabling an $\overline{\mathrm{A}}$-movement dependency into an island that terminates in a surface gap.

It therefore seems that the empirical profile of selective island sensitivity in Asante Twi can be accounted for by treating islands as representational PF constraints that interact with the distribution of resumptive pronouns and pronoun deletion operations (Korsah and Murphy 2024), as well as by viewing them as directly category-sensitive locality domains (Hein and Georgi 2021).

From a more conceptual perspective, both approaches have their separate issues. While the approach in terms of PF-islands is reasonably well worked out, it suffers from the more general question of why these PF constraints are as they are. In other words, why do they prohibit a configuration with a gap but not one with a resumptive pronoun? As a reviewer points out, there is a priori nothing wrong if they happen to be the reverse, militating against overt resumptives in island configurations. As for the approach that takes islands to be directly sensitive to the category, the question is why exactly they are permeable to only nominal elements. In addition, there is not yet a properly formulated account of how these and only these elements can escape from them. While I cannot provide such an implementation for Asante Twi, in what follows, I will present a very similar pattern of island-sensitivity in the Grassfields Bantu language Limbum. For this pattern, an approach in terms of PF-islands that are circumvented by underlying resumptive pronouns is far less well-supported than for Asante Twi. I will tentatively suggest that the selectivity of islands in Limbum is tied to $\varphi$-features and will provide an implementation of this in the grammar.

3. Escaping Islands in Limbum

3.1. The Focus Construction in Limbum

A very similar case of apparent category-sensitivity of islands can be observed in the á-focus construction in the understudied Grassfields Bantu language Limbum (Fransen 1995), which is spoken primarily in the northwest region of Cameroon. The data reported here, if not marked otherwise, stem from a number of elicitation sessions with one native speaker from Nkambe, Cameroon, over a period of several months between August 2018 and May 2019 and some additional elicitation work with the same speaker in August 2023.

The general word order of a neutral declarative clause in Limbum is SVO. Any tense and aspect markers precede the verb. Adverbs including the negation marker kàɁ are restricted to the clause-final position. An example of such a clause is given in (35).

(35)	Limbum neutral declarative clause
	Njíŋwὲ	f$\overline{\mathsf{ɔ}}$	à	mū	yε̄	bō	f$\overline{\mathsf{ɔ}}$	nìŋkòr.
	woman	det	sm	pst2	see	children	det	yesterday
	‘The woman saw the children yesterday.’

The verb commonly appears in root form without any agreement affixes. However, in certain tenses and aspects, a subject marker occurs before the TAM-morpheme that gives the impression of subject agreement (see Hein 2021 for a more detailed description of subject marking).

While (new information) focus is possible in situ (36), the language also comprises a strategy that includes displacement of the focused constituent to the sentence-initial position, where it is preceded by the focus marker á and (optionally) followed by a particle cí (37).8 As briefly mentioned before, unlike in Asante Twi, there is generally no resumption for displaced objects (with the exception of optional resumption for specific bare nominals, as will be discussed in Section 3.4).

(36)	Limbum in situ object focus (Driemel and Nformi 2018, p. 18)
	A:	Wὲ	bí	k$\overline{\mathsf{ɔ}}$nī	ndá?
		2sg	fut1	meet	who
		‘Who will you meet?’
	B:	Mὲ	bí	k$\overline{\mathsf{ɔ}}$nī	Ngàlá.
		1sg	fut1	meet	Ngala
		‘I will meet Ngala.’
(37)	Limbum ex situ object focus (Driemel and Nformi 2018, p. 18)
	A:	Á	ndá1	(cí)	wὲ	bí	k$\overline{\mathsf{ɔ}}$nī ――1	à?
		foc	who	ci	2sg	fut1	meet	q
		‘Who will you meet?’
	B:	Á	Ngàlá1	(cí)	mὲ	bí	k$\overline{\mathsf{ɔ}}$nī ――1.
		foc	Ngala	ci	1sg	fut1	meet
		‘I will meet Ngala.’

The fact that the marker á precedes the focused constituent in conjunction with there being a following element may seem to suggest that the construction is a cleft, i.e., a biclausal structure where á is the copula ‘it is’ and cí is the relative marker ‘that’. Indeed, in her grammar of Limbum, one of the few published works on the language, Fransen (1995) analyses the á-focus construction as a cleft. However, the speaker consulted here generally disagreed with several of the data in this grammar, pointing out either that they had an archaic appeal sounding somewhat old-fashioned or that they might belong to a different variety of the language. Furthermore, Becker et al. (2019) present several arguments for the monoclausality of this construction that are based on the judgments of the same speaker as the ones presented in this paper. For details, the reader is referred to their work. Here, I merely want to briefly add to their argument about the cí-marker not being a relative pronoun. This is corroborated by data showing that the relative pronoun covaries with the head noun (at least) in number, taking the form vǐ in the plural (38a), whereas the element cí in the focus construction is invariant (38b).

(38)

Relative pronoun but not cí covaries with head noun

a.

ŋwὲ

f$\overline{\mathsf{ɔ}}$

rìŋ

bōmbáŋrò

f$\overline{\mathsf{ɔ}}$

{

vǐ

/

*cí

}

njíŋwὲ

f$\overline{\mathsf{ɔ}}$

à

mū

yε̄

man

det

know

boys

det

rel.pl

ci

woman

det

sm

pst2

see

nìŋkòr.

yesterday

‘The man knows the boys whom the woman saw yesterday.’

b.

Á

bōmbáŋrò

f$\overline{\mathsf{ɔ}}$

{

cí

/

*vǐ

}

njíŋwὲ

f$\overline{\mathsf{ɔ}}$

à

mū

yε̄

nìŋkòr.

foc

boys

det

ci

rel.pl

woman

det

sm

pst2

see

yesterday

‘The woman saw the boys yesterday.’

Thus, despite its appearance, the á-focus construction does not show typical properties of a cleft. I will therefore adopt the view that the construction involves a monoclausal structure in which the focused constituent is placed in the left periphery instead of its base position. In what follows, I will argue that it has indeed moved to this surface position rather than being base-generated there.

3.2. Evidence for Movement in Limbum

Like in Asante Twi, there are several movement diagnostics that indicate that the focused constituent originates inside the clause. First of all, the á-focus construction may be used to focus embedded objects: that is, it may apply across a finite clause boundary (39), which is one hallmark of $\overline{\mathrm{A}}$-dependencies.9 While this might not in itself be a sufficient condition for movement, as there are other long-distance dependencies that are not created by movement, it surely is a necessary one.

(39)

Long-distance focus

Á

[DP

njíŋwὲ

f$\overline{\mathsf{ɔ}}$]1

(cí)

mὲ

rìŋ

[CP

mὲ-nε̄

Nfor

bí

k$\overline{\mathsf{ɔ}}$nī ――1 ].

foc

woman

det

cí

1sg

know

1sg-comp

Nfor

fut1

meet

‘I know that Nfor will meet the woman.’

Second, we observe strong cross-over effects, i.e., a focused nominal is reconstructed for Condition C (40).10

(40)	Reconstruction for Condition C
	a.	Í*i/j	rìŋ	í-nε̄	à	cí	ró	Nfòri.
		3sg.s	know	3sg-comp	2sg	prog	search	Nfor
		‘He*i/j knows that you are searching for Nfori.’
	b.	Á	Nfòri	(cí)	í*i/j	rìŋ	í-nε̄	wὲ	cí	ró	――i.
		foc	Nfor	ci	3sg.s	know	3sg-comp	2sg	prog	search
		‘He*i/j knows that you are searching for Nfori.’

A third argument in favour of movement comes from reconstruction for variable binding. In (41a), the variable associated with the possessive marker zhì ‘his’ can be optionally bound in its base position by a c-commanding quantified noun phrase. This binding is unaffected by focus-fronting (41b), which indicates that the focused phrase that includes the possessive reconstructs into its base position.

(41)

Reconstruction for variable binding

a.

[ŋwὲ

nsìp]x

bí

k$\overline{\mathsf{ɔ}}$nī

táā

zhìx/y

ŋgwá.

man

every

fut1

meet

father

3sg.poss

wife

‘Every manx will meet the father of hisx/y wife.’

b.

Á

[táā

ŋgwá

zhìx/y]1

cí

mὲ

kwàɁshí

mὲ-nε̄

[ŋwὲ

nsìp]x

bí

foc

father

wife

3sg.poss

ci

1sg

think

1sg-comp

man

every

fut1

k$\overline{\mathsf{ɔ}}$nī ――1.

meet

‘I think that every manx will meet the father of hisx/y wife.’

In addition, we find that there is reconstruction for relative quantifier scope. In the sentence in (42), the reading where the lower universal njíŋwὲ nsìp ‘every woman’ outscopes the higher existential mbàŋrù ‘male’ is available. This indicates that reconstruction of the existential into the base position of the object is possible. If it were not, one would expect the sentence to be judged odd, which it is not. Example (43) shows that quantifier raising is clause-bound in Limbum (cf. Larson and May 1990; May 1985) and, thus, that the reading in (42) cannot be derived by raising the universal from the embedded clause across the existential in the matrix clause at LF.11

(42)

Reconstruction for scope

Á

mbàŋrù1

(cí)

Shey

à

mū

lā

[CP

í-nε̄

njíŋwὲ

nsìp

à

mū

bzú

foc

male

ci

Shey

sm

pst2

say

3sg-comp

woman

every

sm

pst2

birth

――1 ].

‘Shey said that every woman gave birth to a son.’    ($\forall >\exists$, *$\exists >\forall$)

(43)

Quantifier raising is clause-bound

ŋwὲ-ryēɁnì

à

mū

la

[CP

í-nε̄

mū

nsìp

à

mū

cí

bumi

].

man-teach

sm

pst2

say

3sg-comp

child

every

sm

pst2

prog

sleep

‘A/Some teacher said that every child was sleeping.’    ($\exists >\forall$, *$\forall >\exists$)

It has, however, been argued that reconstruction effects may also hold in base-generation dependencies (Guilliot and Malkawi 2009; Moulton 2013; Rouveret 2008) and therefore cannot constitute unequivocally decisive evidence for $\overline{\mathrm{A}}$-movement. What is argued by the same authors is that a distinction between base-generation reconstruction and movement reconstruction is still possible. While the latter shows what is termed ‘reconstruction conflicts’, the former does not. A reconstruction conflict arises in clauses where the reconstructing XP is subject to two conflicting requirements. Consider the example in (44). The targeted reading is one where the quantifier every student binds the coindexed pronoun he and the R-expression Bresnan is coreferential with the pronoun she. In order to achieve this reading, the extracted XP must be able to reconstruct into a position below the quantifier but above the pronoun: that is, it must reconstruct into an intermediate position (Fox 1999; Lebeaux 1991; Legate 2003; Sauerland 2003). The grammaticality of (44a) attests that this is possible. In contrast, in (44b), there is no position in which both requirements can be satisfied simultaneously. The sentence (with the intended reading) is therefore ungrammatical.

(44)	Reconstruction conflict in English (Lebeaux 1991)
	a.	[DP Which paper that hex gave to Bresnany ]1 did every studentx think [CP ${\mathrm{t}}_1^{\prime }$ that shey would like t1 ]?
	b.	*[DP Which paper that hex gave to Bresnany ]1 did shey think [CP ${\mathrm{t}}_1^{\prime }$ that every studentx would like t1 ]?

Importantly, with reconstruction in base-generation dependencies, as is observable in French and Welsh resumptions, this conflict is absent. That is, whether the necessary intermediate position exists or not has no influence on the grammaticality of the clauses. Rather, either both sentences are acceptable, as in French (45), or both are unacceptable, as in Welsh (46).

(45)

No reconstruction conflict in French (Panitz 2014, p. 18)

a.

[DP

Quel

résumé

qu’ílx

a

donné

à

Hamiday ]1

est-ce

que

chaque

which

abstract

that.he

has

given

to

Hamida

q

that

each

étudiantx

se

demande

[CP t1

si

elley

va

le1

corriger

]?

student

refl

asks

if

she

goes

it

amend

‘Which abstract that hex gave to Hamiday does each studentx wonder if shey will amend it.’

b.

[DP

Quel

résumé

qu’ílx

a

donné

à

Hamiday ]1

est-ce

qu’elley

se

which

abstract

that.he

has

given

to

Hamida

q

that.she

refl

demande

[CP

${\mathrm{t}}_1^{\prime }$

si

chaque

étudiantx

va

le1

réviser

]?

asks

if

each

student

goes

it

revise

‘Which abstract that hex gave to Hamiday does shey wonder if each studentx will revise it.’

(46)

No reconstruction conflict in Welsh (Rouveret 2008)

a.

*[DP

barn

yr

athro

ar

[

eix

mab]y ]1

y

gŵyr

pob

mamx

[CP

${\mathrm{t}}_1^{\prime }$

y

mae

opinion

the

teacher

on

her

son

ci

knows

each

mother

ci

is

efy

yn

ei1

chuddio

]

he

prog

it

conceal

‘the teacher’s opinion of [herx son]y that each motherx knows that hey conceals’

b.

*[DP

barn

yr

athro

ar

[

eix

mab]y ]1

y

gŵyr

efy

[CP

t1

y

mae

pob

opinion

the

teacher

on

her

son

ci

knows

he

ci

is

each

mamx

yn

ei1

pharchu

]

mother

prog

it

respect

‘the teacher’s opinion of [herx son]y that hey knows that each motherx respects’

While the occurrence of a gap in the Limbum focus examples suggests an analysis in terms of movement rather than base-generation, one might still propose that the gap is actually a silent resumptive pronoun in a base-generation dependency (cf. Borer 1984; Chomsky 1977; Cinque 1990). However, reconstruction conflicts, as in the English example (44), also occur in Limbum. Consider the pair of sentences in (47).

(47)

Reconstruction conflict in Limbum

a.

Á

[DP

ŋkār

[

bō

bvix]y

]1

(cí)

njíŋwὲ

nsipx

kwàɁshī

[CP

${\mathrm{t}}_1^{\prime }$

foc

friend

children

3sg.poss

ci

woman

every

think

í-nε̄

óy

mū

cèb

t1

].

3sg-comp

3pl.sm

pst2

insult

‘(It’s) a friend of [herx children]y (that) every womanx thinks that theyy insulted.’

b.

*Á

[DP

ŋkār

[

bō

bvix]y

]1

(cí)

óy

kwàɁshī

[CP

${\mathrm{t}}_1^{\prime }$

ó-nε̄

foc

friend

children

3sg.poss

ci

3pl.sm

think

3pl-comp

njíŋwὲ

nsipx

à

mū

cèb

t1

].

woman

every

sm

pst2

insult

‘(It’s) a friend of [herx children]y (that) theyy think that every womanx insulted.’

In (47a), the embedded object ŋkār bō bvi ‘friend of her children’ has undergone focus-fronting. It contains a bound variable anaphor bvi ‘her’, which in the targeted reading is bound and therefore should be c-commanded by the universally quantified matrix subject njíŋwὲ nsìp ‘every woman’. At the same time, the embedded pronominal subject ó ‘they’ is coreferential with the fronted object and therefore should not c-command it in order to avoid a Principle C violation. There is thus only one position in which both requirements are met: namely, the intermediate position ${\mathrm{t}}_1^{\prime }$ in the left edge of the embedded clause. Just like in English, the corresponding sentence with the matrix and embedded subjects swapped is ungrammatical (47b) because neither in ${\mathrm{t}}_1^{\prime }$ nor in t₁ can both requirements be met at the same time. That (47a) is grammatical indicates that the object reconstructs into the intermediate position. In turn, this provides a strong argument in favour of (successive-cyclic) $\overline{\mathrm{A}}$-movement being involved in the á-focus construction.

Moreover, that the dependency involves $\overline{\mathrm{A}}$-movement is also suggested by a diagnostic that does not rely on reconstruction. We observe superiority effects when both arguments of a transitive verb are questioned. In such a case, only the higher wh-element may undergo fronting (48a). If the lower wh-element is fronted across the higher one, the sentence is severely degraded (48b).

(48)	Superiority effect in Limbum
	a.	Á	ndá1	(cí)	í1	bí	yū	kέέ	à?
		foc	who	ci	3sg.s	fut1	buy	what	q
		‘Who will buy what?’
	b.	??Á	kέέ1	(cí)	ndá	bí	yū	――1	à?
		foc	what	ci	who	fut1	buy		q
		‘What will who buy?’

This receives a straightforward explanation if focus-fronting involves movement, because movement is subject to the Minimal Link Condition (Chomsky 1995) requiring the closest wh-item to move, which in this case is the subject. In order to obtain the same superiority effect under the view that focus involves base-generation, an additional constraint on order preservation would have to be postulated. Note that in Limbum, displacement of the subject obligatorily requires a resumptive pronoun to appear in the regular subject position.

A last piece of evidence for $\overline{\mathrm{A}}$-movement is the observation that focus-fronting licenses parasitic gaps (cf. Bresnan 1977; Nissenbaum 2000; Ross 1967; Taraldsen 1981), which are commonly taken to diagnose movement (Culicover 2001; Engdahl 1983). Thus, focalisation of the object of the matrix clause as either a question (50a) or new information focus (50b) licenses a parasitic gap in the object position of the relative clause.12 The sentence in (49) serves as a baseline.

(49)

Baseline sentence

ŋwὲ

nsìp

[RelCP

zhi

í

rìŋ

ŋkfúú

]

k$\grave{\mathsf{ɔ}}$ŋ

yē.

person

every

3sg.rel

3sg.s

know

chief

like

3sg.anim.obj

‘Everyone who knows the chief likes him.’

(50)

Focus-fronting licenses parasitic gaps

a.

Á

ndá

(cí)

ŋwὲ

nsìp

[RelCP

zhi

í

rìŋ

――pg ]

k$\grave{\mathsf{ɔ}}$ŋ

――1

à?

foc

who

ci

person

every

3sg.rel

3sg.s

know

like

q

‘Who does everyone who knows like?’

b.

Á

ŋkfúú

(cí)

ŋwὲ

nsìp

[RelCP

zhi

í

rìŋ

――pg ]

k$\grave{\mathsf{ɔ}}$ŋ

――1

foc

chief

ci

person

every

3sg.rel

3sg.s

know

like

‘It is the chief who everyone who knows likes’

Taken together, the four reconstruction diagnostics (Principle C, variable binding, relative scope, and reconstruction conflicts) coupled with the two diagnostics that do not rely on reconstruction (superiority and parasitic gaps) suggest that the focused constituent in the á-focus construction has undergone syntactic movement into its surface position.

3.3. Selective Island-Sensitivity

In contrast to the evidence in favour of movement, we find that extraction of nominal objects, just like in Asante Twi, does not trigger any island effects. Thus, focalization of objects and wh-question formation are licit from typically strong island configurations, like a complex noun phrase with a CP-complement (51), a complex noun phrase with a relative clause (52), and an adjunct (53).

(51)

Focus and wh-extraction of DP-object from complex NP (with CP-complement)

a.

Á

njiŋwɛ1

(cí)

mὲ

mū

yōɁ

[DP

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

[CP

zhǐ-nε̄

Nfor

foc

woman

ci

1sg

pst2

hear

rumour

det

3sg.inan-comp

Nfor

bí

kɔnī

――1 ]].

fut1

meet

‘I heard the rumour that Nfor will meet a woman.’

b.

Á

rkar1

(cí)

mὲ

mū

yōɁ

[DP

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

[CP

zhǐ-nε̄

Nfor

foc

car

ci

1sg

pst2

hear

rumour

det

3sg.inan-comp

Nfor

bí

yū

――1 ]].

fut1

buy

‘I heard the rumour that Nfor will buy a car.’

c.

Á

ndá1

(cí)

wὲ

mū

yōɁ

[DP

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

[CP

zhǐ-nε̄

Nfor

bí

foc

who

ci

2sg

pst2

hear

rumour

det

3sg.inan-comp

Nfor

fut1

kɔnī

――1

à

]]?

meet

q

‘Who did you hear the rumour that Nfor will meet?’.

d.

Á

kε̄1

(cí)

wὲ

mū

yōɁ

[DP

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

[CP

zhǐ-nε̄

Nfor

bí

foc

what

ci

2sg

pst2

hear

rumour

det

3sg.inan-comp

N.

fut1

yū

――1

à

]]?

buy

q

‘What did you hear the rumour that Nfor will buy?’

(52)

Wh-extraction of DP-object from complex NP (with relative clause)

a.

Á

ndá1

(cí)

ó

mū

k$\overline{\mathsf{ɔ}}$nī

[DP

njíŋwὲ

f$\overline{\mathsf{ɔ}}$

[CP

zhǐ

í

k$\grave{\mathsf{ɔ}}$ŋ

――1

foc

who

ci

3pl.sm

pst2

meet

woman

det

rel.sg

3sg.s

likes

à

]]?

q

‘Who did they meet the woman who likes?’

Possible answer: ‘They met the woman who likes Shey.’

b.

Á

kέέ1

(cí)

ó

mū

k$\overline{\mathsf{ɔ}}$nī

[DP

njíŋwὲ

f$\overline{\mathsf{ɔ}}$

[CP

zhǐ

í

mū

yū

foc

what

ci

3pl.sm

pst2

meet

woman

det

rel.sg

3sg.s

pst2

buy

――1

à

]]?

q

‘What did they meet the woman who bought?’

Possible answer: ‘They met the woman who bought plantains.’

(53)

Wh-extraction of DP-object from adjunct

a.

Á

ndá1

(cí)

Tanko

bá

Shey

ó

cí

fàɁ

kàɁ

[CP

ànj$\acute{\mathsf{ɔ}}$Ɂ

ó

foc

who

ci

Tanko

and

Shey

3pl.sm

prog

work

neg

because

3pl.sm

cí

ró

――1

à

]?

prog

search.for

q

‘Who are Tanko and Shey not working because they are searching for?’

Possible answer: ‘Tanko and Shey are not working because they are searching for Nfor.’

b.

Á

kέέ1

(cí)

Tanko

bá

Shey

ó

cí

fàɁ

kàɁ

[CP

ànj$\acute{\mathsf{ɔ}}$Ɂ

ó

foc

what

ci

Tanko

and

Shey

3pl.sm

prog

work

neg

because

3pl.sm

cí

sε̄Ɂ

――1

à

]?

prog

fetching

q

‘Who are Tanko and Shey not working because they are fetching?’

Possible answer: ‘Tanko and Shey are not working because they are fetching wood.’

Likewise, object wh-extractions from what are typically weak islands are also grammatical. This is shown for the wh-island in (54). Note that there is no difference depending on whether the island-creating wh-element appears in situ (54b), which is the most natural position in this case, or in the focus position of the embedded clause (54c). This observation holds even when the extracted wh-element is an adjunct rather than an argument (54d).

(54)

Wh-extraction from a wh-island (Hein 2020, pp. 60–61)

a.

Shey

à

mū

bípshī

[CP

í-nɛ

Nfor

bí

zhē

kέέ

].

Shey

sm

pst2

ask

3sg-comp

Nfor

fut1

eat

what

‘Shey asked what Nfor will eat.’

b.

Á

kέέ1

(cí)

Shey

à

mū

bípshī

[CP

í-nɛ

Nfor

bí

zhē ――1

àsíɁkὲ

foc

what

ci

Shey

sm

pst2

ask

3sg-comp

Nfor

fut1

eat

when

].

‘What is it that Shey asked when Nfor will eat (it).’

c.

Á

kέέ1

(cí)

Shey

à

mū

bípshī

[CP

í-nɛ

á

àsíɁkὲ

(cí)

Nfor bí

foc

what

ci

Shey

sm

pst2

ask

3sg-comp

foc

when

ci

Nfor fut1

zhē

――1

à

].

eat

q

‘Shey asked what Nfor will eat when.’

d.

Á

àsíɁkὲ

(cí)

Shey

à

mū

bípshī

[CP

í-nɛ

á

kέέ

(cí)

Nfor bí

foc

when

ci

Shey

sm

pst2

ask

3sg-comp

foc

what

ci

Nfor fut1

zhē

à

].

eat

q

‘Shey asked when Nfor will eat what.’

At first glance, this might suggest that focalization from inside these islands does not involve movement. However, we find that the same reconstruction effects as for non-island focus appear in focalizations from inside an island (using the CNP-island for illustration here). Thus, we observe binding reconstruction for Condition C (55) (compare (40)) and for variables (56) (compare (41)) as well as for relative quantifier scope (57) (compare (42)).

(55)

Reconstruction for Principle C into complex NP island

Á

Nfori

(cí)

í*i/j

mū

yōɁ

[DP

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

[CP

zhǐ-nε̄

à

cí

foc

Nfor

ci

3sg.s

pst2

hear

rumour

def

3.inan-comp

2sg

prog

ró

――i

]].

search

‘He*i/j has heard the rumour that you are searching for Nfori.’

(56)

Reconstruction for variable binding into complex NP island

Á

[táā

ŋgwá

zhìy/x]1

(cí)

mὲ

mū

yōɁ

[DP

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

[CP

foc

father

wife

3sg.poss

ci

1sg

pst2

hear

rumour

def

zhǐ-nε̄

[ŋwὲ

nsìp]x

bí

k$\overline{\mathsf{ɔ}}$nī

――1 ]].

3sg.inan-comp

man

every

fut1

meet

‘I heard a rumour that every manx will meet the father of hisx/y wife.’

(57)

Reconstruction for scope into complex NP island

Á

[mbàŋrù]1

cí

mὲ

mū

yōɁ

[DP

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

[CP

zhǐ-nε̄

njíŋwὲ

foc

male

ci

1sg

pst2

hear

rumour

def

3sg.inan-comp

woman

nsìp

à

mū

bzú

――1

]].

every

sm

pst2

birth

‘I heard a rumour that every woman gave birth to a son.’   (*$\exists >\forall$, $\forall >\exists$)

Importantly, we also find that focalization from inside an island gives rise to the same reconstruction conflicts as from the non-island configuration in (47). Thus, if the binder njíŋwὲ nsip ‘every woman’ of the variable associated with the extracted possessive marker bvi ‘her’ is the subject of the embedded clause that constitutes the complex NP island, focalization is grammatical (58a). This is because the extracted DP ŋkār bō bvi ‘friend of her children’ can reconstruct into the intermediate position ${\mathrm{t}}_1^{\prime }$ below the binder of the possessive bvi, where it is furthermore not c-commanded by the subject pronoun ó ‘they’ of the embedded clause. This is important in order to avoid a violation of Condition C as the embedded subject ó is coreferent with the noun bō ‘children’ inside the focused DP. However, if the universal quantifier is the subject of the embedded clause, focalization is ungrammatical (58b) because reconstruction of the focused DP into both the intermediate (${\mathrm{t}}_1^{\prime }$) and base position (_――1) would lead to a violation of Condition C since the pronominal matrix subject ó ‘they’ would c-command the coreferent R-expression bō ‘children’.

(58)

Reconstruction conflict in complex NP island

a.

Á

[DP

ŋkār

[

bō

bvix]y

]1

(cí)

njíŋwὲ

nsipx

à

mū

yōɁ

[DP

foc

friend

children

3sg.poss

ci

woman

every

sm

pst2

hear

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

[CP

${\mathrm{t}}_1^{\prime }$

zhǐ-nε̄

óy

mū

cèb

――1]].

rumour

def

3sg.inan-comp

3pl.sm

pst2

insult

‘It is a friend of [herx children]y that every womanx heard the rumour that theyy insulted.’

b.

*Á

[DP

ŋkār

[

bō

bvix]y

]1

(cí)

óy

mū

yōɁ

[DP

nsūŋ

(f$\overline{\mathsf{ɔ}}$)

foc

friend

children

3sg.poss

ci

3pl.sm

pst2

hear

rumour

def

[CP

${\mathrm{t}}_1^{\prime }$

zhǐ-nε̄

njíŋwὲ

nsipx

à

mū

cèb

――1 ]].

3sg.inan-comp

woman

every

sm

pst2

insult

‘It is a friend of [herx children]y that theyy heard the rumour that every womanx insulted.’

We may therefore conclude that the á-focus construction always involves $\overline{\mathrm{A}}$-movement independent of whether or not the focused nominal phrase or wh-element originates from inside an island configuration. Note that in contrast to Asante Twi, we do not observe the presence of a resumptive pronoun with any of the nominals, be they animate or inanimate.

As was the case in Asante Twi, the absence of island effects with nominal focus does not entail that island constraints are inactive in Limbum entirely. First note that focus-fronting of VPs (59) and PPs (60) is possible in Limbum.

(59)	VP focus (Becker and Nformi 2016, p. 75)
	Á	r-[VP	yū	msāŋ]1	(cí)	njíŋwὲ	f$\overline{\mathsf{ɔ}}$	bí	gī	――1.
	foc	nmlz-	buy	rice	ci	woman	det	fut1	do
	‘It is buying rice that the woman will do.’
(60)	PP focus
	Á	[PP	ní	pkūh]1	(cí)	Nfòr	n$\grave{\mathsf{ɔ}}$ŋī	――1.
	foc		in	bed	ci	N.	sleep
	‘It is in the bed that Kofi is lying.’

This fronting may also cross a clause boundary, as shown in (61) for VPs and also in (62), although slightly degraded, for PPs.

(61)

Long-distance VP focus (Hein 2020, p. 59)

Á

r-[VP

bò

ndāp]1

(cí)

mὲ

kwàshī

[CP

mὲ-nɛ

Nfor bí

gī

――1.

].

foc

nmlz-

build

house

ci

1sg

think

1sg-comp

Nfor fut1

do

It is building a house that I think that Nfor will do.

(62)

Long-distance PP focus

?Á

[PP

ní

pkūh]1

(cí)

mὲ

kwàshī

[CP

mὲ-nɛ

Nfòr

n$\grave{\mathsf{ɔ}}$ŋī

――1

].

foc

in

bed

ci

1sg

think

1sg-comp

Nfor

sleep

‘It is in bed that I think that Nfor is sleeping.’

However, when the VPs or PPs originate from inside a strong island, their focus-fronting incurs a violation. This is shown for the three familiar configurations, i.e., complex NPs in (63) and (64), relative clauses in (65) and (66), and adjuncts in (67) and (68).13

(63)

VP focus from a complex NP (with CP-complement)

*Á

r-[VP

yū

msāŋ]1

(cí)

mὲ

mū

yōɁ

[DP

nsūŋ

[CP

zhǐ-nε̄

Nfòr

foc

nmlz-

buy

rice

ci

1sg

pst2

hear

rumour

3sg.inan-comp

Nfor

bí

gī

――1

]].

fut1

do

‘I heard a rumour that Nfor will buy rice.’

(64)

PP focus from a complex NP (with CP-complement)

*Á

[PP

ní

pkūh]1

(cí)

mὲ

mū

yōɁ

[DP

nsūŋ

[CP

zhǐ-nε̄

Nfòr

n$\grave{\mathsf{ɔ}}$ŋī

――1

]].

foc

in

bed

ci

1sg

pst2

hear

rumour

3sg-comp

Nfor

sleep

‘It is in the bed that I heard a rumour that Nfor is lying.’

(65)

VP focus from a complex NP (with relative clause)

*Á

r-[VP

yū

msāŋ]1

(cí)

ó

mū

k$\overline{\mathsf{ɔ}}$nī

[DP

njíŋwὲ

f$\overline{\mathsf{ɔ}}$

[CP

zhǐ

í

foc

nmlz-

buy

rice

ci

3pl.sm

pst2

meet

woman

det

rel.sg

3sg.s

bí

gī

――1

]].

fut1

do

‘They met the woman who will buy rice.’

(66)

PP focus from a complex NP (with relative clause)

?Á

[PP

mà

ntāā]1

(cí)

ó

mū

k$\overline{\mathsf{ɔ}}$nī

[DP

njíŋwὲ

f$\overline{\mathsf{ɔ}}$

[CP

zhǐ

í

mū

dò

foc

to

market

ci

they

pst2

meet

woman

det

rel.sg

3sg.s

pst2

go

――1 ]].

‘They met the woman who went to the market.’

(67)

VP-focus from an adjunct

*Á

r-[VP

yū

msāŋ]1

(cí)

Nfor

à

cí

fàɁ

kàɁ

[CP

ànj$\acute{\mathsf{ɔ}}$Ɂ

í

cí

gī

foc

NMLZ-

buy

rice

ci

Nfor

sm

prog

work

neg

because

3sg.s

prog

do

――1 ].

‘Nfor is not working because he is buying rice.’

(68)

PP-focus from an adjunct

?Á

[PP

mà

ntāā]1

(cí)

Nfor

à

cí

fàɁ

kàɁ

[CP

ànj$\acute{\mathsf{ɔ}}$Ɂ

í

mū

dò

――1

].

foc

to

market

ci

Nfor

sm

prog

work

neg

because

3sg.s

pst2

go

‘Nfor is not working because he went to the market.’

An exception to this pattern is the wh-island, from which both VPs (69) and PPs (70) may undergo focus-fronting without incurring an island violation.

(69)

VP focus from wh-island

Á

r-[VP

bò

ndāp]1

(cí)

Shey

à

mū

bípshī

[CP

í-nɛ

á

àsíɁkὲ

(cí)

foc

nmlz-

build

house

ci

Shey

3sg

pst2

ask

3sg-comp

foc

when

ci

Nfor

bí

gī

――1

à

].

Nfor

fut1

do

q

‘Shey asked when Nfor will build a house.’

(70)

PP focus from wh-island

Á

[PP

ní

pkūh]1

(cí)

Shey

à

mū

bípshī

[CP

í-nɛ

á

àsíɁkὲ

(cí)

Nfor

foc

in

bed

ci

Shey

sm

pst2

ask

3sg-comp

foc

when

ci

Nfor

bí

n$\grave{\mathsf{ɔ}}$ŋī

――1

à

].

fut1

sleep

q

‘It is in bed that Shey has asked when Nfor will be sleeping.’

We can thus conclude that complex NPs (of the CP-complement and the relative clause type) and adjuncts do have island status in the language. This status only seems to be suspended if the extracted element is nominal (and an argument).14 Embedded wh-configurations, however, apparently do not constitute an island configuration for focus-fronting, independent of the properties of the extracted element. On the surface, the pattern in Limbum therefore looks very parallel to the one in Asante Twi. Nominal arguments may extract freely from islands, while VPs and PPs incur violations in these contexts despite being focus-frontable in non-island environments.

3.4. No Repair by Resumption in Limbum

Recall that the exemption of nominals from island constraints in Asante Twi could at least partially be explained by them leaving a resumptive pronoun that circumvents a violation of a representational PF-island constraint before it is deleted (for inanimate pronouns). This proposal gains some support in Twi from the fact that the resumptive pronoun appears overtly with animate objects. Given the close similarity between island-sensitivity patterns in Twi and Limbum, one might wonder whether this analysis can be transferred to the Limbum data.

While all examples hitherto contained a gap in the base position of the ex situ focused element, overt resumption is indeed an option in Limbum (71).

(71)	Object focus with resumptive pronoun
	a.	Á	njiŋwɛ1	(cí)	Nfor	bí	kɔnī	{	――1	/	yē1}.
		foc	woman	ci	Nfor	fut1	meet				3sg.o
		‘Nfor will meet a woman.’
	b.	Á	rkar1	(cí)	Nfor	bí	yū	{	――1	/	zhi1}.
		foc	car	ci	Nfor	fut1	buy				3sg.inan
		‘Nfor will buy a car.’

However, this option is severely restricted. Resumption only occurs with focused bare nouns. It is ungrammatical with virtually all other nominal elements, including proper names (72), pronouns (73), definite nouns marked with the definite determiner f$\overline{\mathsf{ɔ}}$ (74), and wh-items (75).

(72)

No resumption with focused proper names

Á

Tanko1

(cí)

Nfòr

à

mū

yε̄

{

*yē1

/

――1}

nìŋkòr.

foc

Tanko

ci

N.

sm

pst2

see

3sg.o

yesterday

‘Nfor saw Tanko yesterday.’

(73)

No resumption with focused pronouns

Á

yē1

(cí)

Nfor

à

mū

yε̄

{

*yē1

/

――1}

nìŋkòr.

foc

3sg

ci

Nfor

sm

pst2

see

3sg.o

yesterday

‘Nfor saw him/her yesterday.’

(74)	No resumption with focused nouns marked with a definite determiner
	Á	[ŋwὲ-ryēɁnì	f$\overline{\mathsf{ɔ}}$]1	cí	í	bí	k$\overline{\mathsf{ɔ}}$nī	{	*yē1	/ ――1}	ntómbzù.
	foc	man-teach	def	ci	Nfor	fut1	meet		3sg.o		first
	‘He will meet the teacher first.’

(75)

No resumption with focused wh-items

Á

ndá1

(cí)

Nfor

à

mū

yε̄

{

*yē1

/ ――1}

nìŋkòr

à?

foc

who

ci

Nfor

sm

pst2

see

3sg.o

yesterday

q

‘Who did Nfor see yesterday?’

This also holds for long-distance focus, where the focused element originates inside an embedded clause, as shown for proper names in (76), pronouns in (77), definite nouns in (78), and wh-items in (79).

(76)

No resumption with long-distance focused proper names

Á

Tanko1

(cí)

mὲ

kwàɁshī [CP

mὲ-nɛ

Nfòr

à

mū

yε̄

{

??yē1 / ――1}

foc

Tanko

ci

1sg

think

1sg-comp

Nfor

sm

pst2

see

3sg.o

nìŋkòr

].

yesterday

‘It is Tanko who I think that Nfor saw (him) yesterday.’

(77)	No resumption with long-distance focused pronouns
	Á	yì1	(cí)	mὲ	kwàɁshī [CP	mὲ-nɛ	Nfòr	à	mū	yε̄	{ ??yì1 / ――1}
	foc	2pl	ci	1sg	think	1sg-comp	Nfor	sm	pst2	see	2pl
	nìŋkò	].
	yesterday
	‘It is you(pl) who I think that Nfor saw (you) yesterday.’

(78)

No resumption with long-distance focused nouns with a definite determiner

Á

[njíŋwὲ

f$\overline{\mathsf{ɔ}}$]1

(cí)

mὲ

rìŋ

[CP

mὲ-nɛ

Nfor

bí

k$\overline{\mathsf{ɔ}}$nī

{

*yē1

/ ――1} ].

foc

woman

det

ci

1sg

know

1sg-comp

Nfor

fut1

meet

3sg.o

‘It is the woman who I know that Nfor will meet (her).’

(79)

No resumption with long-distance focused wh-items

Á

ndá1

(cí)

wὲ

kwàɁshī [CP

wὲ-nɛ

Nfòr

à

mū

yε̄

{

??yē1

/ ――1}

foc

who

ci

2sg

think

2sg-comp

Nfor

sm

pst2

see

3sg.o

nìŋkòr

à

]?

yesterday

q

Closer inspection further reveals that with bare nouns in the focus position, a resumptive only occurs if the noun is interpreted as specific; even in that case, it is only one additional option besides a gap. Thus, in a context that strongly prefers a specific interpretation, both a gap and a resumptive pronoun may occur in the base position of the object (80).

(80)	Context: Nfor has been very nervous today at school. When his classmate and best friend Tanko asks him why that is, Nfor tells him that he has a very important meeting with one of their teachers, Mr. Bassong, next week to talk about a stipend. In the afternoon, Nfor and Tanko meet up with their friend Shey, who is going to a different school. He also notes that Nfor seems very nervous and asks him why. Nfor only says that he has an important appointment coming up. After Nfor has left, Shey asks Tanko:
	Q:	Á	ndá1	(cí)	Nfor	bí	k$\overline{\mathsf{ɔ}}$nī	――1	à?
		foc	who	ci	Nfor	fut1	meet		Q
		‘Who(m) will Nfor meet?’
	Tanko replies:
	A:	Á	ŋwὲ-ryēɁni1	(cí)	Nfor	bí	k$\overline{\mathsf{ɔ}}$nī	{ ――1 /	yē}.
		foc	man-teach	ci	Nfor	fut1	meet		3sg.o
		‘Nfor will meet a teacher.’

In a context that strongly prefers a non-specific interpretation of the focused noun, however, only a gap is acceptable (81).

(81)	Context: Nfor has been very nervous today. When his friend Tanko asks him why that is, Nfor tells him that in today’s mail he received a summons to the police station for a testimony next week. In the afternoon, Nfor and Tanko meet up with their friend Shey. He also notes that Nfor seems very nervous and asks him why. Nfor only says that he has an important appointment coming up. When Nfor has left, Shey asks Tanko:
	Q:	Á	ndá1	(cí)	Nfor	bí	k$\overline{\mathsf{ɔ}}$nī	――1	à
		foc	who	ci	Nfor	fut1	meet		q
		‘Who(m) will Nfor meet?’
	Tanko replies:
	A:	Á	ŋwὲ-rta1	(cí)	Nfor	bí	k$\overline{\mathsf{ɔ}}$nī	{――1 /	#yē1}.
		foc	man-cap	ci	Nfor	fut1	meet		3sg.o
		‘Nfor will meet a policeman.’

In other words, the presence of a resumptive pronoun forces a focused bare noun indefinite to receive a specific interpretation. This property of Limbum object focus extraction is not extraordinary cross-linguistically, as it has been observed for other languages that optional resumption requires a specific/referential antecedent (Aoun et al. 2001; Bianchi 2004; Doron 1982; Sharvit 1999; Sichel 2014; Suner 1998).

Drawing the parallel to Asante Twi, one could, of course, assume that every focus extraction underlyingly leaves a resumptive pronoun. The lack of overt resumptives with proper names, pronouns, definites, and non-specific indefinites would then be due to a PF-deletion rule. However, there are some indications that this is an implausible solution. First, gaps in Twi nominal extraction occur in a natural class of contexts: namely, with inanimate objects. The contexts in which no resumption may occur in Limbum do not form a natural class. Rather, the opposite is true: namely, that the contexts that allow resumptives—inanimate and animate specific bare nouns—form a natural class. If one were to treat resumption in Limbum on a par with resumption in Asante Twi—that is, as the default output of extraction from object position—one would have to restrict the domain of application of a purported PF-deletion rule to all nominals except specific indefinites. In addition, such a rule would have to be applicable to specific indefinites only optionally since an overt resumptive is not mandatory with them (80). Second, unlike in Twi, there is no evidence that an alleged underlying resumptive pronoun appears overtly in any other context except specific indefinites. Clause-final adverbs, for instance, do not force an overt resumptive instead of a gap (72)–(79).

What is more, there already exists a deletion rule in Limbum that is almost identical to the one proposed for Asante Twi. It applies to regular object pronouns in discourse-anaphoric use and optionally deletes them (82) (cf. object (pro-)drop).

(82)

a.

Yaa

à

mū

yε̄

rkār

zhì

í

k$\grave{\mathsf{ɔ}}$ŋ.

Ndū

zhì

bí

yū

Yaa

sm

pst2

see

car

rel.sg

3sg.s

like

husband

3sg.poss

fut1

buy

(zhi)

àyàŋsè.

3sg.inan.o

tomorrow

‘Yaa saw a car that she likes. Her husband will buy it tomorrow.’

b.

Yaa

à

mū

yε̄

mŋkòb

bvì

í

k$\grave{\mathsf{ɔ}}$ŋ.

Ndū

zhì

bí

yū

Yaa

sm

pst2

see

suitcases

rel.pl

3sg.s

like

husband

3sg.poss

fut1

buy

(bvi)

àyàŋsè.

3pl.inan.o

tomorrow

‘Yaa saw suitcases (that she likes). Her husband will buy them tomorrow.’

However, like the deletion rule for object pronouns in Asante Twi, it is restricted to inanimate pronouns. Deletion of animate object pronouns leads to ungrammaticality (83).

(83)

a.

Yaa

à

mū

yε̄

Shey.

Ndū

zhì

bífu

kɔnī

*(yē)

àyàŋsè.

Yaa

sm

pst2

see

Shey

husband

3sg.poss

fut3

meet

3sg.o

tomorrow

‘Yaa saw Shey. Her husband will met him tomorrow.’

b.

Yaa

à

mū

yε̄

Shey

ba

Ngala.

Ndū

zhì

bífu

kɔnī

*(wō)

Yaa

sm

pst2

see

Shey

and

Ngala

husband

3sg.poss

fut3

meet

3pl.o

àyàŋsè.

tomorrow

‘Yaa saw Shey and Ngala. Her husband will meet them tomorrow.’

If, as in Twi, this rule were responsible for the deletion of underlying resumptive pronouns and the occurrence of pseudo-gaps, it should spare animate resumptives. Under the assumption that every nominal extraction first leaves a resumptive pronoun to later be deleted at PF, we would therefore expect that pseudo-gaps would be restricted to inanimate objects, contrary to fact.

In light of these observations, it seems to be the case that island constraints truly are selectively inactive for nominal (argument) focalization. If one can focus a nominal object via movement from inside an island and leave a true gap (as opposed to a silent or PF-deleted resumptive pronoun, as in Asante Twi) the island simply cannot hold. If this conclusion is on the right track, if not for Asante Twi then at least for Limbum, it immediately raises two questions: (i) Why exactly does the split in extractability run between nominal and non-nominal elements? (ii) How can we model this permeability in a formal system? I will address both questions in the following section.

4. An Account of Selective Island Permeability in Limbum

With regard to the first question, it is worth noting that the division between exceptionally extractable and non-extractable categories is basically the same as in English, as discussed in the context of untensed strong islands in Section 1. That is, while DPs can, under certain circumstances, be moved from inside an island, PPs (and VPs) cannot. For English, Cinque (1990) linked this to the availability of silent resumptive pronouns for DPs but not for PPs and VPs. In a similar way, the category-sensitive island permeability in Asante Twi can possibly be traced back to silent resumptive pronouns, too, as argued for by Korsah and Murphy (2020, 2024). However, as we have seen in the previous section, these approaches do not straightforwardly carry over to Limbum. Hence, there must be some other difference between DPs on the one side and PPs and VPs on the other that is responsible for their different behaviour in island extractions in Limbum. I argue that this difference concerns the presence of $\varphi$-features, which are encoded on DPs but arguably not on PPs or VPs. Concerning the second question, note that the clause-level in Limbum shows sensitivity to $\varphi$-features, as reflected by the fact that it exhibits complementizer agreement. It is this $\varphi$-sensitivity and its interaction with $\varphi$-features on other constituents in the clause that, as I argue, leads to the pattern of category-sensitive selective island permeability that we observe in Limbum. More specifically, complementizer agreement shows some exceptional behaviour in island contexts, where non-$\varphi$-bearing elements are blocked from extraction. I suggest that it is the exceptional behaviour of complementizer agreement in islands that leads to the observed unextractability of non-$\varphi$-bearing constituents like PP and VP. In what follows, I will first provide an overview of the complementizer agreement pattern and then propose a possible implementation of its connection to island permeability.

4.1. Complementizer Agreement in Limbum

In patterns of complementizer agreement (CA), the complementizer of an embedded clause agrees for $\varphi$-features with an argument. Languages vary as to whether agreement takes place with an argument of the matrix clause or of the embedded clause. The former type is not untypical for African languages (Baker 2008 on Kinande; Idiatov 2010 on Mende languages; Carstens 2016; Diercks 2013 on Lubukusu; Duncan and Torrence 2017 on Ibibio; Letsholo and Safir 2019 on Ikalanga; Driemel and Kouneli 2024 on Kipsigis), while the latter type has been documented for Germanic varieties (Carstens 2003; Fuß 2008, 2014; Haegeman and van Koppen 2012; Shlonsky 1994; van Alem 2024; van Koppen 2005, 2012; Zwart 1997).

Nformi (2018) documents and discusses a pattern of complementizer agreement of the former type in Limbum. In what follows, I will briefly report his data.15 In Limbum, the complementizer nɛ is prefixed with an element that covaries depending on the $\varphi$-features of the subject of the matrix clause. The different forms of the prefix are given in (84).

(84)

 Complementizer agreement prefixes (Nformi 2018, p. 2)

	sg	pl
1exc.	mɛ-	wέr-
1inc.	—	sí-
2	wέ-	yì-
3	í-	ó-

As is observable in the examples throughout this paper, the agreement prefix tracks the $\varphi$-features of the matrix subject (85).

(85)	a.
		‘The people have said that we should come.’
	b.
		‘We have said that the people should come.’	(Nformi 2018, p. 2)

Furthermore, agreement targets the closest, i.e., the local, matrix subject (86).

(86)
	‘Nfor said that people are reporting that you have come.’          (Nformi 2018, p. 2)

While it has been argued based on Mende languages that agreement of this sort is typically logophorically controlled (Idiatov 2010), Nformi (2018) shows this is not the case for Limbum. In (87), the source of information, i.e., the logophoric centre of the discourse, is the child, whose $\varphi$-features are third-person singular. Nonetheless, the complementizer exhibits a first-person plural agreement prefix, which is exactly the $\varphi$-feature set of the matrix subject.

(87)
	‘We heard from the child that Nfor fetched wood.’          (Nformi 2018, p. 3)

Agreement is with the subject even in cases where that subject is an expletive (88).

(88)	a.
		‘It was good that I went home.’
	b.
		‘It seems that your children have come.’	(Nformi 2018, p. 4)

Interestingly, complementizer agreement with the subject is blocked by an intervening indirect object. In addition, this intervention is defective (Chomsky 2000), as the intervening object cannot itself act as the controller of agreement (89). Instead, the complementizer shows up in its bare, unprefixed form nɛ. In the absence of the object, the complementizer agrees with the matrix subject as expected (90).

(89)	a.
		‘Paul has told me that you have travelled.’
	b.
		‘I have told Paul that you have travelled.’	(Nformi 2018, p. 5)

(90)	a.
		‘Paul has said that you have travelled.’
	b.
		‘I have said that you have travelled.’	(Nformi 2018, p. 5)

The same pattern emerges with experiencers that intervene between the complementizer and the expletive subject of the matrix clause. Agreement cannot be established: neither for the subject nor for the experiencer (91).

(91)	a.
		‘It pained Tanko that I left.’
	b.
		‘It pleased the people that their child suceeded.’	(Nformi 2018, p. 5)

What is more, some intervening phrases seem to not block the agreement relation between the matrix subject and the complementizer even though they appear in the right position to be able to do so. Thus, objects embedded in a PP (92) as well as causees in a causative construction (93) pose no obstacle for CA with the matrix subject.

(92)	a.
		‘Shey said to the children that they will go.’
	b.
		‘My mother cooked food for us so that we might come.’	(Nformi 2018, p. 6)

(93)	a.
		‘The people frightened Nfor that they will report him to me.’
	b.
		‘I made the children lie down so that they can fall asleep.’	(Nformi 2018, p. 6)

The overall pattern seems to be very similar to familiar cases of dative intervention (Bobaljik 2008; Chomsky 2000, 2001; Preminger 2014; Řezáč 2008), where dative-marked arguments (as indirect objects and experiencers often are) block an otherwise expected agreement relation and themselves fail to control agreement.

4.2. Complementizer Agreement as Upwards Agree

Nformi (2018) shows that the Limbum pattern of complementizer agreement can be captured neither by indirect Agree (Diercks 2013) nor by delayed valuation (Carstens 2016), both of which have been proposed to account for upwards CA in Lubukusu. He then goes on to formulate an approach in terms of direct upwards Agree (Bjorkman and Zeijlstra 2019; Zeijlstra 2004, 2012), where the C-head, which hosts the complementizer, carries an unvalued $\varphi$-probe [$u\varphi$: □] that initiates an upwards search for a goal with valued $\varphi$-features that c-commands it (cf. Letsholo and Safir 2019; McFadden and Sundaresan 2021 for other accounts of CA in terms of direct upwards Agree). Thus, CA takes the form in (94), where the $\varphi$-probe on C probes upwards for $\varphi$-features, finds and agrees with the $\varphi$-bearing DP subject, and copies its $\varphi$-features onto C.

(94)	Upwards CA in Limbum

Now, in order to capture the defective intervention, Nformi (2018) adopts a case-based approach (Bobaljik 2008). Assuming that Bantu languages have abstract case (Carstens 2016; Carstens and Diercks 2013), Nformi suggests that the $\varphi$-probe on C is relativized to nominative case. This means that the $\varphi$-probe on C will only copy $\varphi$-features from a nominative-marked goal. In a sentence like (95), the dative-marked indirect object mɛ ‘1sg’ is encountered first by the upward-probing [$u\varphi$: □] feature on C. This halts the probe’s search. However, the indirect object’s $\varphi$-features are unable to value the probe, giving rise to failed agreement. Somewhat unexpectedly, this does not result in default agreement or ungrammaticality. Instead, the complementizer simply appears in its bare form nɛ.

(95)
	‘Paul has told me that you have travelled.’	(Nformi 2018, p. 5)

In contrast to dative-marked indirect objects and experiencers, causees and DP-complements of prepositions located between the matrix subject and the embedded complementizer do not intervene for agreement. Nformi (2018) suggests that this is because they are embedded under an additional structural layer and therefore do not c-command the probe, which is a prerequisite for Upward Agree. To summarize, an embedded C-head bears an unvalued [$u{\varphi }_{nom}$: □] feature that discriminates for nominative case and initiates Upward Agree with a c-commanding argument DP. If that DP bears nominative case, its $\varphi$-features value the probe. If it bears a non-nominative case, it halts the probe’s search but fails to value it, resulting in a bare form of the complementizer.

4.3. Upward Complementizer Agreement Unlocks the Phase

With this in place, we can turn to an account of selective island permeability in Limbum. As mentioned before, I will assume that it is the $\varphi$-features of DPs that allow them to undergo $\overline{\mathrm{A}}$-movement from an island. What is more, I suggest that $\overline{\mathrm{A}}$-movement leaving any embedded CP is contingent on the moving constituent bearing $\varphi$-features. As evidenced by CA, the C-layer is in some sense $\varphi$-sensitive in Limbum. Adopting a phase-based account of successive-cyclic movement with vP and CP as phases (Chomsky 2000, 2001, 2008), this can be achieved by restricting the edge of the CP-phase, i.e., SpecCP, to $\varphi$-bearing elements. Thus, an embedded C-head will check its specifier for $\varphi$-features. If this checking is not successful, as is the case with non-DP elements, like a PP in (96), the derivation crashes. Without further modifications, this restriction on the edge of the CP-phase would prevent any $\overline{\mathrm{A}}$-extraction of PPs and VPs from an embedded clause, as they do not bear $\varphi$-features (though elements inside them might well do). This is clearly empirically false.

(96)	Restriction of SpecCP to ϕ-bearing constituents

This is where CA becomes important. It has been argued that Agree has the potential to ‘unlock’ an already completed phase thereby allowing an otherwise illicit extraction from it (Branan 2018; Davis and Branan 2019; Halpert 2016, 2018; Rackowsky and Richards 2005). Commonly, this unlocking is achieved by standard downwards Agree between the matrix verb and the embedded CP. Here, I argue that Upward Agree between the embedded complementizer and an argument in the matrix clause also makes an already completed CP-phase accessible again.16 Thus, while VPs and PPs may not initially be accessible for extraction to outside the embedded clause because they cannot land in the intermediate escape hatch position in SpecCP, they become available for movement again when the CP-phase is unlocked by complementizer agreement. Adopting the weak version of the Phase Impenetrability Condition (PIC, Chomsky 2001) that states that the domain of a phase head becomes inaccessible only when the next-higher phase head is merged, let us consider the derivation of a regular long-distance focus of a PP.

In (97), as a first step, the PP will have to be moved into the edge of the embedded vP phase. When C is merged, the vP phase becomes locked. Upon merging of matrix v, the CP-phase domain will become locked, making the PP at the edge of the embedded vP no longer accessible for movement. It can therefore not undergo movement to matrix SpecvP. In order for that to be possible, it would have had to move to SpecCP in a previous step. This movement, however, is pre-empted by the restriction of SpecCP to $\varphi$-bearing constituents (96).

(97)	Locked embedded CP

As shown in (98), once the nominative-bearing matrix subject is merged into the structure, the embedded complementizer will agree with it and will thereby unlock the CP-phase. This makes the PP in the embedded SpecvP accessible again. It will then undergo intermediate movement into matrix SpecvP above the subject.

(98)	Unlocking of CP-phase via Upward Agree

In the next step, T is merged and checks for nominative case on the subject, triggering its movement to SpecTP. The PP either reaches its final landing site in the focus position of the matrix clause or can move to a SpecvP of a yet higher clause in a similar fashion as it has moved from the embedded SpecvP into matrix SpecvP.

4.4. Explaining Island Permeability: Complementizer Agreement in Clausal Islands

We can now turn to the analysis of selective island permeability. In most of the island cases above, we are dealing with a clausal constituent, i.e., a complement clause, a relative clause, or an adjunct clause. Interestingly, the latter two clause types do not show any complementizer agreement. The first clause type does, although its complementizer’s form is different from those listed in (84), a fact I will come back to momentarily (in Section 4.4.2).

4.4.1. No Complementizer Agreement in Relative Clauses and Adjunct Clauses

The adjunct island is introduced by a non-agreeing complementizer ànj$\acute{\mathsf{ɔ}}$Ɂ ‘because’ in (53a), repeated below as (99a), and the relative clause only shows a relative pronoun zhǐ (or vǐ for a plural head noun), which is arguably a realization of the relative operator, in (52a), repeated below as (99b).

(99)	a.
		‘Who are Tanko and Shey not working because they are searching for?’
		Possible answer: ‘Tanko and Shey are not working because they are searching for Nfor.’
	b.
		‘Who did they meet the woman who likes?’
		Possible answer: ‘They met the woman who likes Shey.’

The absence of any agreement prefixes on the complementizers in these cases, as I argue, shows that the relevant C-heads do not carry a $\varphi$-probe that could trigger Upward Agree. Nonetheless, the restriction of SpecCP to $\varphi$-bearing constituents still holds. Assuming that whatever makes relative clauses and adjunct clauses islands in languages like English is absent in Limbum, it is then straightforward as to why DPs may freely evacuate from them but PPs and VPs may not. A $\varphi$-bearing DP can be moved to the edge of the CP phase before it is locked and is therefore accessible for further movement into higher phase edges or final landing sites. A PP or VP must remain in the edge of the embedded vP and therefore depends on CA to unlock the embedded CP-phase in order to be able to reach the matrix vP edge, and, consequently, any landing site in higher clauses. Since such unlocking by CA is absent in relative clauses and adjunct clauses, they constitute impermeable domains for PP and VP extraction.

4.4.2. Object Complementizer Agreement in Complement Clauses Is Too Early

The only island cases where there is an overt complementizer that is prefixed with what looks like an agreement morpheme are complement clauses such as the one in (51a), repeated below as (100).

(100)
	‘I heard the rumour that Nfor will meet a woman.’

However, this complementizer clearly does not agree with the matrix subject. If this were the case, the complementizer should take the form í-nɛ. In fact, the prefix zhǐ- does not appear at all in the list of prefix forms (84) provided by Nformi (2018), repeated in (101). One could try to argue that it is the default agreement form that occurs in the absence of agreement or when agreement has failed. However, as noted above for example (95), failure of CA in, e.g., dative intervention cases results in an unprefixed complementizer rather than a purported default form zhǐ-nε̄. Therefore, I will argue here that the complementizer in (100) actually agrees with the noun that embeds it: that is, it agrees with the matrix direct object nsūŋ (f$\overline{\mathsf{ɔ}}$) ‘(the) rumour’. Consequently, Nformi’s (2018) assessment of CA in Limbum must be incomplete, as CA does not skip all objects but only the ones that do not bear accusative. Indeed, in all of his examples, the interveners were either arguably dative-marked indirect objects or experiencers but never straightforwardly accusative-marked direct objects.

That (100) constitutes a genuine case of agreement with a direct object, meaning that the $\varphi$-probe on C must be relativized to structural case rather than nominative, is supported by a closer investigation of the actual morphological form of the agreement prefix. First, note that the prefixes for subject agreement provided by Nformi (2018) in (84), repeated below as (101), essentially take the same form as the subject pronouns (102).17

(101)	Subject CA prefixes (Nformi 2018, p. 2)
		sg	pl
	1exc.	mɛ-	wέr-
	1inc.	—	sí-
	2	wέ-	yì-
	3	í-	ó-
(102)	Subject pronouns (Hein 2021, p. 136)
		sg	pl
	1exc.	mὲ	wὲr
	1inc.	—	sì
	2	wὲ	yì
	3.anim	í	wōyè
	3.inan	í	bvī

While the object series of personal pronouns in (103) is largely syncretic with the subject one, it crucially differs in the third-person. Interestingly, the third-person object pronoun for inanimates zhī is nearly identical with the prefix zhǐ- on the complementizer in (100).

(103)	Object pronouns (Hein 2021, p. 136)
		sg	pl
	1exc.	mὲ	wὲr
	1inc.	—	sì
	2	wὲ	yì
	3.anim	yé	wō
	3.inan	zhī	bvī

Assuming that object agreement prefixes, like subject agreement prefixes, closely resemble their respective pronouns, this speaks against treating zhǐ- in (100) as a default prefix and in favour of it being an actual object agreement prefix. If this is correct, we would expect the prefix to vary depending on the number and animacy of the matrix object. Thus, turning the object nsūŋ ‘rumour’ in (100) into the plural msūŋ ‘rumours’ should yield a complementizer prefix that looks very much like the object pronoun bvī. As shown in (104), this is indeed the case.

(104)
	‘I heard rumours that Nfor will meet a woman.’

I conclude from this that the complementizer in examples such as (100) and (104) agrees with its embedding noun, i.e., the matrix object. More generally, this means that CA in Limbum is not restricted to matrix subjects (pace Nformi 2018) but also holds for direct objects. Since in most cases the embedded clause itself acts as the direct object of the matrix clause, the occurrence of object CA is restricted to structures in which the embedded clause is embedded into a direct object DP. In terms of probing, this means that the $\varphi$-probe on the complementizer is not relativized to nominative case but rather to structural case, which encompasses both nominative and accusative but, crucially, not dative.

Given that CA takes place in complex NP islands with complement clauses, which should enable the extraction of PPs and VPs, why can they not be moved from inside the complement clause? The answer is because object agreement takes place before matrix v is merged. It therefore comes too early to unlock the CP phase, which will only become locked when the matrix v head enters the structure. Since the probe on embedded C has already entered into an Agree relation with the matrix object DP, it cannot again Agree with the subject and can thus not unlock the CP phase (105).

(105)	No unlocking of CP-phase via CA with the object

Although the data are somewhat limited and more work on other islands and on complementizer agreement in Limbum is needed, I have suggested that there is a connection between selective island permeability and complementizer agreement that rests on the assumption that SpecCP is reserved for $\varphi$-bearing constituents only. This means that non-$\varphi$-bearing constituents like PPs and VPs must rely on the phase-unlocking property of complementizer agreement to be able to leave the embedded clause. However, CPs that constitute islands, in contrast to other embedded CPs, do not show $\varphi$-agreement with the matrix subject. They either do not show any agreement at all, in which case non-$\varphi$-bearing constituents are trapped inside the CP-phase, or they exhibit agreement with the direct object. In this case, PPs and VPs also get trapped inside the CP-phase because this agreement comes too early in the derivation to unlock the CP-phase for extraction to matrix SpecvP.

5. Conclusions

Both Asante Twi and Limbum exhibit a pattern of $\overline{\mathrm{A}}$-movement where nominal elements may freely be extracted from inside what are commonly taken to be strong islands. In contrast, movement of VP and PP constituents from these configurations incurs an island violation. This pattern is reminiscent of the situation with strong islands in English, which under certain conditions allows for nominal gaps but may never contain gaps of PPs. However, an account of the licit nominal gaps as base-generated empty resumptive pronouns as suggested for English, which are not available for PPs and VPs, seems implausible. The dependencies in Asante Twi and Limbum also have properties of $\overline{\mathrm{A}}$-movement when the gap appears in an island. Moreover, both languages show overt grammatical resumptions subject to language-specific conditions, and at least in Asante Twi, the resumptives, be they inside or outside an island domain, behave like gaps for the purposes of several movement diagnostics. If one were to treat the island escapes as cases of repair by resumption nevertheless, one would have to assume that every movement underlyingly leaves a resumptive pronoun that satisfies island constraints at the point of their evaluation (either at PF, see Lasnik 2001; Merchant 2001; Pesetsky 1998; Shlonsky 1992, or in the syntax, Boeckx 2003; Klein 2017; Müller 2014) and subsequently undergoes PF-deletion to give the appearance of a gap on the surface.

As Korsah and Murphy (2020, 2024) suggest, this might be a viable analysis for Asante Twi. For the Limbum pattern, however, I have argued that it seems implausible, as overt resumptive pronouns have a much narrower distribution than in Asante Twi, and the required deletion rule would have to encompass several different contexts that do not form a natural class. Instead, I suggest that the permeability of islands for nominal elements is due to their $\varphi$-features. Starting from the assumption that syntactic islands are entirely absent in Limbum, I propose that only $\varphi$-bearing elements may pass through the left edge of a clause. As PPs and VPs do not carry $\varphi$-features, this effectively restricts clause-external movement to nominal elements in both island and non-island configurations. The availability of long-distance movement of VPs and PPs from the latter is then accounted for by the observation that these configurations show complementizer agreement with the matrix subject, which unlocks the CP phase and thereby allows movement to skip the CP edge. In the relevant island configurations, complementizer agreement is either entirely absent or takes place with the matrix object, which comes too early to unlock the CP phase. Hence, under this view, the Limbum pattern of selective island permeability is not so much one of unexpected extraction of nominal constituents from islands but rather one of exceptional extraction of non-$\varphi$-bearing elements from regular embedded clauses.

More research on islands and on complementizer agreement is required to determine whether this approach holds up to a larger and more diverse set of Limbum data or, indeed, to similar patterns of asymmetric extraction from islands and non-islands cross-linguistically.