On the Absence of Certain Island Effects in Mende

Abstract

The distinction between weak and strong islands has been extensively explored in the literature from both a descriptive and analytical perspective. In this paper, I document and analyze island constructions and constraints in Mende, an understudied Mande language spoken in Sierra Leone. Mende has both weak islands (left branch and wh-islands) and strong islands (adjunct clauses, sentential subjects, and coordinate structures). Intriguingly, it has a third class of islands, that I call mixed islands which show a subject–non-subject asymmetry in allowing for movement out of relative clauses, only when they modify the subject. As such, subject-modifying RCs cannot be classified as (strong/weak) islands in Mende. This is the first systematic work on islands and island constraints in the Mande language family, and, as such, it brings novel data from an understudied language family to bear on our understanding of A-bar dependencies and the study of island escape in African languages. It also calls into question a neat paradigm of cross-linguistic island constraints. Importantly, this work also lays down a baseline for future research on island constraints in the broader Mande language family. In order to discuss island constraints, this paper also lays out the first analysis of relative clauses in Mende, while integrating new research on the left periphery, focus constructions, and wh-constructions.

1. Introduction

The literature on resumptive pronouns (RPs) indicates that they can amnesty island violations in many languages (c.f. Christensen and Nyvad 2014; McCloskey 2017). Recent work on African languages, however, shows that numerous languages do not fit neatly within the traditional categories of islandhood (c.f. Wolof: Torrence 2005, 2012; Krachi: Torrence and Kandybowicz 2015; Asante Twi: Hein and Georgi 2020; Igbo: Georgi and Amaechi 2020; Shupamem: Schurr et al. 2024; Akan: Murphy and Korsah, forthcoming; and Ikpana: Kandybowicz et al. 2021, 2023). In this paper, I show that islandhood in Mende, an understudied OV Mande language spoken in Sierra Leone, challenges the current theory. While there are both weak and strong islands in the language (Smith 2023), I propose that relative clauses are Mixed Islands, as they show a subject–non-subject asymmetry in permitting extraction, only when they modify the subject. This asymmetry can be seen in (1) and (2).

In (1a), the subject is modified by the string-adjacent relative clause, while (1b) shows that wh-movement out of the relative clause to the matrix left periphery is permitted with the 3rd person plural RP ti (glossed 3pl.rp) surfacing in the relative clause.1 In (2a), the pre-verbal direct object is modified by an obligatorily stranded post-verbal relative clause. In this context, extraction out of the relative clause is blocked, even with the RP (2b).

(1)	a.	nyápú-í-sìà	[tí	netí-í-síà	vέlὲ-ngá]	Subject-modifying RC
		girl-def-pl	3pl.sm	net-def-pl	weave-prf
		tí	Mὲlí	gbáfà-ngà
		3pl.sm	Mary	insult-prf
		‘The girls who wove the nets insulted Mary.’
	b.	gbὲ-ngái	mìá	nyápú-í-sìà	[tí
		what-pl	foc.l	girl-def-pl	3pl.sm
		tíi	vέlὲ-ngá]	tí	Mὲlí	gbáfà-ngà
		3pl.rp	weave-prf	3pl.sm	Mary	insult-prf
		‘What are they that the girls who wove them insulted Mary?’ what is x, such that the girls who wove x insulted Mary?
(2)	a.	Mὲlí	ndúpú-í-sìà	{*[tí	nétí-í-sìà	Object-modifying RC
		Mary	child-def-pl	3pl.sm	net-def-pl
		vέlὲ-ngá]}	l$\mathsf{ɔ}́$  -ngá	{[tí	nétí-í-sìà	vὲlέ-ngá]}
		weave-prf	see-prf	3pl.sm	net-def-pl	weave-prf
		‘Mary saw the children who wove the nets.’
	b.	*gbέ-ngái	mìá	Mὲlí	ndúpú-í-sìà
		what-pl	foc.l	Mary	child-def-pl
		l$\mathsf{ɔ}́$  -ngá	[tí	tíi	vέlὲ-ngá]
		see-prf	3pl.sm	3pl.rp	weave-prf
		Intended: ‘What did Mary see the children who wove them?’ what is x, such that Mary saw the children who wove x?

In this paper, I investigate this distinction. I discuss the structure and distribution of relative clauses, investigating the clausal structure of Mende, while also considering the obligatory stranding of relative clauses below their head. I conclude by comparing the factors that influence the permissibility of extraction in Mende with the factors that influence extractability out of relative clauses in Mainland Scandinavian languages.

I show that extraction out of relative clauses is sanctioned only when the RC modifies the subject and is blocked for object-modifying RCs. As such, I refer to them as mixed islands, since their status as an island is dependent on the position of the relative clause within the matrix clause. When the relative clause CP is in a high position in the matrix clause (SpecExtraP_h), which immediately dominates SubjP, extraction is possible. On the other hand, when it is below the matrix verb in SpecExtraP_l, movement is blocked. Since wh-movement is sanctioned out of subject-modifying RCs in Mende, they cannot be classified as (strong/weak) islands.

To my knowledge, this is the first systematic work on islands and island constraints in the Mande language family. As such, it brings novel data from an understudied language family to bear on our understanding of A-bar dependencies. Similar to the other African languages noted above, it calls into question a neat paradigm of cross-linguistic island constraints. Importantly, this work also lays down a baseline for future research on island constraints in the broader Mande language family. In order to discuss island constraints, this paper also lays out the first analysis of relative clauses in Mende, while integrating new research on the left periphery, focus constructions, and wh-constructions (see Smith 2021b, 2023 for initial research on these topics).

The remainder of the paper is structured as follows: Section 2 introduces the language and previous research while also sketching relevant Mende syntax. Section 3 argues that a movement analysis, and not base generation, best captures the facts of A-bar constructions in Mende. Section 4 introduces and describes relative clause mixed islands. Section 5 is a conclusion.

2. Previous Research and Mende Syntax

Mende is part of the broader Mande language family spoken throughout west Africa and is classified as a Western Mande language, most similar to Loko and Bandi. The Mande language family is considered an early offshoot of the Niger-Congo family (Williamson and Blench 2000).

Mende is a Southwestern Mande language, with about 2.5 million speakers in Sierra Leone and Liberia. There are four main dialects: Kpa, Kɔɔ, Wanjaama, and Sewama with a high degree of lexical similarity (Eberhard et al. 2023). While most previous research has focused on the Kɔɔ and Kpa dialects, this research focuses on Sewama Mende, a dialect that has been relatively unstudied. The data were gathered both in person and over Zoom, working with native speakers in Bo, Sierra Leone.

A number of Mende grammars have been produced over the years, including work by Schön (1884), Migeod (1908), Aginsky (1935), Crosby and Ward (1944), Innes (1961, 1967, 1969), Spears (1967), and Brown (1982). Most previous research in the language focuses on tone (c.f. Dwyer 1971, 1978; Leben 1973, 1978; and Goldsmith 1976) and word-initial consonant mutation (c.f. Dwyer 1969; Conteh et al. 1983, 1986). In my investigation of Sewama Mende, its tone does not align with that reported in the literature. I have marked surface tone throughout. I will indicate contexts in which word-initial consonant mutation occurs when relevant to the discussion.

Very little syntactic analysis has been developed for Mende. Sengova (1981) considers tense and aspect in the language and includes a good deal of syntactic description to lay out his arguments regarding the connection between syntax and the semantics of tense and aspect. Other syntactic analyses include Smith (2023, forthcoming a, forthcoming b).

2.1. Mende Syntax

Typical of Mande languages (Vai: Welmers 1976; Lorma: Dwyer 1981; Wan: Nikitina 2009, 2011, 2012, 2019; Bambara: Fofana and Traoré 2003; Mandinka: Creissels 2024), Mende has SOVX word order. In this remainder of Section 2, I move beyond this generalization and describe the clausal structure of Mende in more detail.

(3)	S(ubject)	SM	O(bject)	V(erb)	X
	nyápú-í-sìà	tí	mángú-í-sìà	màjíá-ngá	sùkú-í-sìà	hùn	gbóì
	girl-def-pl	3pl.sm	mango-def-pl	sell-prf	school-def-pl	in	yesterday
	‘The girls sold the mangoes at the schools yesterday.’

Mende does not have a noun classification system as seen in the forms of the plural nouns in (3), which show no noun class marking. Welmers (1971, p. 131) notes that this is true of the Mande languages more broadly.

In many Mande languages (Bearth 2009; Creissels 2019), including Mende (Innes 1967; Sengova 1981; Smith, forthcoming a) various clauses contain a subject marker that follows the subject in the TAM position agreeing with it in number and person, while also encoding habituality and negation, when present (glossed as hab for habitual constructions and sm (subject marker) for non-habitual constructions).2 I provide a more detailed analysis in Section 2.3, but for now note the contrast between the subject markers in (3) and (4). While the 3rd person plural subject marker for the past tense in (3) is ti, it is ta for the present/habitual tense in (4a). The 3rd person singular subject marker for the habitual tense in (4b) surfaces as a, pointing towards a polymorphemic structure.

(4)	a.	nyápú-í-sìà	t.á	mángú-í-sìà	màjíà	l$\mathsf{ɔ}̀$	sùkú-í-sìà	fóló	gbí	Habitual
		girl-def-pl	3pl.hab	mango-def-pl	sell	nm	school-def-pl	day	all
		‘The girls sell the mangoes at the school every day.’
	b.	nyápú-í	ø.á	mángú-í-sìà	màjíà	l$\mathsf{ɔ}̀$	sùkú-í-sìà	fóló	gbí
		girl-def	3sg.hab	mango-def-pl	sell	nm	school-def-pl	day	all
		‘The girl sells the mangoes at the schools every day.’

Negation is marked immediately after the subject marker.

(5)	S	SM	NEG	O	V-TNS	X
	nyápú-í-sìà	tí	ì	mángú-í-sìà	màjíà-ní	sùkú-í-sìà	hùn	gbóì
	girl-def-pl	3pl.sm	neg	mango-def-pl	sell-pst	school-def-pl	in	yesterday
	‘The girls did not sell the mangoes at the schools yesterday.’

Tense is marked as a suffix on the verb, as seen in comparing the past tense marker—(n)i in (3) with the future marker ma in (6).3

(6)	S	SM	O	V-TNS		X
	nyápú-í-sìà	tí	mángú-í-sìà	màjíà-má	á	sùkú-í-sìà	hùn	síná
	girl-def-pl	3pl.fut	mango-def-pl	sell-fut	nm	school-def-pl	in	tomorrow
	‘The girls will not sell the mangoes at the schools tomorrow.’

Anticipating a more detailed discussion in Section 3.1, I highlight three types of focus in the language here. Mende has two mutually exclusive focus markers: mia (glossed as foc.l), which occurs in the left periphery and lɔ which surfaces in TP. The marker lɔ occurs in two distinct contexts—it can either mark a narrow focus within the clause (glossed as foc.i) or surface as part of the verbal complex (glossed as nm—neutral marker). When mia is used, lɔ cannot surface in the clause, nor can lɔ surface twice. In contexts where lɔ is used, it can surface as a lengthening of the preceding word’s final vowel, as seen in (6) where the a following the future marker -ma is a manifestation of lɔ. While they seem to be related, I set aside questions related to the connection between lɔ as an in situ focus and lɔ that occurs within the verbal complex as a topic for future research.4

(7)	a.	Kpànâ	nyápú-í-sìà	làtù-(n)í	*(l$\mathsf{ɔ}\mathbf{̀}$  )		Neutral focus
		Kpana	girl-def-pl	praise-pst	nm
		‘Kpana praised the girls.’
	b.	nyápú-í-sìà	mìà	Kpànâ	tíi	làtù-ngá	Left peripheral focus
		girl-def-pl	foc.l	Kpana	3pl.rp	praise-prf
		‘It is the girls that Kpana praised.’
	c.	Kpànâ	nyápú-í-sìà	l$\mathsf{ɔ}\mathbf{̀}$	làtù-ngá		In situ focus
		Kpana	girl-def-pl	foc.i	praise-prf
		‘Kpana praised the girls.’

Similar to most Mande languages (c.f. Mahou: Koopman 1984; Wan: Nikitina 2012; Jalkunan: Heath 2017), Mende is postpositional.

(8)	Kpànâ	nyápú-í-sìà	làtù-í	l$\mathsf{ɔ}̀$	sùkú-í-sìà	hùn
	Kpana	girl-def-pl	praise-pst	nm	school-def-pl	in
	‘Kpana praised the girls in the schools.’

In the following sections, I review in more detail some foundational aspects of Mende syntax that lay the groundwork for the ensuing discussion on island constructions and constraints in Mende.

2.2. Verbal Complements

In this section, I motivate an analysis of Mende clauses, working upwards from VP.

The Mande languages have traditionally been classified as having a strict SOVX order (Gensler 1994; Nikitina 2012; Creissels 2024). In Mende, the subject and object precede the verb with the dative object (encoded in a PP) and adjuncts following it.

(9)	S	O	V	X
	Kpànâ	nìké-í-sìà	g$\mathsf{ɔ}́$  k$\mathsf{ɔ}̀$  -ngá	kpáá	hùn	gbóì
	Kpana	cow-def-pl	find-prf	farm	on	yesterday
	‘Kpana found the cows on the farm yesterday.’

(10)	S	O	V	Dat		X
	Kpànâ	nèsí-í-sìà	vè-ngá	Mὲlí	wέ	nj$\mathsf{ɔ}̀$  p$\mathsf{ɔ}́$  wà	hùn
	Kpana	pineapple-def-pl	give-prf	Mary	to	market	at
	‘Kpana gave the pineapples to Mary at the market.’

Following Kayne (1994), I argue that underlyingly Mende has a head-initial verb phrase and that the surface structure is derived via leftward movement (Smith 2021a). Similar to Aboh’s (2004) analysis for the Gbe languages, I argue that both the object and verb raise out of the vP. The verb head raises, through the functional head hosting the locative oblique, before adjoining the aspect (Asp) head, then the DP direct object raises into a higher position for Case (c.f. Koopman (1984, 1992) for the Mande languages Mahou and Bambara respectively.)5 Specifically, I argue that this position is the specifier of a Kase head (c.f. Major and Torrence, forthcoming for a similar analysis of Avatime).

(11)	a.	Kpànâ	kέyὲpὲ-í-sìà	mέnì-ngá	sùkú-í-sìà	hùn
		Kpana	rumor-def-pl	hear-prf	school-def-pl	at
		‘Kpana heard the rumors at school.’
	b.	Deriving SOV

In contrast to DP objects of verbs, CP objects do not raise into a pre-verbal Case position (along the lines of Stowell’s (1981) Case Resistance Principle). CP objects instead surface in an ‘extraposed’ position, a construction which has been documented in a number of languages (Major and Torrence, forthcoming for Avatime, Aboh (2004) for Gungbe, Zwart (1997) for Dutch).

(12)	CP objects
	Kpànâ	{hùngέ-ngá}	[kέ	nyápú-í-sìà	tí	mángú-í-sìà	yéyà-ngá]	{*hungɛ-nga}
	Kpana	explain-prf	c	girl-def-pl	3pl.sm	mango-def-pl	buy-prf	explain-prf
	‘Kpana explained that the girls bought the mangoes.’

CP complements of the verb can raise, however, it is to a position below the surface position of the verb, as suggested by (13). Following Cinque (1999), I suggest that the celerative adverb floflo ‘quickly’ is in a fixed position above the verb phrase. In (13a), both the DP direct object and verb must raise into a position above the adverb, while in (13b) the verb obligatorily raises out of vP above the adverb, while the CP can optionally do so. My language consultant has confirmed that floflo ‘quickly’ in (b) modifies the matrix verb, not the embedded verb.

(13)	a.	S		Odp		V	ADV
		Kpànâ	{*floflo}	nd$\mathsf{ɔ}̀$  mí-í-sìà	{*floflo}	hùngέ-ngá	{flófló}
		Kpana	quickly	story-def-pl	quickly	explain-prf	quickly
		‘Kpana quickly explained the story.’
	b.	S	V	ADV	Ocp
		Kpànâ	hùngέ-ngá	{flófló}	[kέ	nyápú-í-sìà	tí
		Kpana	explain-prf	quickly	c	girl-def-pl	3pl.sm
				ADV
		mángú-í-sìà	yéyà-ngá]	{flófló}
		mango-def-pl	buy-prf	quickly
		‘Kpana quickly explained that the girls bought the mangoes.’

I sketch this derivation out in (14). The verb hungɛ ‘explain’ can take either a DP or CP object as its complement. The verb raises out of vP, head moving to the aspect marker. If the direct object is a DP, it raises into SpecKaseP (indicated by the solid-line arrow), but if it is a CP, it can either remain in situ, or raise. If it raises, an extraposition phrase (ExtraP_L) with a null head merges above vP, and the CP modifier moves into its specifier (indicated by the dotted arrow).6

(14)	DP/CP raising

2.3. Subject Markers

As noted above, Mende subjects are followed by an obligatory subject marker. In (15a), the 3rd person plural subject marker is ta, and in (15b), the 3rd person singular is a.

(15)	a.	nyápú-í-sìà	t.á	Kpànâ	làtú	l$\mathsf{ɔ}̀$	Habitual SM
		girl-def-pl	3pl.hab	Kpana	praise	nm
		‘The girls praise Kpana.’
	b.	nyápú-í	ø.á	Kpànâ	làtú	l$\mathsf{ɔ}̀$
		girl-def	3sg.hab	Kpana	praise	nm
		‘The girl praises Kpana.’

In past tense constructions, however, the 3rd person plural subject marker surfaces as ti, while the singular counterpart is null.

(16)	a.	nyápú-í-sìà	tí	Kpànâ	làtú-í	l$\mathsf{ɔ}̀$	Past tense SM
		girl-def-pl	3pl.sm	Kpana	praise-pst	nm
		‘The girls praised Kpana.’
	b.	nyápú-í	ø	Kpànâ	làtú-í	l$\mathsf{ɔ}̀$
		girl-def	3sg.sm	Kpana	praise-pst	nm
		‘The girl praised Kpana.’

In light of these data, I suggest that subject markers are polymorphemic. Even though these subject markers have traditionally been orthographically written as a unit, they are the surfacing of a series of heads at the top of the middlefield. Syntactically, subject markers are structured as in the tree in (17), though I use the SMP (subject marker phrase) shorthand through the rest of the paper. The [t] encodes 3rd person plural agreement with the subject, which is triggered when the subject moves through SpecSubjP, while [a] encodes habitual aspect. The subject of the clause moves into a higher position, namely, SpecFinP (Rizzi 1997, 2001; Cardinaletti 1997; Smith, forthcoming a).

(17)	The articulated structure of a Mende subject marker

2.4. The Left Periphery

In this section, I consider in more detail the articulated structure of the left periphery in Mende, in which both topics and focused constituents can surface (Smith, forthcoming a).

(18)	a.	nyápú-í-sìà	tí		Kpànâ	làtú-ngá		sùkú-í-sìà	hùn
		girl-def-pl	3pl.sm		Kpana	praise-prf		school-def-pl	at
		‘The girls praised Kpana at the schools.’
	b.	Kpànâi	vá,		nyápú-í-sìà	tí		ngíi	TOP
		Kpana	for		girl-def-pl	3pl.sm		3sg
		làtú-ngà	sùkú-í-sìà		hùn
		praise-prf	school-def-pl		at
		‘As for Kpana, the girls praised him at the schools.’
	c.	Kpànâi	vá,	sùkú-í-sìàj		mìà	nyápú-í-sìà		TOP, FOC
		Kpana	for	school-def-pl		foc.l	girl-def-pl
		tí	ngíi	làtú-ngá		tìj	hùn
		3pl.sm	3sg.rp	praise-prf		3pl.rp	at
		‘As for Kpana, it is the schools that the girls praised him at.’
	d.	sùkú-í-sìàj	mìà,	Kpànâi		vá,	nyápú-í-sìà		FOC, TOP
		school-def-pl	foc.l	Kpana		for	girl-def-pl
		tí	ngíi	làtú-ngá		tíj	hún
		3pl.sm	3sg.rp	praise-prf		3pl.rp	at
		‘It is the schools that, as for Kpana, the girls have praised him at.’

Based on (18a), in (18b) the direct object Kpana surfaces in the left periphery as a topicalized constituent introduced by the postposition va ‘(as) for.’ In (18c), the topicalized constituent is followed by the focused constituent, while in (18d), the order of the topic and focused constituent changes, indicating that they are unordered in the left periphery.

Focus and topic constructions can also occur in embedded phrases where they follow the complementizer. The data in (19) show that both a topicalized and focused constituent can follow kɛ, the complementizer for an embedded statement.

(19)	Mende left periphery (ForceP > TopP/FocP > FinP)
			forceP	topP		focP
	Mὲlí	kítí-ngá	[kὲ	Kpànâj	vá,	sùkú-í-sìài	mìà
	Mary	doubt-prf	c	Kpana	for	school-def-pl	foc
	finP	tp
	nyápú-í-sìà	tí	ngíj	làtú-ngá	tíi	hún]
	girl-def-pl	3pl.sm	3sg.rp	praise-prf	3pl.rp	in
	‘Mary doubted that, as for Kpana, it is the schools in which the girls praised him.’

Following Rizzi (1997, 2001) in Smith, forthcoming a, I argue that kɛ heads a Force Phrase, taking the remainder of the clause as its complement and that mia heads a Focus Phrase, with the focused constituent moving into its specifier. Similarly, va heads a Topic Phrase, with the topicalized constituent moving into its specifier. In the previous section, I argued that the matrix subject moves into the specifier of a Finite Phrase. In (19), I have marked each of these phrases in the embedded clause.

3. Movement

3.1. Focus

Evidence for A-bar movement in Mende includes quantifier stranding and reconstruction effects. In (20a), the DP and quantifier are shown, while (20b) shows that they can surface in the left periphery and (20c) shows the quantifier stranded with the resumptive 3rd person plural pronoun ti surfacing in the canonical position. Following the reasoning in Sportiche (1988) and Fitzpatrick (2006), I take this as evidence that the DP in SpecFocP moved from its immediate preverbal position.

(20)		Stranded DP object quantifier
	a.	Kpànâ	lùmbé-í-sìà	kpέlέ	gáw$\mathsf{ɔ}̀$  -ngá
		Kpana	lemon-def-pl	all	peel-prf
		‘Kpana peeled all of the lemons.’
	b.	[lùmbé-í-sìà	kpέlέ]i	míá	Kpànâ	tii	gáw$\mathsf{ɔ}̀$  -ngá
		lemon-def-pl	all	foc.l	Kpana	3pl.rp	peel-prf
		‘It is all the lemons that Kpana peeled.’
	c.	lùmbé-í-sìài	mìà	Kpànâ	tíi	kpέlέ	gáw$\mathsf{ɔ}̀$  -ngá
		lemon-def-pl	foc.l	Kpana	3pl.rp	all	peel-prf
		‘It is all the lemons that Kpana peeled.’

In the remainder of this section, I use evidence from reconstruction effects to argue for a movement analysis, with resumptive pronouns functioning as traces of moved constituents. Reconstruction effects occur when a constituent surfaces in one part of the clause but acts as if it were in a lower part. I show two examples of these effects: Principle A binding and ideophone movement.

I turn first to Principle A binding. In (21a), the embedded statement has a DP subject Mary, which binds its reflexive object ta kpe ‘herself.’ In example (21b), the reflexive object surfaces in SpecFocP, where it cannot be bound by Mary. Since this sentence is grammatical and the reflexive is not bound in the left periphery, I conclude that it behaves as if it were in its canonical position, where it is bound by Mary.

(21)		Principle A binding
	a.	ndùpú-í-sìà	tí	ngí-ngá	{kὲ	Mὲlíi	[tà	kpé]i
		child-def-pl	3pl.sm	remember-prf	c	Mary	3sg	self
		l$\mathsf{ɔ}́$  -ngá	mὲmὲ	hùn}
		see-prf	mirror	in
		‘The children remembered that Mary saw herself in the mirror.’
	b.	[tà  kpé]i	mìà	ndùpú-í-sìà	tí	ngí-ngá
		3sg self	foc.l	child-def-pl	3pl.sm	remember-prf
		{kὲ	Mὲlíi	ngíi	l$\mathsf{ɔ}́$  -ngá	mὲmὲ	hùn}
		c	Mary	3sg.rp	see-prf	mirror	in
		‘It is herself that the children remembered that Mary saw in the mirror.’

A second example of a reconstruction effect concerns movement of an ideophone. Ideophones are relatively common in Niger-Congo languages and have been described as vivid sensory words (Dingemanse 2018; Downing 2019). Following Tamba et al. (2012) and Torrence (2013), I analyze ideophones as similar to adverbs in modifying a verb, with there being a strong selectional relationship between the ideophone and the verb, such that ideophones cannot appear with any verb, but only a few, or even just one. A surface discontinuity, therefore, is a result of movement. The following data show two ideophones: kpe which describes ‘a cut all the way through something’ and fikifiki which describes ‘a back-and-forth sawing motion.’ In (22a), the verb lewe ‘cut’ is modified by the ideophone kpe. In (22b), the verb bɔ ‘shoot’ is used. While in English something can be described as having been shot ‘clean through’ (e.g., ‘He was shot clean through the leg.’), kpe cannot be used with the verb bɔ ‘shoot’ in Mende (22c). Note that given the close selectional relationship, the verb and ideophone are string adjacent.

(22)	a.	Pìtá	nèsí-í-sìà	lèwè-ngá	kpé	Ideophone
		Peter	pineapple-def-pl	cut-prf	clean.through
		‘Peter cut the pineapple clean through.’
	b.	Pìtá	k$\mathsf{ɔ}́$  lí-í	b$\mathsf{ɔ}̀$  -ngá
		Peter	leopard-def	shoot-prf
		‘Peter shot the leopard.’
	c.	*Pìtá	k$\mathsf{ɔ}́$  lí-í	b$\mathsf{ɔ}̀$  -ngá	kpé
		Peter	leopard-def	shoot-prf	clean.through
		Intended: ‘Peter shot the leopard clean through.’

Based on (23a), in (23b) the verb lewe ‘cut’ is modified by the ideophone fikifiki, indicating the cutting motion as being ‘back and forth.’ In (23c), the ideophone surfaces in the left periphery, while the verb remains in its canonical position. The most straightforward explanation for this separation of the ideophone from the verb it modifies is that the ideophone has moved from its canonical post-verbal position to the left periphery, from which it still modifies the verbal action.

(23)	a.	Pìtá	mbèké-í-síà		lèwè-ngá
		Peter	branch-def-pl		cut-prf
		‘Peter cut the branches.’
	b.	Pìtá	mbèké-í-síà		lèwè-ngá	fíkífìkì
		Peter	branch-def-pl		cut-prf	sawing.motion
		‘Peter cut the branches with a sawing motion.’
	c	fíkífíkí	mìà	Pìtá	mbèké-í-síà	lèwè-ngá	Ideophone movement
		sawing.motion	foc.l	Peter	branch-def-pl	cut-prf
		‘It is with a sawing motion that Peter cut the branches.’

3.2. Wh-Questions

Wh-questions in Mende are structured similarly to focus constructions in that they can occur in situ or in the left periphery, they require a focus particle (lɔ or mia, respectively), and they utilize RPs in movement constructions. In this section, I briefly introduce relevant wh-constructions. For further description and analysis, see Smith (2021b).

Table 1. Wh-expressions in Mende.

Mende Interrogative	Translation	Resumptive
ye (-ni)	‘who (PL)’	ngi (sg)/ti (pl)
gbɛ (-nga)	‘what (PL)’	ø (sg)/ti (pl)
mi-ndo (-nga)	‘where (PL)’	na
mi-gbe (-nga)	‘when (PL)’
gbɛ-va	‘why’
ye	‘how’
ye (nda) Xindef	‘whose’	ø (sg)/ti (pl)
Xdef ye-gbɛ	‘which’	ø (sg)/ti (pl)
Xdef/indef lolɛ	‘how much / many’

is a summary of Mende wh-phrases, an indication of which can be marked as plural, as well as the corresponding resumptive pronouns.

Questions in Mende can be formed either in situ or in the left periphery. Based on (24a), in (24b) the subject has been transformed into the wh-word yɔ-ɔ, which is derived from a phonological process that occurs when ye ‘who’ is focus marked by lɔ.7 In (24c), an in situ construction, the direct object surfaces as plural-marked wh-word gbɛ-nga-a ‘what (pl),’ with the focus marker lɔ, surfacing as a lengthening of the morpheme final [a]. In (24d), it is shown that the wh-word can also surface in the left periphery.

(24)		Argument wh-questions
	a.	ndùpú-í	nìké-í-síà	g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}̀$  -ngá
		child-def	cow-def-pl	find-prf
		‘The child found the cows.’
	b.	y$\mathsf{ɔ}\mathbf{̀}$  -$\mathsf{ɔ}̀$	nìké-í-síà	g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}̀$  -ngá				Wh-subject
		who-foc.i	cow-def-pl	find-prf
		‘Who found the cows?’
	c.	ndùpú-í	gbὲ-ngá-á	g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}̀$  -ngá				In situ object
		child-def	what-pl-foc.i	find-prf
		‘What did the child find?’
	d.	gbὲ-ngái	mìà	ndùpú-í	tíi	g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}̀$  -ngá		Fronted object
		what-pl	foc.l	child-def	3pl.rp	find-prf
		‘What did the child find?’

Turning next to adjuncts, in (25a) the locative phrase njɔpɔwa hun ‘at the market’ and temporal phrase gboi ‘yesterday’ both occur post-verbally. In (25b), a locative wh-phrase mindo ‘where’ focused in situ is shown, while (25c) shows that the wh-phrase can surface in the left periphery. In (25d), it is shown that the temporal adverb can be transformed into the wh-phrase migbe ‘when’, with (25e) showing that migbe can move to the left periphery. Note that there is no resumptive pronoun that surfaces for migbe.

(25)		Adjunct wh-questions
	a.	Pìtá	mángú-í-sìà	mὲ-ngá	nj$\mathsf{ɔ}̀$  p$\mathsf{ɔ}̀$  wá
		Peter	mango-def-pl	eat-prf	market
		hùn	gbóí
		at	yesterday
		‘Peter ate the mangoes at the market yesterday.’
	b.	Pìtá	mángú-í-sìà	mὲ-ngá	míndò		In situ adjunct
		Peter	mango-def-pl	eat-prf	where
		l$\mathsf{ɔ}̀$	gbóí
		foc.i	yesterday
		‘Where did Peter eat the mangoes yesterday?’
	c.	míndói	mìà	Pìtá	mángú-í-sìà	mὲ-ngá	Fronted adjunct
		where	foc.l	Peter	mango-def-pl	eat-prf
		nài	gbóí
		loc.rp	yesterday
		‘Where is it that Peter ate the mangoes yesterday?’
	d.	Pìtá	mángú-í-sìà	mὲ-ngá	nj$\mathsf{ɔ}̀$  p$\mathsf{ɔ}̀$  wá	hùn	In situ adjunct
		Peter	mango-def-pl	eat-prf	market	at
		mígbè	l$\mathsf{ɔ}̀$
		when	foc.i
		‘When did Peter eat the mangoes at the market?’
	e.	mígbè	mìà	Pìtá	mángú-í-sìà	mὲ-ngá	Fronted adjunct
		when	foc.l	Peter	mango-def-pl	eat-prf
		nj$\mathsf{ɔ}̀$  p$\mathsf{ɔ}̀$  wá	hùn
		market	at
		‘When is it that Peter ate the mangoes at the market?’

In multi-clause wh-constructions, Mende permits in situ, partial movement, and full movement. In (26a), the verb mɛni ‘hear’ takes a CP complement. In (26b), the wh-phrase surfaces in situ in the embedded clause, while in (26c) it partially moves to the left periphery of the embedded clause. In both (26b-c), the wh-phrase in the embedded clause has a matrix scope. Full movement is shown in (26d).

(26)		Multi-clause Wh-movement
	a.	Kpànâ	mὲnì-ngá	[kὲ	nyàpú-í-sìà	tí	Embedded clause
		Kpana	hear-prf	c	girl-def-pl	3pl.sm
		mángú-í-sìà	yèyà-ngá]
		mango-def-pl	buy-prf
		‘Kpana heard that the girls bought the mangoes.’
	b.	Kpànâ	mὲnì-ngá	[kὲ	nyàpú-í-sìà	tí	Wh-In situ
		Kpana	hear-prf	c	girl-def-pl	3pl.sm
		gbὲ-ngá-á	yèyà-ngá]
		what-pl-foc.l	buy-prf
		‘What (PL) did Kpana hear that the girls bought?’
	c.	Kpànâ	mὲnì-ngá	[kὲ	gbὲ-ngái	mìà	Partial movement
		Kpana	hear-prf	c	what-pl	foc.l
		nyàpú-í-sìà	tí	tíi	yèyà-ngá]
		girl-def-pl	3pl.sm	3pl.rp	buy-prf
		‘What (PL) did Kpana hear that it is that the girls bought?’
	d.	gbὲ-ngái	mìà	Kpànâ	mὲnì-ngá	[kὲ	Full movement
		what-pl	foc.l	Kpana	hear-prf	c
		nyápú-í-sìà	tí	tíi	yèyà-ngá]
		girl-def-pl	3pl.sm	3pl.rp	buy-prf
		‘What (PL) is it that Kpana heard that the girls bought?’

I turn next to relative clauses and their status as mixed islands.

4. Mixed Islands

The distinction between weak and strong islands has been extensively explored in the literature from both a descriptive and analytical perspective (Ross 1967; Chomsky 1977; Szabolcsi and den Dikken 2002; Szabolcsi and Lohndal 2017). In this section, I discuss relative clause islands, which do not neatly fit into either category. Instead, I refer to them as mixed islands, because, as I show, there is a subject–object extraction asymmetry. Before considering these constructions, I first lay the groundwork by describing the structure of DPs. I then look at relative clauses and their distribution before considering their status as islands.

4.1. Mende DP Structure

The noun is the leftmost element in Mende DPs, with number and (in)definite markers following (27a). Adjectives precede numerals, and both precede the definite and number markers (27b-c). Demonstratives surface between the definite marker and number (27d). I assume minimally that NPs raise into SpecDP (Ritter 1991). I further assume that adjectives are in a fixed hierarchy above the NP, in the specifier of Functional Phrases (Sproat and Shih 1990; Cinque 1994). The surface structure of a Mende DP is shown in (27e).

(27)	a.	nyàpú-í-sìà				n-def-pl
		girl-def-pl
		‘the girls’
	b.	nyàpú	gùtú-í-sìà			n-adj-def-pl
		girl	short-def-pl
		‘The short girls.’
	c.	nyàpú	gùtú	w$\mathsf{ɔ}́$  tέ-í-sìà		n-adj-numb-def-pl
		girl	short	six-def-pl
		‘The six short girls.’
	d.	nyàpú	gùtú	w$\mathsf{ɔ}́$  tέ-í	ná-sìà	n-adj-numb-def-dem-pl
		girl	short	six-def	dem-pl
		‘Those six short girls’
	e.	N-(Adj)-Det-(Dem)-Num

I propose that the order in (40d) is derived as follows in (41). Immediately above the NP is a functional phrase (FP₁), which hosts the adjective in its specifier (Cinque 1993; Crisma 1993). Above it is FP₂, into whose specifier the NP raises.8 The NP raises again, pied-piping FP₂ with it, into the specifier of the Numeral Phrase. The entire Numeral Phrase then raises into the SpecNumP, before raising again into SpecDemP, and finally into SpecDP, yielding the surface structure.9

(28)	Complex NP raising to SpecDP

4.2. Relative Clauses in Mende

As noted in the introduction, this is the first analysis of relative clauses in Mende in the generative tradition of which I am aware.10 As such, the following sections provide a detailed description and analysis of their structure. I analyze relative clauses in Mende under a Kaynian (D+CP) approach (Kayne 1994), arguing that the head of the relative clause originates inside of a CP complement to a head in the DP functional structure, and raises to SpecDP. As I show below, this is strikingly similar to what is found in ordinary DPs.

In (29a), a relativized subject nyapuisia ‘the girls’ as the head of the relative clause is shown. The object and verb follow, with temporal and locative adjuncts occurring post-verbally. Mende RCs do not have a relative pronoun, and relativized subjects do not have a resumptive pronoun. Differing from a matrix clause, it is ungrammatical for a neutral marker (lɔ) to surface on the verb. In (29b), a relativized object is shown, which precedes the clause. In contrast to relativized subject in (29a), the 3rd person plural resumptive pronoun ti surfaces in the object position, resuming the object tɛisia ‘the chickens.’ The verb is marked for past tense, but, again, no neutral marker can appear.11

(29)	a.	Subject Relativization
		nyàpú-í-sìài	[ti	tá	tέ-í-sìà		màjíá
		girl-def-pl		3pl.hab	chicken-def-pl		sell
		(*l$\mathsf{ɔ}̀$  )	wàtì	gbí	nj$\mathsf{ɔ}̀$  p$\mathsf{ɔ}̀$  wá		hùn]
		nm	time	all	market		at
		‘The girls who always sell the chickens at the market.’
	b.	Object Relativization
		tέ-í-sìàj		[nyàpú-í-sìà	tí	tíj	yèyá-nì
		chicken-def-pl		girl-def-pl	3pl.sm	3pl.rp	buy-prf
		(*l$\mathsf{ɔ}̀$  )		nj$\mathsf{ɔ}̀$  p$\mathsf{ɔ}̀$  wá	hùn]
		nm		market	at
		‘The chickens that the girls bought at the market.’

In (30), the parallel structure between a relativized subject (a), relativized object (b), and ordinary DP (c) is shown.

(30)		Structure of relative clauses
	a.	srel	def	num	[__	sm	do	v	tns	x]
	b.	orel	def	num	[s	sm	rp	v	tns	x]
	c.	np	def	num	[__	]

Having argued that the head of the relative clause begins within the clause, I next show that the relative clause is a CP. Evidence for this includes the presence of topic and focus constructions in the left periphery of the relative clause. In (31a), the head of the relative clause is the subject of the matrix clause, while (31b) is a topic construction and (31c) is a focus construction. In Section 2.4, I argued that both topics and focus constructions occur in the left periphery, and these data show that left peripheral constructions can occur in a relative clause. We can conclude, therefore, that the relative clause is, in fact, a CP.

(31)		Relative clause left periphery
	a.	nyàpú-í-sìài	[Kpànâ	tìi	l$\mathsf{ɔ}́$  -nì	kpàà	hùn
		girl-def-pl	Kpana	3pl.rp	see-pst	farm	on
		gbóí]	tí	mbè-í	m$\mathsf{ɔ}̀$  l$\mathsf{ɔ}́$  -ì	l$\mathsf{ɔ}̀$
		yesterday	3pl.sm	rice-def	burn-pst	nm
		‘The girls that Kpana saw yesterday on the farm burned the rice.’
	b.	nyàpú-í-sìài,	[gbóí	vá,	Kpànâ	tíi	l$\mathsf{ɔ}́$  -nì
		girl-def-pl	yesterday	for	Kpana	3pl.rp	see-prf
		kpàà	hùn]	tí	mbè-í	m$\mathsf{ɔ}̀$  l$\mathsf{ɔ}́$  -ì	l$\mathsf{ɔ}̀$
		farm	on	3pl.sm	rice-def	burn-pst	nm
		‘The girls, as for yesterday, Kpana saw them on the farm (they) burned the rice.’
	c.	nyàpú-í-sìài	[kpáá	hùn	mìà	Kpànâ	tìi
		girl-def-pl	farm	on	foc.l	Kpana	3pl.rp
		l$\mathsf{ɔ}́$  -nì	nà	gbóí]	tí	mbè-í	m$\mathsf{ɔ}̀$  l$\mathsf{ɔ}́$  -ì	l$\mathsf{ɔ}̀$
		see-pst	loc	yesterday	3pl.sm	rice-def	burn-pst	nm
		‘The girls, it is on the farm that Kpana saw them yesterday (they) burned the rice.’

4.2.1. The Derivation of Relative Clauses

The derivation of a relative clause is set out in (45). The relativized nominal (‘NP’) begins within the CP. The NP raises to SpecCP, and it subsequently raises into the specifier of a relative phrase (RelP) (as proposed by Collins 2015). Paralleling the structure laid out in (28), I suggest, instead, that it is the Num head that selects the Relative Phrase with the NP raising into the DP left periphery, via SpecNumP before surfacing in SpecDP. In the remainder of this section, I motivate this analysis.

(32)	Derivation of a relative clause

I want to briefly point out that the tree in (32) does not capture the full derivation. Crucially, under this analysis the DP head of the relative clause is not a constituent. I propose that relative clause constructions in Mende are derived when the CP portion of the clause raises into the specifier of a functional phrase that I call an ExtraP (see Section 2.2 for a similar argument for CP complements). The DP remnant then raises into a higher position. This can be seen in the following data, where in (33a) the object appears in a pre-verbal position, with the CP portion remaining post-verbal. Though not as clearly evident, I propose that a similar process yields the word order in (33b). I discuss ‘extraposition’ in greater detail in the following section.

(33)

a.

Object-modifying RC

Kpànâ

njè-í-sìài

màjìà-í

l$\mathsf{ɔ}̀$

[ti

tí

ngì

lέ-ngá

gbálè-nì]

Kpana

goat-def-pl

sell-pst

nm

3pl.sm

3sg

chicken-pl

hurt-pst

‘Kpana sold the goats that hurt his chickens.’

b.

Subject-modifying RC

nyàpú-í-sìài

[ti

tí

njé-í-sìà

g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}́$  -ní]

tí

káté-í

kpàyà-í

l$\mathsf{ɔ}̀$

girl-def-pl

3pl.sm

goat-def-pl

find-pst

3pl.sm

fence-def

strengthen-pst

nm

‘The girls who found the goats strengthened the fence.’

I turn next to reconstruction effects to argue that the head of the relative clause originates within the clause. The first evidence that I consider is reflexive binding. In (34a) and (34b), the DO is bound by the subject, with the object in (34a) being non-reflexive, while the object in (34b) is reflexive. In both cases, it is a possessive construction. In (34c), the nominal constituent containing the reflexive ta kpe ‘himself’ is the head of the relative clause and has raised into a pre-verbal position, with the relative clause in a post-verbal position. In this position, ta kpe ‘himself’ can take either Kpana or John as antecedent. Since ta kpe can be understood as referring to John, we can conclude that it reconstructs into the relative clause. In other words, in regards to binding, it behaves as if it were in its canonical position between the subject and verb of the relative clause. In (34d), the reflexive surfaces in the left periphery of the matrix clause, but can reconstruct into the c-command domain of John (as indicated by the subscripts).

(34)		Reconstruction effects: reflexive binding
	a.	J$\mathsf{ɔ}́$  ni	[ngìj/Kpànâj		nὲnέ-í-sìà		gbá	kòló	má]	ndàlá-í	l$\mathsf{ɔ}̀$
		John	3sg/Kpana		shadow-def-pl		stuck	paper	on	draw-pst	nm
		‘John drew pictures of him/Kpana’
	b.	J$\mathsf{ɔ}́$  ni	{[tá	kpé]i	nὲnέ-í-sìà		gbá
		John	3sg	self	shadow-def-pl		stuck
		kòló	má}	ndàlá-í	l$\mathsf{ɔ}̀$
		paper	on	draw-pst	nm
		‘John drew pictures of himself.’
	c.	Kpànâj	{[tá	kpé]i/j	nὲnέ-í-sìà		gbá		kòlò
		Kpana	3sg	self	shadow-def-pl		stuck		paper
		má}k	l$\mathsf{ɔ}̀$  -í	l$\mathsf{ɔ}̀$	[J$\mathsf{ɔ}́$  ni		tìk		ndálá-nì
		on	see-pst	nm	John		3pl.rp		draw-pst
		‘Kpana saw pictures of himself that John drew.’
	d.	{[tá	kpé]i/j	nέnέ-í-sìà		gbá	kòlò		mà}k
		3sg	self	shadow-def-pl		stuck	paper		on
		mìà	Kpànâj	tìk		l$\mathsf{ɔ}̀$  -nì	[J$\mathsf{ɔ}́$  ni		tìk	ndálá-nì]
		foc.l	Kpana	3pl.rp		see-pst	John		3pl.rp	draw-pst
		‘It is pictures of himself that Kpana saw that John drew.’

Further evidence for the promotion analysis comes from quantifier scope. In (35a), it is scopally ambiguous. Under the subject wide scope reading, every boy works on a possibly different book (∀ > ∃). When the object takes a wide scope, this corresponds to a situation in which there is a particular book that every boy worked on (∃ > ∀). In (35b), the direct object of the RC-internal verb has been relativized (and the RC extraposed so that it follows the matrix verb). Crucially, (35b) is still scopally ambiguous. Specifically, ‘book’ can reconstruct into a relative clause and scope under the RC-internal subject. (The wide scope reading of ‘book’ would be expected from its surface position). Given the evidence from binding and quantifier scope, I conclude that the promotion analysis is on the right track for Mende.

(35)		Quantifier scope
	a.	híndólópò	gbí	tí	k$\mathsf{ɔ}̀$  l$\mathsf{ɔ}́$	jèwé-í	l$\mathsf{ɔ}̀$
		boy	all	3pl.sm	book	write-pst	nm
		‘Every boy wrote a book.’ (√ A book x, such that every boy wrote x  ∃ > ∀ √ Every boy x wrote a possibly different book y  ∀  > ∃)
	b.	Kpànâ	k$\mathsf{ɔ}̀$  l$\mathsf{ɔ}́$	yèyá-í	l$\mathsf{ɔ}̀$	[híndólòpò	gbí	tí	sèwè-nì]
		Kpana	book	buy-pst	nm	boy	all	3pl.sm	write-pst
		‘Kpana bought a book that every boy wrote.’ (√ A book x, such that every boy wrote x  ∃ > ∀ √ Every boy x wrote a possibly different book y  ∀ > ∃)

4.2.2. Relative Clause Extraposition as Stranding

In order to more fully understand the process, in this section I explore the distribution of subject and object relative clauses in Mende. Crucially, I will argue that relative clauses always extrapose in Mende.

I begin by setting out in the following examples the position of object and subject- modifying relative clauses.12 In the object-modifying relative clauses in (36), the head of the RC functions as the direct object of the matrix clause. The DO/head of the RC surfaces pre-verbally, with the actual RC occurring post-verbally. In both (36a) and (36b), the subject marker in the RC agrees in number with the RC subject and in tense/aspect with the RC verb. In (36a), a subject relativizing clause, there is no resumptive pronoun for the head of the clause nyapuisia ‘the girls’, which surfaces as the clausal direct object above the verb, while in (36b), an object relativizing clause, the 3rd person plural resumptive pronoun ti surfaces in the canonical direct object position of the head of the RC.

(36)		Object-modifying RC
	a.	Kpànâ	nyàpú-í-sìài		màlé-í	l$\mathsf{ɔ}̀$	[ti
		Kpana	girl-def-pl		meet-pst	nm
		tá	tέ-í-sìà		màjíà	wátí	gbí]
		3pl.hab	chicken-def-pl		sell	time	all
		‘Kpana met the girls who always sell the chickens.’
	b.	Kpànâ	wátí	gbí	à	tέ-í-sìàj		vàwé	l$\mathsf{ɔ}̀$
		Kpana	time	all	3sg.hab	chicken-def-pl		disturb	nm
		[nyàpú-í-sìà	tí	tíj	yéyà-nì	nj$\mathsf{ɔ}̀$  p$\mathsf{ɔ}́$  wá		hùn]
		girl-def-pl	3pl.sm	3pl.rp	buy-pst	market		at
		‘Kpana always disturbs the chickens that the girls bought at the market.’

In (37a), a grammatical construction with a pre-verbal DP object and post-verbal modifying RC is shown, which I argue results from stranding. In (37b), it is shown that it is ungrammatical for the DP and RC to both surface pre-verbally, while (37c) shows that they cannot both follow the verb.

(37)

Stranded RC modifier of a DP object

a.

Pìtá

nìké-í-sìài

mὲní-í

l$\mathsf{ɔ}̀$

[nyàpú-í-sìà

tí

tíi

g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}̀$  -nì]

Peter

cow-def-pl

hear-pst

nm

girl-def-pl

3pl.sm

3pl.rp

find-pst

‘Peter heard the cows that the girls found.’

b.

*Pìtá

nìké-í-sìài

[nyàpú-í-sìà

tí

tíi

g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}̀$  -nì]

mὲní-í

l$\mathsf{ɔ}̀$

Peter

cow-def-pl

girl-def-pl

3pl.sm

3pl.rp

find-pst

hear-pst

nm

‘Peter heard the cows that the girls found.’

c.

*Pìtá

mὲní-í

l$\mathsf{ɔ}̀$

nike-i-siai

[nyàpú-í-sìà

tí

tíi

g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}̀$  -nì]

Peter

hear-pst

nm

cow-def-pl

girl-def-pl

3pl.sm

3pl.rp

find-pst

‘Peter heard the cows that the girls found.’

Consider next the data in (38), in which the Relative Clause CP can surface either in a position to the left or right of the aspectual adverb kpɔ ‘already.’ If the CP surfaces above the adverb, I assume that it has raised out of the verb phrase. In either case, the CP is stranded below the surface position of the verb.

(38)	CP raising
	Pìtá	nìké-í-sìài	mὲní-í	l$\mathsf{ɔ}̀$	{kp$\mathsf{ɔ}\mathbf{́}$  }	[nyàpú-í-sìà	tí	tíi	g$\mathsf{ɔ}̀$  k$\mathsf{ɔ}̀$  -nì]	{kp$\mathsf{ɔ}\mathbf{́}$  }
	Peter	cow-def-pl	hear-pst	nm	already	girl-def-pl	3pl.sm	3pl.rp	find-pst	already
	‘Peter already heard the cows that the girls found.’

The structure of an object-modifying relative clause is laid out in (39) with the source and surface position of the relativized constituent underlined. The head of the relative clause has raised into a pre-verbal position, while the relative clause CP is stranded.

(39)	Object-modifying relative clause
	S  SM  O  V  {X]  [RC]  {X}

In the following example, I suggest a derivation for how the surface structure of a direct object-modifying relative clause surfaces in Mende. Consider (40), in which the head of the relative clause nyapu ‘girl’ merges within the relative CP. It raises through SpecCP and SpecRelP into SpecNumP then SpecDP. The CP portion of the relative clause, which is a constituent, subsequently raises into the specifier of the Extraposition Phrase.13 I refer to this extraposition phrase above the vP as ExtraP_l. This leaves the derived DP to remnant movement, raising for Case into SpecKaseP.

(40)	a.	Kpànâ	nyàpú-í-sìài	màlé-í	l$\mathsf{ɔ}̀$	[ti	tá	tέ-í-sìà
		Kpana	girl-def-pl	meet-pst	nm		3pl.hab	chicken-def-pl
		màjíà	wátí	gbí]
		sell	time	all
		‘Kpana met the girls who always sell the chickens.’
	b.	DP and CP Raising from an Object-Modifying Relative Clause

This analysis aligns with Kayne’s (1994, p. 121) proposal that relative clauses remain stranded in a non-Case marked position below the normal Case-marked positions, with the verb raising above this position, as well (c.f. Bianchi 1999).

I turn next to subject-modifying relative clauses. In (41), the head of the RC nyapuisia ‘the girls’ functions as the matrix subject. The modifying RC follows the subject and is in turn followed by the subject marker of the matrix clause. The direct object, verb, and any post-verbal material follow. Note in (41a), with a subject relativizing clause, that the subject marker in the RC ta agrees with the matrix subject (the promoted subject of the RC) and encodes habitual aspect. The matrix subject marker ti surfaces after the relative clause and agrees with the matrix subject. In (41b), with an object relativizing clause, the subject marker of the RC ti agrees with the number of the RC subject and the tense of the verb, while the clausal subject marker ta agrees in number with the matrix subject and in habitual aspect with the matrix verb. As expected, the resumptive pronoun ti occurs in the direct object position in the relative clause from which the clausal subject raised.

(41)		Subject-modifying relative clauses
	a.	nyàpú-í-sìài	[ti	tá	tέ-í-sìà		màjíá	wátí
		girl-def-pl		3pl.hab	chicken-def-pl		sell	time
		gbí]	tí	Kpànâ	màlé-í		l$\mathsf{ɔ}̀$
		all	3pl.sm	Kpana	meet-pst		nm
		‘The girls who always sell the chickens met Kpana.’
	b.	tὲ-í-sìài	[nyàpú-í-sìà		tí	tíi	yèyà-nì	nj$\mathsf{ɔ}̀$  p$\mathsf{ɔ}́$  wá	hùn]
		chicken-def-pl	girl-def-pl		3pl.sm	3pl.rp	buy-pst	market	at
		tá	Kpànâ		vàwé	l$\mathsf{ɔ}̀$	wátí	gbí
		3pl.hab	Kpana		disturb	nm	time	all
		‘The chickens that the girls bought at the market always disturb Kpana.’

Similar to object-modifying relative clauses, the head of a subject-modifying relative clause raises out of the clause into a higher position. Being the matrix subject, I argue that it raises into SpecFinP. The surface structure of a subject-modifying relative clause is laid out in (42).

(42)	Subject-modifying relative clause
	S  [RC]  SM  O  V  {X}

In (43), it is shown that an adverb can intervene between the subject and the relative clause, and I conclude that the DP and relative CP have split.

(43)	Stranded RC modifier of subject
	nyàpú-í-sìài	{kp$\mathsf{ɔ}́$  }	[ti	tá	tὲ-í-sìà	màjíá	wátí
	girl-def-pl	already		3pl.hab	chicken-def-pl	sell	time
	gbí]	tí	Kpànâ	màlé-í	l$\mathsf{ɔ}̀$
	all	3pl.sm	Kpana	meet-pst	nm
	‘The girls who always sell the chickens (already) met Kpana.’

In this construction, the subject is in SpecFinP with the relative clause intervening between the subject nyapuisia and the matrix subject marker ti. I propose the following derivation (leaving out the adverb for simplicity). The DP subject (including the RC) raises from the vP into SpecSubjP, triggering agreement with the Subj head. At this point, similar to the object-modifying relative clause, the CP portion of the relative clause raises into an extraposition phrase (SpecExtraP_h). The DP subject then raises into SpecFinP, yielding the surface word order.

(44)	a.	nyàpú-í-sìài	[ti	tá	tέ-í-sìà	màjíá	wátí
		girl-def-pl		3pl.hab	chicken-def-pl	sell	time
		gbí]	tí	Kpànâ	màlé-í	l$\mathsf{ɔ}̀$
		all	3pl.sm	Kpana	meet-pst	nm
		‘The girls who always sell the chickens met Kpana.’
	b.	Derivation of subject-modifying RC

In summary, I have shown a parallel process for both object- and subject-modifying relative clauses. In both instances, the CP portion of the relative clause raises into an extraposition phrase, leaving the DP portion to raise into its surface position (SpecKaseP for the object and SpecFinP for the subject). Baltin (1981, 2006) argues that constituents extraposed from subjects adjoin to IP, while constituents extraposed from objects adjoin to VP. He argues that an extraposed phrase adjoins to the first maximal phrase that dominates its phrase of origin (Baltin 2006, p. 241). While my analysis does not support this assertion, it does seem that in Mende the extraposition phrases are adjoined at the positions for which he argues.

4.2.3. Relative Clause Mixed Islands

Having established the distribution of subject- and object-modifying relative clauses, I turn next to their status as mixed islands. Consider the subject-modifying relative clauses in (45), in which the head of the relative clause is promoted from within the clause to the matrix subject position. The DP (including the relative clause) raises to SpecSubjP, triggering agreement with the Subject Marker. The relative clause CP then raises into SpecExtraP_h, while the matrix subject raises into SpecFinP.

(45)	a.	nìké-í-sìài	[hùmám$\mathsf{ɔ}́$  -í	tìi	hùmá-ngá]	tí	lùgbá-ì	l$\mathsf{ɔ}̀$
		cow-def-pl	thief-def	3pl.sm	steal-prf	3pl.sm	stumble-pst	nm
		pὲlέ-í	hùn
		road-def	on
		‘The cows that the thief stole stumbled on the road.’
	b.	Subject-modifying RC

In this type of construction, the subject of the relative clause can be wh-questioned and move out of the relative clause and into the left peripheral focus position.

(46)	yèi	míà	nìké-í-sìàj	[ìi	tìj	hùmá-ngá]
	who	foc.l	cow-def-pl	3sg.rp	3pl.rp	steal-prf
	tíi	lùgbá-ní	pὲlὲ-í	hùn
	3pl.sm	stumble-pst	road-def	on
	‘Who is it that stole the cows that stumbled on the road?’

In (62), the wh-questioned subject of the relative clause moves into SpecFocP, with a resumptive pronoun surfacing in its pre-movement position. In this construction, the relative clause is not an island.

(47)	Movement out of subject-modifying RC

Consider next a subject-modifying RC (bracketed in (48a)) in an embedded clause (in curly brackets). In (48b), partial movement to the left periphery of the embedded clause is shown, while (48c) shows that wh-word can move to the clausal left periphery.

(48)		Embedded clause with subject-modifying RC: partial and full movement
	a.	Mὲlí		mὲní-í		{kὲ	nìké-í-sìài	[hùmám$\mathsf{ɔ}́$  -í	tíi
		Mary		hear-pst		c	cow-def-pl	thief-def	3pl.rp
		hùmá-ngá]		tí		lùgbá-ì	l$\mathsf{ɔ}̀$	pὲlὲ-í	hùn}
		steal-prf		3pl.sm		stumble-pst	nm	road-def	on
		‘Mary heard that the cows that the thief stole stumbled on the road.’
	b.	Mὲlí	mὲní-í		{kὲ	yèj	míà	nìké-í-sìài	[ìj
		Mary	hear-pst		c	who	foc.l	cow-def-pl	3sg.rp
		tíi	hùmá-ngá]		tí	lúgbà-nì	pὲlὲ-í	hùn}
		3pl.rp	steal-prf		3pl.sm	stumble-pst	road-def	on
		‘Mary heard that it was who that stole the cows that stumbled on the road?’
	c.	yèj	míà		Mὲlí	mὲní-í	{kὲ	nìké-í-sìài	[ìj
		who	foc.l		Mary	hear-pst	c	cow-def-pl	3sg.rp
		tìi	hùmá-ngá]		tí	lúgbà-nì	pὲlὲ-í	hùn}
		3pl.rp	steal-prf		3pl.sm	stumble-pst	road-def	on
		‘Who is x such that Mary heard that x stole the cows that stumbled on the road?’

In contrast to the subject-modifying relative clause, movement out of an object-modifying relative clause is not sanctioned. As noted above, the DP head of an object-modifying relative clause obligatorily moves into a pre-verbal position for Case assignment.14 The relative CP moves into SpecExtraP_l, remaining below the verb, as it does not need to raise for Case. This structure is shown in (49).

(49)	a.	Kpànâ	[kúlé-í-sìà]i	wúá-ì	l$\mathsf{ɔ}̀$	[màhé-í	tìi	yéyá-nì]
		Kpana	cloth-def-pl	wash-pst	nm	chief-def	3pl.rp	buy-pst
		‘Kpana washed the clothes that the chief bought.’
	b.	Object-modifying relative clause

In contrast to subject-modifying clauses, movement out of an object-modifying relative clause is blocked, as seen in (50) with the analysis in (51). The head of the RC raises into SpecKaseP with the relative CP stranded in SpecExtraP_l. In this construction, the relative clause is an island.

(50)	*yèi	míà	Kpànâ	[kúlé-í-sìà]j	wúá-nì	[ìi	tìj	yéyá-nì]
	who	foc.l	Kpana	cloth-def-pl	wash-pst	3sg.rp	3pl.rp	buy-pst
	‘Who is it that Kpana washed the clothes that he bought?’

(51)	*Movement out of an object-modifying RC

This same pattern occurs in partial- and full-movement constructions. In (52a), the verb mɛni ‘hear’ takes a CP complement (in curly brackets) containing an object-modifying relative clause (bracketed). Partial movement out of the relative clause to the embedded clause left periphery is blocked (52b), as is movement out of the RC to the matrix left periphery (52c).

(52)		Embedded clause with an object-modifying RC: partial and full movement
	a.	Mὲlí	mὲní-í	l$\mathsf{ɔ}̀$	{kὲ	Kpànâ	[kúlé-í-sìà]i
		Mary	hear-pst	nm	c	Kpana	cloth-def-pl
		wúá-ì	l$\mathsf{ɔ}̀$	[màhé-í	tìi	yéyá-nì]}
		wash-pst	nm	chief-def	3pl.rp	buy-pst
		‘Mary heard that Kpana washed the clothes that the chief bought.’
	b.	*Mὲlí	mὲní-í	l$\mathsf{ɔ}̀$	{kὲ	yèj	míà
		Mary	hear-pst	nm	c	who	foc.l
		Kpànâ	[kúlé-í-sìà]i	wúá-nì	[ìj	tìi	yéyá-nì]}
		Kpana	cloth-def-pl	wash-pst	3sg.rp	3pl.rp	buy-pst
		Intended: Who is x such that Mary heard that Kpana washed the clothes that x bought?
	c.	*yèi	míà	Mὲlí	mὲní-í	{kὲ	Kpànâ
		who	foc	Mary	hear-pst	c	Kpana
		[kúlé-í-sìà]j	wúá-ì	l$\mathsf{ɔ}̀$	[ìi	tìj	yéyá-nì]}
		cloth-def-pl	wash-pst	nm	3sg.rp	3pl.rp	buy-pst
		Intended: Who is x such that Mary heard that Kpana washed the clothes that x bought?

The previous data highlight an asymmetry between subject-modifying and object-modifying relative clauses in regards to their status as islands. While movement is possible out of subject-modifying relative clauses, it is blocked for object-modifying RCs. As such, I refer to them as mixed islands, since their status as an island is dependent on the position of the relative clause within the matrix clause. When the relative clause CP is in SpecExtraP_h, which immediately dominates SubjP, extraction is possible. On the other hand, when it is in SpecExtraP_l, movement is blocked.

In both cases, the relative clause is in SpecExtraP, and, as such, it is the position of the extraposition clause which determines whether movement out of the RC is permitted. This can be seen in comparing the structures in (53a), where the subject-modifying RC is in SpecExtraP_h and (53b), where the object-modifying RC is in SpecExtraP_l.

(53)	a.	Subject-Modifying configuration
	b.	Object-Modifying configuration

Extraction out of relative clauses has been indicated in the literature, particularly in African (Kandybowicz et al. 2021, 2023; Schurr et al. 2024; Murphy and Korsah, forthcoming) and Mainland Scandinavian languages (Engdahl 1997; Kush et al. 2013, 2019; Müller 2014, 2015). Studies indicate that there is an A-bar movement from relative clauses (Christensen and Nyvad 2014; Lindahl 2014, 2017, 2022), and there is consensus that relative clauses are weak islands in various languages, including Danish (Müller and Eggers 2022) and Swedish (Lindahl 2014), as well as English (Vincent et al. 2022) and Hebrew (Sichel 2018). However, acceptability ratings have been shown to be low (Poulsen 2008; Müller 2015, 2019). Factors such as the embedding verb (Erteschik-Shir 1973; Lindahl 2022), whether the sentence is existential or not (Kush et al. 2021; Lindahl 2022; Vincent et al. 2022), as well as the content of what is extracted (Müller and Eggers 2022) influence acceptability.

While both Mainland Scandinavian and Mende allow for extraction out of relative clauses, the contexts in which it is permitted vary. For Mainland Scandinavian, factors include the embedding verb, whether the clause is existential or not, and the content of what is fronted. In Mende, the factor seems to be syntactic. Movement out of subject-modifying RCs is sanctioned, and it is otherwise blocked. Mende does not fit the traditional notion of a weak island, in the sense that some phrase types can be moved out while others cannot (Szabolcsi and den Dikken 2002), and it therefore seems to represent a unique island variety cross-linguistically. Further research on other types of relative clauses in Mende, as well as the structure and distribution of relative clauses in other Mande languages might prove insightful.

In this section, I have sought to describe the structure and distribution of relative clauses in Mende. I have argued that they obligatorily raise, so that the head of the clause can move as a constituent into a higher position (either SpecKaseP or SpecFinP.) The relative clause raises first into SpecExtraP before the DP head of the clause remnant moves into its higher position. I have shown that while wh-movement is possible out of a subject-modifying RC, it is blocked out of object-modifying RCs. Therefore, subject-modifying RCs do not have the status of (strong/weak) islands in Mende.

5. Conclusions

The literature on island constraints has typically argued for a bifurcation—weak islands permit some types of extraction while strong islands categorically block it. In this paper, I argue for a third type of island, namely, mixed islands, where permeability seems to be conditioned by the syntactic position of the island.

Specifically, I propose that relative clauses are mixed islands: when they modify the subject, extraction is permitted; when they modify the object, extraction is blocked. I show how in both constructions the CP portion of the relative clause raises into the specifier of an Extra(position) P(hrase). The remaining DP portion then raises into a higher position (SpecFinP for subjects and SpecKaseP for objects). When the relative clause modifies the object, the ExtraP is indicated as ExtraP_l, out of which wh-movement is blocked. When the relative clause modifies the subject, the ExtraP is indicated as ExtraP_h, and wh-movement is sanctioned.

Beyond this distinction in extraction out of relative clauses, this paper makes several further contributions to research. First, it joins the other papers in this issue in contributing to the role of African languages in shaping the landscape of island research.

Second, it provides support for a promotion analysis of relative clauses from a language family that has thus far not been analyzed. By looking at reconstruction effects and the obligatory raising of the head, I have argued for a promotion analysis of relative clauses in Mende.

Third, these data have shown that resumptive pronouns do not play a role in ameliorating island violations. In some contexts, they allow for movement out of islands (e.g., subject-modifying RCs), while in other contexts they do not sanction movement (e.g., object-modifying RCs). As such, they do not line up with McCloskey’s (2017) assertion that resumptive pronouns are insensitive to constraints on movement.

Finally, it provides a detailed syntactic analysis of a Mande language, a family for which there is little research in the generative enterprise. As such, it highlights some syntactic characteristics of these understudied languages providing a basis for further research on subject markers, verbal complements, focus constructions, and the left periphery. Each of these areas warrants further study, not just in Mende, but in the broader Mande family.