1. Introduction

1Swahili [ swh] is a Bantu language, widely spoken in East Africa where it functions as a lingua franca. It is one of the official languages of Tanzania, Kenya, Uganda, and Rwanda, and the national language of Tanzania and Kenya. Like many other Bantu languages, Swahili has a highly agglutinative verb structure, as shown in (1).

(1)	Nilimwandikia
	ni‑li‑m‑andik‑i‑a
	sm‑tam‑om‑verb_root‑ds‑fv
	1sg‑pst‑3sg‑write‑appl‑ind
	‘I wrote to him.’

2The subject marker (sm) and object marker (om) are prefixes that agree with the subject and object, respectively. The prefix expressing the tense/aspect of the verb (tam) follows the sm. The verb stem consists of the verb root, optional derivational suffixes (ds), and the final vowel (fv). Derivational suffixes can be added, or replaced by other suffixes, to form pairs of noncausal (nC)/causal (C) verbs that share the same root, as shown in (2)‑(4).

		Verb root	(‑ds)	Final vowel
(2)	a.	simam		‑a	‘stand, stop (nC)’
	b.	simam	‑ish	‑a	‘stand, stop (sth.) (C)’
(3)	a.	chan		‑a	‘tear (sth.) (C)’
	b.	chan	‑ik	‑a	‘be torn (nC)’
(4)	a.	fum	‑uk	‑a	‘become unstitched (nC)’
	b.	fum	‑u	‑a	‘unstitch (C)’
cf.	c.	fum		‑a	‘knit (C)’

• Causative alternation: The causal verb is derived from the noncausal verb.
• Anticausative alternation: The noncausal verb is derived from the causal verb.
• Equipollent alternation: Both noncausal/causal verbs are marked.
• Labile alternation: The same form denotes both the noncausal and causal meaning.
• Suppletive alternation: The noncausal and causal meanings are expressed by different, lexically unrelated verb roots.

4According to this classification, the verb pairs in (2), (3), and (4) display causative, anticausative, and equipollent alternations, respectively.

5Haspelmath (1993) examined 31 pairs of noncausal/causal verb pairs in 21 languages, including Swahili, to investigate which type of strategy different languages use to express noncausal/causal verb alternations. According to his analysis, Swahili falls into neither the causative nor the anticausative type, since his data show 11 equal pairs of causative and anticausative alternations (Haspelmath 1993: 101). However, this study finds Swahili as clearly belonging to the causative type, that is, the causative alternation has the highest rate of occurrence among all five types in the set of 31 noncausal/causal verb pairs. Crucial to this analysis is a diachronic phonological change that was not considered in Haspelmath’s (1993) study. This diachronic sound change is called “spirantization”, a well-known phenomenon in the study of Bantu languages (see, among others, Nurse & Hinnebusch 1993; Schadeberg 1994; Bostoen 2008).

6The aim of this paper is to revisit Haspelmath’s (1993) analysis of noncausal/causal verb pairs in Swahili and, taking into account this diachronic sound change, argue that Swahili is a causative type language contrary to Haspelmath’s analysis. Section 2 elucidates each alternation type in Swahili, and Section 3 compares this data with Haspelmath’s data. Section 4 compares data from Swahili with those from two other Bantu languages, Herero and Matengo. Section 5 discusses the correlation between the spontaneity value of noncausal verbs and markedness, as compared to that found in Comrie’s analysis.

2. Types of alternations in Swahili

7This section discusses Haspelmath’s five alternation strategies based on his sample of 31 noncausal/causal verb pairs in Swahili.

2.1 Causative alternation type

9Examples of the suffix ‑ish are shown in (5)‑(9).

(5)	a.	gand‑a	‘freeze (nC)’
	b.	gand‑ish‑a	‘freeze (sth.) (C)’

(6)	a.	zam‑a	‘sink (nC)’
	b.	zam‑ish‑a	‘sink (sth.) (C)’

(7)	a.	simam‑a	‘stand, stop (nC)’
	b.	simam‑ish‑a	‘stand, stop (sth.) (C)’

(8)	a.	pand‑a	‘raise (nC)’
	b.	pand‑ish‑a	‘rise (C)’

(9)	a.	yumb‑a	‘swing, rock (nC)’
	b.	yumb‑ish‑a	‘swing, rock (sth.) (C)’

10Example (10) illustrates the use of the suffix ‑y. The causal verb ponya ‘cure’ is derived from the noncausal verb pona ‘recover’ by adding the suffix ‑y.

(10)	a.	pon‑a	‘recover (nC)’
	b.	pon‑y‑a	‘cure (C)’

11In (10), the suffix‑y is simply attached to the root pon. However, this suffix is not usually visible on the surface. For example, in (11), chemka is a noncausal verb meaning ‘boil’, and the causative verb chemsha is considered to have historically been chemk‑y‑a, that became chemsha as a result of spirantization: k + y > sh. The same can be seen in (12)‑(16).

(11)	a.	chemk‑a	‘boil (nC)’
	b.	chemk‑y‑a > chemsha	‘boil (sth.) (C)’

(12)	a.	kauk‑a	‘dry (nC)’
	b.	kauk‑y‑a > kausha	‘dry (sth.) (C)’

(13)	a.	amk‑a	‘wake up (nC)’
	b.	amk‑y‑a > amsha	‘wake (so.) up (C)’

(15)	a.	zunguk‑a	‘turn (nC)’
	b.	zunguk‑y‑a > zungusha	‘turn (sth.) (C)’

(16)	a.	bingirik‑a	‘roll (nC)’
	b.	bingirik‑y‑a > bingirisha	‘roll (sth.) (C)’

(17)	a.	ungu‑a	‘burn (nC)’
	b.	ungu‑y‑a > unguza	‘burn (sth.) (C)’

(18)	a.	ja‑a	‘fill (nC)’
	b.	ja‑y‑a > jaza	‘fill (sth.) (C)’

(19)	a.	ene‑a	‘spread (nC)’
	b.	ene‑y‑a > eneza	‘spread (sth.) (C)’

(20)	a.	endele‑a	‘develop (nC)’
	b.	endele‑y‑a > endeleza	‘develop (sth.) (C)’

(21)	a.	pote‑a	‘get lost (nC)’
	b.	pote‑y‑a > poteza	‘lose (C)’

(22)	a.	bembe‑a	‘swing, rock (nC)’
	b.	bembe‑y‑a > bembeza	‘swing, rock (sth.) (C)’

13In Swahili, 18 out of the 31 pairs examined conform to this type according to my data. I will return to a discussion of the phonological changes observed here in Section 3.

2.2 Anticausative alternation type

14Pairs of the anticausative alternation type include acausal verb and a noncausal verb derived from the causal verb by adding ‑ik. Examples (23)‑(29) are of this type. The causal verb haribu in (29), meaning ‘destroy’, is a loanword, without the final vowel ‑a. Because a derivational suffix is usually inserted between a verb root and the final vowel ‑a, loanwords without the final vowel are modified to the prototypical verbal structure through the addition of the suffix ‑ik and the final vowel ‑a.

(23)	a.	zim‑ik‑a	‘go out (nC)’
	b.	zim‑a	‘put out (C)’

(24)	a.	maliz‑ik‑a	‘finish (nC)’
	b.	maliz‑a	‘finish (sth.) (C)’

(25)	a.	ung‑ik‑a	‘connect, be connected (nC)’
	b.	ung‑a	‘connect (sth.) (C)’

(26)	a.	kusany‑ik‑a	‘gather, be gathered (nC)’
	b.	kusany‑a	‘gather (sth.) (C)’

(27)	a.	vunj‑ik‑a	‘break, be broken (nC)’
	b.	vunj‑a	‘break (sth.) (C)’

(28)	a.	fung‑ik‑a	‘close, be closed (nC)’
	b.	fung‑a	‘close (sth.) (C)’

(29)	a.	haribu	‘destroy, be destroyed (nC)’
	b.	haribu‑ika > haribika	‘destroy (sth.) (C)’

15Example (30) illustrates a different type of derivation of the noncausal verb from the causal form.

(30)	a.	ji‑funz‑a	‘learn (nC)’
	b.	funz‑a	‘teach (C)’

2.3 Equipollent alternation type

17In the equipollent alternation type, both the noncausal and causal forms are equally marked by derivational suffixes. The noncausal verb is derived by adding ‑uk to the verb root, and ‑u to the causal verb in Swahili.

(31)	a.	fung‑uk‑a	‘open, be open (nC)’
	b.	fung‑u‑a	‘open (sth.) (C)’
cf.	c.	fung‑a	‘close (sth.) (C)’

(32)	a.	pas‑uk‑a	‘split, be split (nC)’
	b.	pas‑u‑a	‘split (sth.) (C)’

19Another equipollent suffix alternation is shown in (33).

20The causal verb geua ‘change (C)’ in (33c) is historically geula. With the addition of the causative suffix ‑y, it became geuza in (33b), due to the spirantization rule l + y > z. Thus, both geua and geuza are causal. However, geua only appears in an old dictionary by Johnson (1939), while recent dictionaries treat geuza as the transitive counterpart of the intransitive geuka (BAKITA 2010; 2015; Mohamed 2001; 2013: TATAKI; 2014). Haspelmath (1993: 119) lists both geua and geuza as causal verbs, and analyses them as anticausative and equipollent types, respectively.

(34)	Haspelmath (1993: 119); cf. verb pair #25 in Appendix 1
	a.	geu‑a ‘change (C)’/geu‑k‑a ‘change (nC)’	Anticausative type
	b.	geu‑z‑a ‘change (C)’/geu‑k‑a ‘change (nC)’	Equipollent type

21While guea as in (34a) is not incorrect, geuza is much more commonly used nowadays. Almost 100 years ago, Johnson (1939: 113) noted that geuza was generally used to mean ‘change, make (sth.) different, alter’. Therefore, this study uses geuza only as the causal counterpart of the noncausal verb geuka.

22Overall, 3 out of the 31 pairs are of the equipollent alternation type in Swahili.

2.4 Labile alternation type

23In the labile alternation type, the same form is used for both the noncausal and causal counterparts of the pair. Only one of the 31 Swahili pairs is of this type. This is shown in (35).

(35)	a.	Shule	i‑me‑anz‑a.	‘begin, start (nC)’
		9.school	sm9‑prf‑begin‑fv
		‘School has started.’
	b.	Tu‑me‑anz‑a	shule.	‘begin, start (sth.) (C)’
		sm1pl‑prf‑begin‑fv	9.school
		‘We have started school.’

2.5 Suppletive alternation type

24Pairs of the suppletive alternation type use different verb roots for the noncausal and causal meanings. Again, only one pair of this type is found in Swahili, that is, ‘die’ and ‘kill’, as shown in (36).

(36)	a.	f‑a	‘die’
	b.	u‑a	‘kill’

25The pair ‘die/kill’ belongs to the suppletive type in 16 out of the 21 languages investigated by Haspelmath (1993), and Swahili is one of them.

26In the total of 31 verb pairs considered here, only 5 pairs in Swahili are of the non-directed type, that is, an equipollent, labile, or suppletive alternations.

3. Comparison with Haspelmath’s analysis

27As already seen, of the 31 Swahili verb pairs, 18 are causative, 8 are anticausative, 3 are equipollent, 1 is labile, and another 1 is suppletive. Although Haspelmath’s (1993) analysis suggests that Swahili prefers neither the causative nor anticausative type, the present study argues that Swahili clearly leans towards the causative type. Table 1 compares the results of this study with those of Haspelmath’s.

Table 1 — Number of pairs per alternation type

	My analysis	Haspelmath (1993)
Causative	18	11
Anticausative	8	11
Equipollent	3	8
Labile	1	0
Suppletive	1	1

28According to Haspelmath’s analysis, 11 of the 31 pairs belong to the causative type, and 11 pairs belong to the anticausative type. However, the current analysis finds 18 pairs belonging to the causative type, and only 8 pairs belonging to the anticausative type. The crucial difference between these two analyses lies in the treatment of pairs such as those presented in (37). While Haspelmath treated them as equipollent, this study classifies them as belonging to the causative type (see Appendix 1).

(37)		Noncausal	Causal
	a.	chemka	chemsha	‘boil’
	b.	kauka	kausha	‘dry’
	c.	amka	amsha	‘wake up’
	d.	yeyuka	yeyusha	‘melt, dissolve’
	e.	zunguka	zungusha	‘turn’

29At first glance, these pairs appear to be derived from the derivational suffix ‑k(a) for noncausal verbs, and ‑sh(a) or ‑z(a) for causal verbs. In other words, they seem to be of the equipollent type, wherein both the noncausal and causal verbs are derived from the same root. However, this is only an analysis of the verbs’ surface forms. As discussed in Section 2.1, spirantization has occurred on the causal verbs in these pairs, with the addition of the causative suffix ‑y to the root-final k. This phonological change, triggered by the suffix y, occurred not only with k, but also with other consonants (Polomé 1967; Schadeberg 1992; Mpiranya 2015), as shown in (38)‑(41). The noncausal verb ingia in (41a) was originally ingila, with the root-final *l (Nurse & Hinnebusch 1993: 98‑104), and spirantization occurred when that l was followed by the causative derivation suffix ‑y (Schadeberg 1992: 10).

t + y > sh
(38)	a.	pit‑a	‘pass (nC)’
	b.	pit‑y‑a > pisha	‘let (so.) pass (C)’

p + y > fy
(39)	a.	ogop‑a	‘be frightened (nC)’
	b.	ogop‑y‑a > ogofya	‘frighten (C)’

l + y > z
(40)	a.	lal‑a	‘lay down, sleep (nC)’
	b.	lal‑y‑a > laza	‘lay (so.) in bed (C)’

(41)	a.	ingil‑a > ingia	‘get in (nC)’
	b.	ingil‑y‑a > ingiza	‘put (sth.) in (C)’

31The reason why Haspelmath did not find this clear causative tendency in Swahili is because spirantization was not considered. Moreover, there are additional problems with the list he used. While anza, meaning ‘begin’, is the only verb pair belonging to the labile type as found in this study, Haspelmath considers it as representative of the causative alternation type, as shown in (42) (cf. verb pair 16 in Appendix 1).

(42)			Noncausal	Causal
	a.	Haspelmath (1993)	anza	anz‑ish‑a	‘begin’
	b.	This paper	anza	anza	‘begin’

32However, anzisha, meaning ‘establish, set up’ (Mohamed 2011: 21) is not the causal counterpart of anza, meaning ‘begin’. Other examples of inappropriate translations include yonga, that is, ‘stagger’ for ‘rock’ (verb pair #23 in Appendix 1), and the pair of fanya ujambo/pata ujambo, that is, ‘make goodness/get goodness’ for ‘improve’ (verb pair #22 in Appendix 1). Moreover, Haspelmath provides passive forms for the noncausal counterparts of the causal verbs unga, meaning ‘connect’ (verb pair #24 in Appendix 1), and funga, meaning ‘close’ (verb pair #29 in Appendix 1). However, these verbs have the noncausal forms ungika and fungika as noncausal counterparts, respectively. This could have made a difference in the analysis.

4. Comparison with other Bantu languages: Herero and Matengo

33Table 2 compares similar datasets from Herero, spoken in Namibia, and Matengo, spoken in Tanzania, with the Swahili data. The Swahili data are provided in Appendix 1, and the Herero (Yoneda 2014a) and Matengo (Yoneda 2014b) data are provided in Appendix 2.

Table 2 — Comparison of Swahili with Herero and Matengo (Yoneda 2014a; Yoneda 2014b; Yoneda 2014c; updated 2022)

	Swahili	Herero	Matengo
Causative	18	15	5
Anticausative	8	4	4
Equipollent	3	8	16
Labile	1	2	1
Suppletive	1	2	3
Total	31	31	29

34The Matengo data are incomplete because only 29 of the 31 pairs are attested, yet, Table 2 shows a clear tendency in Swahili for the causative alternation type. Herero, where 15 of the 31 pairs belong to the causative alternation type, shows a similar tendency but the prominence of the causative alternation is slightly more robust in Swahili. Furthermore, Swahili has only a small number of non-directed alternations. Unlike Swahili and Herero, Matengo shows a clear tendency for the equipollent type.

35Herero exhibits only one morphological strategy for causative correspondence, in the surveyed noncausal/causal pairs, that is, the derivational suffix ‑is. On the other hand, the causative strategy for noncausal/causal alternations in Matengo is more complicated. Matengo does not have a productive causative suffix, such as ‑ish in Swahili, and its derivational processes are not always transparent. There are 5 causative pairs in the Matengo dataset, each of which has a different derivational process. In (43), the causal verb is formed by deriving the noncausal member with the suffix ‑ik, in (44) with the suffix ‑i, no suffix is added in (45) and (46), and the root-final consonant k changes to h in (47).

(43)	a.	dʒem‑a	‘stop (nC)’
	b.	dʒem‑ik‑a > dʒemeka	‘stop (sth.) (C)’

(44)	a.	lam‑a	‘improve (nC)’
	b.	lam‑i‑a	‘improve (C)’

(45)	a.	bel‑a (> be‑el‑a?)	‘boil (nC)’
	b.	be‑a	‘boil (water) (C)’

(46)	a.	hɔb‑a	‘get lost (nC)’
	b.	hɔ‑a	‘lose (C)’

(47)	a.	ɲuk‑a	‘rock (nC)’
	b.	ɲu(h)‑a	‘rock (sth.) (C)’

(48)		Noncausal	Causal
	a.	booka	booa	‘move/move (sth.)’
	b.	pita	pia	‘be out/let (sth.) out’
	c.	pjopa	pjoa	‘be warm/make (sth.) warm’
	d.	dʒɔgɔpa	dʒɔgɔa	‘be frightened/frighten’

37This could be a case of over-application caused by the appearance of the ‑uk/‑u, or ‑ik/‑i pairs. The causal forms seem to have been derived from their noncausal counterparts by deleting the root-final consonant. Another account of this could be the deletion of the Cj at the verb root-final. Yoneda (2000: 158) noted that since the combination Cj never appears at the verb root-final in the present form of Matengo, it is possible that the pair was originally CVC vs. CVC‑i‑a > CVCja, wherein the ‑i was suffixed for causative alternation, and the Cj at the root-final was diachronically deleted, resulting in CVa.

5. The correlation between the spontaneity of noncausal verbs and markedness

40However, in Haspelmath’s Swahili data, 5,5 causative pairs appear in the first half, while 6 pairs appear in the second half (Comrie 2006: 306). Therefore, in Swahili, a slightly higher number of causative pairs are attested in the second half than in the first half. Based on this observation, Comrie (2006: 308) concludes that Swahili is an exception to his own hypothesis.

Table 3 — Correlation between the spontaneity of noncausal verbs and their markedness

	Analysis in this paper		Analysis in Comrie (2006) based on Haspelmath’s (1993) data
Verb pairs	#1‑15	#16‑30	#1‑15	#16‑30
Anticausative	4	4	4	7
Causative	11	7	5.5	6

41However, as Table 3 shows, according to the alternative analysis that takes spirantization into account, 11 causative alternation pairs appear in the first half, while 7 pairs appear in the second half (see also Appendix 1).

(49)	Causative alternation type among verb pairs #1‑15
	a.	chemka/chemsha	‘boil’
	b.	ganda/gandisha	‘freeze’
	c.	kauka/kausha	‘dry’
	d.	amka/amsha	‘wake up’
	e.	zama/zamisha	‘sink’
	f.	yeyuka/yeyusha	‘melt, dissolve’
	g.	simama/simamisha	‘stop’
	h.	zunguka/zungusha	‘turn’
	i.	ungua/unguza	‘burn’
	j.	jaa/jaza	‘fill’

(50)	Causative alternation type among verb pairs #16‑30
	a.	enea/eneza	‘spread’
	b.	bingirisha/bingirisha	‘roll’
	c.	endelea/endeleza	‘develop’
	d.	potea/poteza	‘get lost/lose’
	e.	panda/pandisha	‘rise/raise’
	f.	pona/ponya	‘improve (heal)’
	g.	yumba/yumbisha	‘rock’

42Although the number of anticausative types among the verb pairs #1‑15 and #16‑30 is the same, and does not support the markedness hypothesis, the number of causative types confirms that Swahili is not an exception to, but rather a confirmation of, Comrie’s (2006) hypothesis.

6. Conclusion

43This paper argues that, based on a sample of 31 verb pairs, the causative strategy is preferred for encoding the noncausal/causal alternation in Swahili. This study refines previous findings by considering the historical sound change of spirantization. The prominence of the causative strategy in Swahili is made even clearer through comparison with other Bantu languages such as Herero and Matengo, with the latter showings. The causative type, therefore, is not a cross-Bantu tendency. Some Bantu languages, like Matengo, show an equipollent tendency. Swahili, instead, has a low number of noncausal/causal verb pairs with a non-directed alternation.

44Interestingly, this investigation of the noncausal/causal alternation indicates variation not only in the type but also in the strategies of both causative and anticausative alternations across Bantu languages. Although a micro-variation study is not the topic of this paper, further research would be worthwhile, not only on the tendency of alternation types, but also on cross-Bantu variation in causative and anticausative strategies.

Acknowledgments

45This paper is based on my presentation at the 10th World Congress of African Linguistics (WOCAL), 9 June, 2021, held online at Leiden University. I would like to thank all those who gave me helpful comments during and after the conference. I am also grateful to the anonymous reviewers and the editors for their helpful comments. Of course, any mistakes and misunderstandings are my own responsibility. This research has been conducted under the financial support of Grants-in-Aid for Scientific Research (Nos. 18K00538 and 19K00550) from the Japan Society for the Promotion of Science.

Abbreviations

appl	applicative suffix
caus	causative suffix
C	causal
ds	derivational suffix
fv	final vowel
ind	indicative
nC	noncausal
om	object marker
pl	plural
prf	perfect
pst	past
sg	singular
sm	subject marker
tam	tense/aspect marker