The Consonant Inventory of Proto-Tsonga-Copi

Abstract

Recent studies have greatly furthered our understanding of the Southern Bantu languages, but questions about the internal relationships of the Southern Bantu language subgroups and the validity of the clade as a whole still remain. This study attempts to reconstruct the consonant inventory of one proposed genetic clade, that of Tsonga-Copi (S50–S60). Using published dictionaries and reference works for each language of the subgrouping, a corpus of cognate vocabulary was assembled. Each term was then matched, where possible, to a reconstruction in the Bantu Lexical Reconstructions 3 (BLR3) database. Sound correspondences were identified and used to reconstruct the consonant inventory of Proto-Tsonga-Copi. In addition to the discovery of several typologically unusual sound changes, the results of this study also lend support to existing and developing hypotheses about both the internal relationships of Southern Bantu clades, as well as the nature of language contact in (pre)historic Southern Africa, particularly the influence of Khoisan and other Bantu languages.

1. Introduction

The Southern Bantu language group consists of ~80 living languages (Maho, 2009) divided among six subgroups: Shona (S10), Venda (S20), Sotho-Tswana (S30), Nguni (S40), Tsonga (S50), and Copi (S60). It has long been a question of how these groups within Southern Bantu are related to one another, as well as whether Southern Bantu as a whole constitutes a valid genetic language family. Reconstructions of Proto-Bantu (PB) (Meinhof, 1899; Meeussen, 1967; Guthrie, 1971) based on lexical and phonological comparison generally noted the high degree of similarity shared by the six clades but did not reconstruct lower-order protolanguages such as Proto-Southern Bantu (PSB) itself. This task was attempted by Van der Spuy (1989), whose research unfortunately lacked any representation of the Copi clade. Van der Spuy’s study (summarized in Van der Spuy, 1990) yielded a Proto-Southern Bantu which could not clearly be distinguished from the earlier reconstructions of Proto-Bantu, a finding which was supported by later comparative work by Janson (1991–1992). This casts doubt on the validity of Southern Bantu as a sufficiently distinct unit from (late) Proto-Bantu, but of course does not discount the possibility that lower-order groupings indeed represent valid genetic clades.

Bantu classification has also been attempted through lexicostatistic studies (Ehret, 1972; Borland, 1986; Bastin et al., 1999) which compare basic vocabulary that is supposed to be more resistant to borrowing. The former two studies only examined Southern Bantu, effectively presupposing it as a valid clade. However, Bastin et al. (1999) incorporated data from 450 Bantu languages, and its results did indeed point to Southern Bantu as a valid internal clade. More recent Bantu classification studies (Currie et al., 2013; Grollemund et al., 2015; Koile et al., 2022) use Bayesian phylogenetic models, which improve upon lexicostatistic models by focusing on shared innovations rather than overall similarity to establish genetic relationships. These studies were all conducted using over 400 Bantu languages, and each again pointed to the validity of Southern Bantu as a clade. When it comes to the internal classification of Southern Bantu, however, these studies suffer from a poverty of data. Currie et al. (2013) include 23 Southern Bantu languages, while Grollemund et al. (2015) and Koile et al. (2022) include only 11, and none of these studies include any Copi languages.

The newest phylogenetic study of Bantu, Gunnink et al. (2025), incorporates ~120 Bantu languages, 34 of these being Southern Bantu, including both languages of the Copi group. As in their prior studies (Sengupta et al., 2021; Gunnink et al., 2022), the authors employ Bayesian models with lists of approximately 100 semantically controlled lexemes to determine the overall phylogeny of the languages sampled. The result of this study both supports the existence of Southern Bantu as a clade and groups the Tsonga (S50) and Copi (S60) clusters together as a distinct subclade within Southern Bantu to a high degree of probability. This implies that the languages of these two groups share a single common ancestor, Proto-Tsonga-Copi (PTC). This ancestor should theoretically be reconstructable using comparative methodology.

This study investigates the phonological relationships between the Tsonga (or Tswa-Ronga) and Copi (or Copi-Tonga) clades, and attempts to reconstruct the consonant inventory of their nearest ancestor, Proto-Tsonga-Copi. First, an account of the synchronic phonological inventories is given with a comparison to Meeussen’s (1967) reconstruction of Proto-Bantu. Afterwards, the specific methodology of this particular study is elaborated on. Next, an account of the reflexes of each Proto-Bantu phoneme is given for the languages under consideration, along with discussion of the implications for the reconstruction of Proto-Tsonga-Copi. Finally, the reconstructed consonant inventory will be examined typologically, and notable findings and future avenues for research will be discussed.

It is important to note that the present study does not attempt to demonstrate the validity of Tsonga-Copi as a clade, though the implications of this study and the further research necessary to demonstrate the clade’s validity will be discussed later on. This work is principally a comparative phonological study much in the vein of Guthrie’s work, with a much narrower focus. While Guthrie’s (1971) comparative analysis is impressive, it is limited in usefulness for determining familial relationships and shared phonological innovations at a micro level. To illustrate, Guthrie’s phonological correspondences for S50 are based primarily on Tswa, for which he cites 152 lexemes. For Tsonga and Ronga, on the other hand, he cites 11 and 15 lexemes, respectively. From this data alone, it is hardly possible to identify any divergence within S50 languages. While S60 is comparatively better represented, with 100 Copi and 69 Tonga lexemes, there are only 31 reconstructed forms where Guthrie cites a lexeme in both S50 and S60. Van der Spuy’s comparative study also has its limits in usefulness, arbitrarily taking a single language from each Southern Bantu clade to be representative of the whole, in addition to the aforementioned omission of Copi data. The present study provides an abundance of data on all S50 and S60 languages, which allows lexemes in each language to be compared directly to their cognates in another, making it easier to tease apart dialectical divergence, loanwords, and other otherwise overlooked phenomena. Beyond that, this study integrates in its analysis of correspondences the results of decades of development in Bantu historical research which have occurred since Guthrie’s study. The result is an expanded understanding of the historical phonology of both S50 and S60, as well as a strong basis for which future comparative studies on the internal relationships of Southern Bantu languages can be conducted.

1.1. Tonga

The Tonga language (S62), also known as Gitonga or Guitonga, is spoken by approximately 327,000 speakers in the vicinity of Inhambane in Mozambique (Eberhard et al., 2022). The language should not be confused with the Tonga language of Zambia and Zimbabwe (M64) or the Tonga language of Malawi (N15), and so it is sometimes called Tonga of Inhambane or Mozambican Tonga to disambiguate. All sources consulted (Lanham, 1955; Amaral et al., 2007; Ngunga & Faquir, 2012) identify five dialects of Tonga: Gitonga gya Khogani/Khoga, Nyambe, Khumbana, Sewi, and Morrumbene/Rombe. All sources make use of the Khoga dialect, spoken on the coast surrounding Inhambane Bay, as a standard. Lanham (1955, p. v) also incorporates some words from Morrumbene, while noting that no great differences between the two occur. Lanham (1955, p. 109) further notes that Sewi, spoken within the city of Inhambane proper, has the largest number of Portuguese borrowings. Amaral et al. (2007, p. 109) suggest that Nyambe, spoken on the oceanic coast of the Jangamo district, is the most archaic dialect, but do not exemplify this. The consonant inventory given in

Table 1. The consonant inventory of Tonga ¹.

		Bilabial	Labio-Dental	Alveolar	Lateral	Palatal	Velar	Glottal
Nasal		m		n		ɲ	(ŋ)
Stop	Voiceless	p		t			k
	Voiceless Aspirated	ph		th			kh
	Voiced	b		(d)
	Implosive	ɓ	v’ 1	ɗ
Affricate	Voiceless		pf	ts
	Voiceless Aspirated		pfh	tsh
	Voiced		bv	dz
Fricative	Voiceless		f	s	(ɬ)			h
	Voiced	β		(z)			ɣ
Liquid				r	l
Glide		w				j

¹ The phonemes /z/ and /ɬ/ are said to only occur in non-native vocabulary by Lanham (1955), though he does not list potential source languages besides Tswa. The phonemes /ŋ/ and /d/ are not given by Lanham, but are present in both Amaral et al. (2007) and Ngunga and Faquir (2012).

attempts to synthesize the consonant inventories reported by these three sources. In this and other inventories, labialization, palatalization, and prenasalization are not indicated, as they are assumed to reflect sequences of multiple phonemes. Phonemes given in parentheses are claimed to be loan phonemes, and clicks (which are loan phonemes in all S50 and S60 languages) are omitted entirely.

The Tonga consonant inventory shows four series of stops, those being voiceless, voiceless aspirated, voiced, and implosive. These series are largely paralleled in the affricates, with the exception of the implosives. Only the voiced stops and affricates can be prenasalized. A notable asymmetry is the lack of voiced and implosive velar stops. Lanham (1955) gives a phoneme /β/, which corresponds to both /β/ and /v’/ in Amaral et al. (2007). As will be discussed in Section 3.1 and Section 3.5, the distinction in Amaral et al. is etymological, and so Lanham’s description must reflect a dialectical merger of these phonemes.

1.2. Copi and Lenge

The Copi language (S61), often written Chopi or Cicopi, is spoken by around 1,100,000 speakers in Mozambique. These speakers primarily live in a strip along the coast of Gaza province (Eberhard et al., 2022). Both Dos Santos (1949) and Bailey (1976) divide Copi dialects into three groups: Southern, Central, and Northern. Dos Santos (1949, p. 9) claims that the Northern group shows the greatest number of dialectical differences relative to the Central group, which he takes to be the “pure” Copi. He further notes that the Southern dialects show the greatest amount of hybridism with Changana, due to the bilingualism of the people living there. This observation is echoed by Bailey (1976, p. 1), who adds that women in the southernmost regions, until recently, spoke the Lenge variety of Copi, which was highly divergent and influenced by Changana, while the men of the region principally spoke Changana. In addition to Changana, Bailey describes a high level of bilingualism in both Tswa and Tonga as well, and Eberhard et al. (2022) report that in modern times, approximately 50% of Copi speakers speak Tswa.

The modern dialectical picture is not entirely clear. Ngunga and Faquir (2012) and Eberhard et al. (2022) cite six dialects: Ndonge/Ndonje, Lenge/Lengue, Tonga (distinct from the Tonga language detailed above), Lambwe, Khambani, and Copi. Ngunga and Faquir use the latter of these as their reference variant, while noting the need for in-depth dialectological studies to verify the existence and degree of divergence of these dialects. Maho (2009) treats Lenge as a distinct language in his classification of Bantu languages, also marking the variety as possibly extinct. Hammarström (2019) instead lists Lenge as a dialect of Copi.

Table 2. The consonant inventory of Copi.

		Labial	Labiodental	Labio-Alveolar	Alveolar	Lateral	Postalveolar	Palatal	Velar	Glottal
Nasal	Plain	m			n			ɲ	ŋ
	Aspirated	(mɦ)			(nɦ)
Stop	Voiceless	p			t				k
	Voiceless Aspirated	ph			th				kh
	Voiced	b			d			ɟ	g
	Implosive	ɓ			ɗ
Affricate	Voiceless		pf	pʂ	ts	(tɬ)	tʃ
	Voiceless Aspirated		pfh	pʂh	tsh	(tɬh)	tʃh
	Voiced		bv~v	bʐ	dz	(dɮ)	dʒ
Fricative	Voiceless		f	ʂ	s	(ɬ)	(ʃ)
	Voiced									ɦ
Liquid	Plain				r	l
	Aspirated				rɦ
Glide		w	ʋ					j

attempts to synthesize the consonant inventories proposed in the various reference sources into a single overview. The information is drawn principally from Dos Santos (1949) and Bailey (1976), of which neither attempts to represent a single variety of Copi. Smyth and Matthews (1902) do not include particularly detailed phonetic descriptions in their dictionary of Lenge, but the data is nevertheless worthy of particular note due to its treatment of a singular language variety and the unclear status of Lenge as a dialect or a separate language. For that reason, phonemes which appear in Lenge are given in bold, with the phoneme /r^ɦ/ only reported for Lenge further italicized. In any case, the inventory below should not be accepted without caution; Bailey (1995, p. 138) notes the “considerable” ideolectical and regional variation that still exists among Copi speakers, and the present study would benefit greatly from modern fieldwork data on Copi, as well as all of the languages under consideration. Nevertheless, the following paragraphs attempt to discuss and balance the various authors’ reports and their implications for the overall phonology of Copi.

A clear difference between the consonant inventories of Copi and Tonga is in the number of affricates. Copi has both labiodental and alveolar affricates, as does Tonga, but also has a series of affricates described as “labioalveolar” (Bailey, 1976, p. 14) and a postalveolar series. The former of these is generally notated in complexes which use the symbols for the retroflex sibilants <ʂ> and <ʐ>. This is simply notational, and these sounds are not necessarily phonetically retroflex.

Copi shows the same four-way stop and affricate contrast as Tonga. Voiced stops and affricates can be prenasalized, and these prenasalized obstruents can also be aspirated. Bailey (1995, p. 146) states that most lexemes containing voiced aspirates are clear borrowings from Tsonga, but some appear to be native Copi words. In either case, he notes a large degree of dialectical variation in the realization of these words. Besides the voiced aspirates, Bailey also claims that aspirated nasals, an entire series of lateral affricates, and a postalveolar fricative /ʃ/ are present but exclusively confined to Tsonga borrowings.

When it comes to the phonetic nature of certain stops, the various sources consulted are not in agreement. Ngunga and Faquir (2012) include both a voiceless and voiced palatal stop in the consonant inventory, and cite [ca] ‘dawn’ and [ɟika] ‘diverge’ as examples. The former of these is given in Bailey (1995) as [tʃa] with an affricate, and the respective spellings tcha and djika in Dos Santos (1949) also suggest an affricate. As such, these phonemes will be spelled in this paper as affricates. The voiced aspirated affricate [tʃ^h] is not present in Ngunga and Faquir (2012), but is reported by Bailey (1976, 1995). Neither work cites many lexical examples, but the former gives [tʃ^hisa] ‘abort’, spelled in Dos Santos (1949) as tchisa, identical to the spelling of [tʃ]. It is therefore not possible to distinguish between [tʃ] and [tʃ^h] following Dos Santos, and so all instances will be cited as [tʃ].

Bailey (1976) does nevertheless report the existence of a palatal stop, which can optionally be implosive. The representation of this sound is not clear in Dos Santos (1949). Dos Santos’ dictionary contains two sections with ‘d’ sounds, the first being “palatal d” and the latter being “dental d”. Both of these are spelled <d>. This does not seem to overlap with the palatal stop distinction noted by Bailey. Bailey (1995) cites the lexemes [ɟa] ‘eat’ and [ɗuka] ‘try’, but these are both listed in Dos Santos (1949) under “palatal d”. Dos Santos spells these dia and duka, respectively, so it could be suggested that the sequence <di> indicates a palatal stop. Unfortunately, Bailey (1995) claims that the sequence [ɗj] exists and phonetically contrasts with [ɟ], and gives [ɗjanda] ‘egg’ as an example. This is listed in Dos Santos (1949) as a dialectical variant which is spelled dianda. Dos Santos is also not consistent in reflecting the phoneme Bailey identifies as [ɗ] as “palatal d”, with the lexeme [ɗaja] ‘kill’ given under “dental d” in Dos Santos (1949). Since it is unclear how to reflect the phoneme Dos Santos spells <d>, such entries will always be cited with [ɗ].

It is also worth noting that while Dos Santos’ orthography seems to consistently reflect the difference between [ɓ] and [b] (the former spelled <b> and the latter <bh>), the same cannot be said for [ɗ] and [d]. Ngunga and Faquir (2012) have the term [dula] ‘expensive’, which is spelled dula in Dos Santos (1949) and listed under “dental d”. Bailey (1976) has the term [dana] ‘call’, which Dos Santos spells dana and lists under “palatal d”. Dos Santos (1949) only spells two words with <dh>, both listed under “dental d”. These are dhani ‘species of tree from which bows are made’, and dhuti ‘hut’; neither of these are present in Ngunga and Faquir (2012) or Bailey (1976, 1995), so it cannot be made clear what contrast is spelled here.

One final bit of contention between sources has to do with the voiced labiodentals. Bailey (1976) lists both /bv/ and /v/ in the phonemic inventory, and gives /bvarula/ ‘tear’ and /va/ ‘be’ as examples. In Bailey (1995), /bv/ is no longer reported, but /ʋ/ is, with exemplary terms being /vuna/ ‘help’ and /ʋa/ ‘be’. Ngunga and Faquir (2012) again report a contrast of /v/ and /ʋ/, with /varula/ ‘tear’ and /ʋeka/ ‘put’ as examples. All of these words are spelled with initial <v> in Dos Santos (1949), with the exception of ‘tear’; this is cited with two spellings, bvarula and varula. All of this suggests two phonemes, /ʋ/ and /bv ~ v/, with the latter showing either free variation, dialectical variation, or a sound shift in progress. When Dos Santos spells <v>, this will be denoted as /ʋ/, but the potential for conflation with /v/ should be noted.

1.3. Tsonga and Changana

The Tsonga/Changana (S53) language is spoken natively by approximately 5,680,000 speakers in South Africa and Mozambique. Of these, 2,280,000 live in South Africa in the northeastern Limpopo and Mpumalanga provinces, and 4,200,000 live in Mozambique in the northern parts of Gaza and Maputo provinces and the west of Inhambane province. The language is spoken by an additional 100,000 people living in Zimbabwe, where it is often called Shangani (Eberhard et al., 2022). The term Tsonga here refers to the South African standard of the language, which has a written form distinct from Changana, the Mozambican standard. Unsurprisingly, the Mozambican varieties have a large number of loans from Portuguese which are not found in South African Tsonga.

Teasing apart the picture of Tsonga/Changana and its dialects is difficult; to start, the term Tsonga is often used to refer not only to Changana but also to Ronga and Tswa as well. Descriptions of the various dialects therefore often appear to arbitrarily include or exclude certain varieties based on what is perceived to fall under the umbrella of a given language. Maho (2009) distinguishes eight dialects of Tsonga/Changana: Xiluleke, N’gwalungu, Hlave, Nkuna, Gwamba, Nhlanganu, Djonga/Jonga, and Bila. He lists Hlengwe as a distinct language, and incorporates Dzibi and Dzonga under Tswa and Konde, Ssonga, and Xonga under Ronga. Baumbach (1974, 1987) identifies four dialect clusters: the nucleus cluster, encompassing Changana, Nkuna, Gwambe, Hlave, and N’walungu; a peripheral dialect cluster with only Konde named, an intermediate cluster ‘A’ consisting of Luleke and Nhlanganu, and an intermediate cluster ‘B’ consisting of Xonga and Ssonga (with Ssonga omitted in his 1987 grammar). Both Sitoe (1996) and Ngunga and Faquir (2012) report five dialects of Changana: Hlanganu, Dzonga, N’walungu, Bila, and Hlengwe.

It is clear that there is substantial overlap between what is considered Tsonga, Ronga, or Tswa, and that within and between these groups there is a substantial degree of variation. As with the other languages under consideration in this study, this is compounded by the high levels of bilingualism among speakers. The consonant inventory for Tsonga in

Table 3. The consonant inventory of Tsonga.

		Bilabial	Labiodental	Alveolar	Lateral	Postalveolar	Retroflex	Palatal	Velar	Glottal
Nasal	Plain	m		n				ɲ	ŋ
	Aspirated	mɦ		nɦ					ŋɦ
Stop	Voiceless	p		t					k
	Voiceless Aspirated	ph		th					kh
	Voiced	b		d					g
	Voiced Aspirated	bɦ		dɦ					gɦ
Affricate	Voiceless		pf	ts	tɬ	tʃ	ʈʂ
	Voiceless Aspirated		pfh	tsh	tɬh	tʃh	ʈʂh
	Voiced		bv	dz	dɮ	dʒ	ɖʐ
	Voiced Aspirated		bvɦ	dzɦ	dɮɦ	dʒɦ
Fricative	Voiceless		f	s	ɬ	ʃ	ʂ
	Voiced	β	v	z			ʐ			ɦ
Liquid	Plain			r	l
	Aspirated			rɦ
Glide		w						j

draws primarily on the work of Baumbach (1987), whose data hails from the Nkuna and Gwamba dialects. Changana phonology and lexical forms come from the works of Sitoe (1996) and Ngunga and Faquir (2012), whose data derives mostly from the Dzonga dialect. For the most detailed account of the phonological differences between several of the Tsonga dialects, readers should consult Baumbach (1974).

The consonant inventory of Tsonga has often been reported as massive, with Baumbach (1987) citing exactly 100 non-click consonantal phonemes. Part of this can be due to analysis; Baumbach (1987) and Cuénod (1967) both treat palatalization and labialization as phonemic, rather than as reflecting consonants followed by glides. These contrasts exist on top of the voicing, aspiration, and prenasalization contrasts which are also found. If these features are taken to all be mutually compatible, the number of permutations of these features becomes immense. This means that most gaps in the phonological system should be taken as coincidental rather than reflecting any theoretical impossibility.

That said, it is clear that not all features are mutually compatible. Bilabial consonants are never labialized. Lateral, postalveolar, velar, and glottal consonants are not palatalized. Labiodental, “whistled”, and palatal consonants show neither labialization nor palatalization. When it comes to prenasalization, Tsonga is distinct from Copi and Tonga in that voiceless consonants, including fricatives, can be prenasalized. Among the aspirates, it is exceptional that there is a series of prenasalized voiced aspirates which lack a non-prenasalized counterpart. The typological rarity of this is noted in Janson (2001), who suggests that phonetic aspiration here is not phonologically contrastive.

The phonetic character of certain consonants described varyingly as “aspirated” or “breathy” may be called into question, namely the nasals and the rhotic /r^ɦ/. The latter of these was also remarkable to Janson (2001, p. 27), who remarked that “rhotics cannot really be aspirated”. It seems that the primary phonological difference between these consonants and their unaspirated counterparts is their interaction with tone. Lee (2009) states that the aspirated nasals and rhotics are depressor consonants, which block the process of high tone spread; this is a property that these phonemes share with all the other voiceless and voiced aspirates, but not their unaspirated counterparts. Therefore, even if these consonants do not truly have phonetic aspiration, they still display properties which make them distinct phonemes, and they will continue to be marked with a symbol /^ɦ/.

The other set of sounds which deserves attention is the series that Cuénod (1967) and Baumbach (1987) refer to as the “whistled” affricates and sibilants. These are phones which have a distinctive labial element that gives them their whistling sound. Such sounds are found in Tsonga and Tswa, as well as the Southern Bantu language group Shona (S10), but are virtually unheard of outside of the family, only perhaps being attested in certain Caucasian and South Arabian languages (Shosted, 2006). These phonemes seem to be the labialized counterparts of the alveolar affricates and fricatives, and, like the Copi labio-alveolar phonemes, they are often spelled with the retroflex symbols <ʂ> and <ʐ>. The matter of notation is further complicated by allophony within Tsonga; the alveolar affricates may be freely pronounced as retroflex affricates. To make it clear that these phonemes are effectively parallel series, they will all be spelled as retroflex, with the “whistled” affricates being denoted by additional labialization <^w>, and the whistled sibilants simply being written <ʂ> and <ʐ>.

The consonant inventory of Changana is almost identical to that of Tsonga. The most significant difference between the two languages is that Changana shows two mergers (Janson, 2001). The first of these is a merger of the Tsonga palatalized bilabial stops and the whistled affricates, with both of these reflected in Changana as labio-alveolar affricates. The second merger is between the Tsonga palatalized alveolar stops and the postalveolar affricates, which are both reflected in Changana as postalveolar affricates. A schematic of this is given in

Table 4. Phonological correspondences between Changana and Tsonga.

Labio-Alveolars			Alveolars
Changana	Tsonga		Changana	Tsonga
pʂ	pj	tʂw	tʃ	tj	tʃ
mpʂ	mpj	ntʂw	ntʃ	ntj	ntʃ
pʂh	phj	tʂhw	tʃh	thj	tʃh
mpʂh	mphj	ntʂhw	ntʃh	nthj	ntʃh
bʐ	bj	dʐw	dʒ	dj	dʒ
mbʐ	mbj	ndʐw	ndʒ	ndj	ndʒ

below.

It is also worth mentioning that Changana has two additional implosive stops, /ɓ/ and /ɗ/. Noting that all the languages from Mozambique have such stops, but that South African Tsonga lacks them, it remains a question whether this represents a loss in South African Tsonga or a borrowing in Changana.

1.4. Tswa

Tswa (S51), also written as Tshwa or Xitswa, is spoken by 1,000,000 speakers in Mozambique, with an additional 20,000 speakers in Zimbabwe (Eberhard et al., 2022). Ngunga and Faquir (2012) list five varieties of Tswa: Khambani, Rhonga, Hlengwe, Mhandla, Dzhonge/Donge, and Dzivi. As mentioned in the previous section, the Hlengwe variety is treated as a distinct language by Maho (2009), though Ethnologue (Eberhard et al., 2022) and Hammarström (2019) both list Hlengwe as a dialect. The consonant inventory given in

Table 5. The consonant inventory of Tswa.

		Bilabial	Labiodental	Labio-Alveolar	Alveolar	Lateral	Postalveolar	Palatal	Velar	Glottal
Nasal		m			n			ɲ	ŋ
Stop	Voiceless	p			t			c	k
	Voiced	b			d			ɟ	g
	Implosive	ɓ			ɗ
Affricate	Voiceless		pf	pʂ	ts	tɬ
	Voiced		bv	bʐ	dz	dɮ
Fricative	Voiceless		f	ʂ	s	ɬ	ʃ			h
	Voiced		v	ʐ	z		ʒ
Liquid					r	l
Glide		w	ʋ				j

and the lexical items cited from Persson (1928) throughout this paper are drawn principally from the Dzivi dialect.

Unlike the sources consulted for Tsonga, Ngunga and Faquir (2012) and Gundane (2015) do not provide a detailed account of the distribution of prenasalization, aspiration, and labialization in Tswa. In the absence of descriptive linguistic data, some observations about the spellings used in Persson (1928) may provide insights into the phonological system. Spellings <mp>, <nt>, and <nk> occur, suggesting that voiceless stops may be prenasalized. The spellings <ph>, <th>, and <kh> also occur, though never after <m> or <n>, suggesting the presence of voiceless aspirated stops which do not allow prenasalization. The spellings <bh> and <gh> only ever occur after <m> and <n>, suggesting that voiced aspirated stops only occur when prenasalized, as was the case in Tsonga and Changana. Interestingly, the spelling <dh> never occurs. Finally, the spellings <mh> and <nh> occur, indicating at least the presence of aspirated bilabial and alveolar nasals.

Tswa is distinct among the languages under consideration in having the greatest time gap between the dictionary used and the linguistic resources consulted, this being almost a century. Because of this, it is particularly useful to follow Persson’s descriptions of Tswa’s phonology when appropriate. Persson (1928) makes it clear, for example, that phoneme corresponding to /ʋ/ in Ngunga and Faquir (2012) and Gundane (2015) was, at the time, a voiced bilabial fricative [β]. Furthermore, the phoneme /bʐ/ is consistently spelled by Persson as <by>, suggesting a historic palatalized bilabial stop [b^j]. Finally, the palatal stops /c/ and /ɟ/ are claimed to be pronounced like the English affricates [tʃ] and [dʒ]. All of these phonemes are used to transcribe Persson’s data throughout this paper.

1.5. Ronga

Ronga (S54), also written Rhonga or Xironga, is spoken by approximately 618,000 speakers. Almost all of these live in Mozambique, in the region immediately surrounding the capital Maputo, though there is a small group of approximately 1000 speakers living in South Africa, just below the border of KwaZulu-Natal province and Mozambique (Eberhard et al., 2022). As with other S50 languages, the dialectical picture is difficult to decipher. Ngunga and Faquir name four dialects of Ronga: Lwandle/Kalanga, Nondrwana, Zingili, and Hlanganu, with Nondrwana being used as their reference variety. Da Conceição (1999) presents a table from a newspaper article which lists three dialects: Kalanga, Ronga, and Putru. Eberhard also lists three dialects, Konde, Putru, and Kalanga, though it is unclear if this implies that Konde is equivalent to Da Conceição’s Ronga. Maho (2009) gives three Ronga varieties: Konde, Ssonga, and Xonga. While Baumbach (1974) works on what he calls Tsonga varieties, he includes separate entries and phonological inventories for varieties called Konde, Ssonga, and Ronga, among others. Based on the map he includes, his Ronga data comes from the immediate vicinity of Maputo, where Ngunga and Faquir claim Nondrwana is spoken. It is unfortunately unclear which variety Quintão references, if it is any one in particular.

Table 6. The consonant inventory of Ronga.

		Bilabial	Labiodental	Labio-Alveolar	Alveolar	Lateral	Postalveolar	Retroflex	Palatal	Velar	Glottal
Nasal	Plain	m			n				ɲ	ŋ
	Aspirated	mɦ			nɦ					ŋɦ
Stop	Voiceless	p			t					k
	Voiceless Aspirated	ph			th					kh
	Voiced	b			d					g
	Voiced Aspirated				dɦ
Affricate	Voiceless		pf	pʂ	ts	tɬ	tʃ	tʂ
	Voiceless Aspirated		pfh	pʂh	tsh	tɬh	tʃh	tʂh
	Voiced		bv	bʐ	dz	dɮ	dʒ	dʐ
Fricative	Voiceless		f	ʂ	s	ɬ	ʃ
	Voiced	v~β		ʐ	z		ʒ	ʐ			ɦ
Approximant		w				l			j

combines the speech sounds of Ronga listed in Baumbach (1969, 1974). These sources indicate that most non-labial obstruents can be labialized, but do not list any palatalized consonants. Both these and other sources which have been consulted for Ronga do not provide a detailed account of which phonemes can be prenasalized, and so spellings will again be taken into consideration. Quintão (1951) has spellings which suggest that both voiceless stops and voiceless aspirated stops can indeed be prenasalized. Quintão gives no spellings which indicate voiced aspirates, whether prenasalized or not. Baumbach’s example of /d^ɦ/, [nd^ɦawu] ‘place’, does at least point to the existence of prenasalized voice aspirates; Hargus and Da Conceição (1999) also offer the form [ng^ɦwendza] ‘bachelor’. These words are spelled ndawu and nkwendja in Quintão (1951), suggesting that the voiced aspirates are either innovations or restricted to particular dialects.

In addition to the phonemes listed above, Hargus and Da Conceição (1999) note the sparse existence of implosive stops /ɓ/ and /ɗ/, suggesting that such forms only come from Changana loans. Ngunga and Faquir (2012) also list implosives in the inventory of Ronga, and corroborate their rarity within the language.

Unlike most of the languages under consideration, Ronga seems to possess a genuine and contrastive series of retroflex affricates. Cross-referencing lexemes cited in Quintão (1951) and Baumbach (1969) shows that Quintão systematically distinguishes the spelling of the voiceless retroflex affricates from the voiceless postalveolar affricates, but spells both voiced retroflex and voiced postalveolar affricates as <dj>. In this paper, a Ronga form cited with /dʒ/ may therefore actually reflect a retroflex affricate.

It is also pertinent to discuss the representation of the labio-alveolar obstruents in Tswa. The phoneme given as /bʐ/ in Baumbach (1969) and Ngunga and Faquir (2012), and /bʐ^w/ in Hargus and Da Conceição (1999), is consistently spelled <by> by Quintão. This will be represented in this paper as a palatalized bilabial stop /b^j/. The phoneme given as /pʂ/ in Baumbach (1969) and Ngunga and Faquir (2012), and /pʂ^w/ in Hargus and Da Conceição (1999) is consistently spelled <bs> by Quintão. This will be represented as /pʂ/, as it seems unlikely that the two components of the affricate would disagree in voicing. Baumbach (1969) only gives one lexeme attesting /pʂ^h/, which is [pʂ^ha] ‘burn’. Quintão (1951) has two entries for this, spelled bsa and pya. The former will be transcribed as the phoneme /pʂ/, and the latter as a palatalized voiceless bilabial stop /p^j/. As for the labio-alveolar sibilant /ʂ/, the form given as [ʂoʂi] ‘now’ in Baumbach (1969) is spelled bsobsi by Quintão. It is not clear whether this reflects a merger with /pʂ/ or is simply a deficient spelling. Nevertheless, it will be transcribed /pʂ/, and the potential ambiguity should be kept in mind. The voiced labio-alveolar sibilant /ʐ/ is only cited in a clan name, which is unattested in Quintão (1951).

Finally, attention must be called to the phonemes /b/ and /v~β/. Both of these are spelled in Quintão (1951) with <b>. Interestingly, all the reflexes of /b/ cited by Baumbach (1969) occur before /u/, suggesting that these are not two distinct phonemes but rather conditioned allophones of a single phoneme. Nevertheless, both will be reflected as /b/ in this paper as Quintão’s spelling makes it impossible to differentiate.

1.6. Proto-Bantu

The reconstructed consonant inventory of Proto-Bantu, seen in

Table 7. The reconstructed consonant inventory of Proto-Bantu from Meeussen (1967).

	Bilabial	Alveolar	Palatal	Velar
Nasal	*m	*n	*ɲ
Voiceless Stop	*p	*t	*c	*k
Voiced Stop	*b	*d	*j	*g
Geminate Nasal	*mm	*nn	*ɲɲ
Prenasalized Voiceless Stop	*mp	*nt	*ɲc	*ŋk
Prenasalized Voiced Stop	*mb	*nd	*ɲj	*ŋg

, is exceptionally small in comparison to all of Tswa-Ronga and Copi-Tonga. In practice, these phonemes likely had a variety of allophonic variants, and Meeussen (1967, p. 83) suggests *d might as well have been transcribed as *l, *c as *s, and *j as *z or *y. As the language developed, allophony at this level would be vulnerable to sound changes which might phonemicize it. Yet another significant source of development in the daughter Bantu languages were nasal/stop combinations. These sequences often formed units with distinct phonological outcomes, as will be seen for Tsonga-Copi in the subsequent sections.

Proto-Bantu is reconstructed with a 7-vowel inventory as shown in

Table 8. The reconstructed vowel inventory of Proto-Bantu.

	Front	Back
Close	*i	*u
Near-Close	*ɪ	*ʊ
Mid	*e	*o
Open	*a

. This includes the canonical five vowels *a, *e, *i, *o, *u, as well as two near-close vowels *ɪ, and *ʊ (Bastin et al., 2002).

The PB close vowels, when following a stop, triggered a spirantization process which generally yielded affricates or fricatives in the daughter languages (Bostoen, 2008). The immediate result of this change would have been allophonic, but it was often phonemicized by a subsequent merger of the close and close-mid vowels, the 7-to-5 vowel merger (Bostoen, 2008). While not every language underwent spirantization or the 7-to-5 vowel merger, every language which underwent the merger also underwent spirantization (Schadeberg, 1994). The Tswa-Ronga and Copi-Tonga languages all underwent both changes, and all languages in each family have a 5-vowel system. As will be seen in Section 3, spirantization is a major source of innovation in the consonant inventory of PTC and its daughter languages.

2. Materials and Methods

The first step was to assemble an initial list of lexemes to search for in each language. These were compiled from a composite of the 200-word Swadesh list (Swadesh, 1952), the entries in the word list in Gunnink et al. (2022), and the glosses in Van der Spuy’s (1989) reconstruction of Proto-Southern Bantu.

Seven dictionaries were then consulted to identify corresponding lexemes in the daughter languages, these being Amaral et al. (2007), a bidirectional Portuguese-Tonga dictionary; Dos Santos (1949), a bidirectional Portuguese-Copi dictionary; Smyth and Matthews (1902), a unidirectional Lenge-English dictionary; Cuénod (1967), a unidirectional Tsonga-English dictionary; Sitoe (1996), a unidirectional Changana-Portuguese dictionary; Persson (1928), a unidirectional English-Tswa dictionary; and Quintão (1951), a bidirectional Portuguese-Ronga dictionary. Many of the dictionaries selected did not include sufficiently descriptive information about the phonology of the languages, so additional sources had to be sought out to convert the orthographic representations into phonological transcriptions. The lexemes were entered into the list according to the principles outlined in Section 1.1, Section 1.2, Section 1.3, Section 1.4 and Section 1.5. Tone information was not included, as only the Tsonga and Changana dictionaries (inconsistently) marked it.

Whenever a dictionary entry gave multiple translations for a given lexeme, all were entered into the list of results. For those dictionaries which were bidirectional, all English or Portuguese terms in the reverse direction were also copied and searched for, in order to control for slight semantic alterations. This ultimately created something of a word “web”, with one entry having multiple potential translations in English, Portuguese, and the various Bantu languages.

After this initial set of words had been obtained, additional words were added to the list; search items were chosen by going through each dictionary to find words containing rarer phonemes or sequences, such as different combinations of aspiration, palatalization, and labialization. Only those words which did not seem to have meanings too unclear or too abstract were chosen. Once these were added to the list, their correspondents in the other languages were filled in. The resulting list constituted the final search list.

Next, the data was pruned. For each lexeme, it was considered a valid cognate series if there was at least one correspondent from Tonga and Copi/Lenge, and two correspondents from Tswa, Ronga, and Tsonga/Changana1. If a lexeme had multiple translations, and these seemed to show multiple cognate sets, these were then separated into unique entries. Given the absence of prior research on phonological correspondences between the given languages, correspondence was determined in an impressionistic manner. The potential for error that such a process invites should be noted. With the resulting data, the Bantu Lexical Reconstructions 3 (BLR3) database (Bastin et al., 2002) was searched to find a reconstruction2 corresponding to each lexeme, if possible. The resulting cognate sets were then used to determine which sound changes took place in the languages under consideration and determine a Proto-Tsonga-Copi reconstruction, the results of which are given in the following section.

3. Results

The following section details the outcomes of each of the traditionally reconstructed Proto-Bantu phonemes. Prenasalized stops are treated in the same section as their oral counterpart. For each phoneme, an overview of the identified sound changes is given, followed by tables with illustrative cognate sets of the relevant phoneme present in a variety of conditioning environments. A proposed Proto-Tsonga-Copi reconstruction is provided for each set; a class prefix is present in nominal reconstructions when the daughter languages point to the same original noun class; otherwise, these reconstructions begin with a bare hyphen. Finally, an existing Bantu reconstruction from the BLR3 database is included with each entry where a likely match was identified. These forms are given directly as cited in BLR3; this means that final vowels are generally omitted from verbal forms, and nominal forms do not include a class prefix.

3.1. *p

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC
ph	ph	p	mɦ	mɦ	mɦ	mɦ	*mph	*N_
s	s	s	f	f	f	f	*ps	*_i
tsh	tsh	tʃ	ns	ns	s	ns	*psh	*N_i
f	f	f	f	f	f	f	*f	*_u
ph	ph	p	ɦ	h	h	ɦ	*ph	*V(#)i_ in nouns
s	psw	sw	ʂ	ʂ	ʂ	ʂ	*phj	*V(#)i_ in verbs
βj	ɦj	hj	ʈʂhw	pʂh	ʂ	pʂ/pj	*bɦj	*_ɪV
phj	phj	phj	ɳʈʂhw	mpʂh	ʂ	mpʂ/ntʃ	*mphj	*N_ɪV
βw	phw	hw	phj	pʂh	pʂ	-	*bɦw	*_ʊV
p	p	p	p	p	p	p	*p	early borrowings?
β	ɦ	h	ɦ	h	h	ɦ	*bɦ	elsewhere

As seen in

Table 9. Reflexes of PB *p in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
give	βa	ɦa	ha	ɦa	ha	ha	-	*bɦa	*pá
thresh	βula	ɦula	hula	ɦula	hula	hula	ɦula	*bɦula	*pʊ́ʊd
nerve	mi-siβa 1	in-siɦa	-	n-siɦa	n-siha	siha	n-siɦa	*mu-sibɦa	*kìpà
end	βedza	ɦeta	heta	ɦeta	heta	-	ɦeta	*bɦeta	*pédi
near	βa-fuβi	ɦa-fuɦi	ku-fuhi	ku-suɦi 2	ku-suhi	ku-suhi	ku-suɦi	*-fubɦi	*kúpí
live	ꞵaɲa	ɦaɲa	haɲa	ɦaɲa	haɲa	haɲa	ɦaɲa	*bɦaɲa	-
fruit	handro 3	ɦando	handu	m-bɦaɳɖʐu	m-bɦandzu	mɦanzu	mɦandʒu	*mu-bɦando	-
finger	-	li-tiɦo	li-tsiho	ri-tiɦo	ri-tiho	li-tiho	li-tiɦo	*li-dzibɦo	-

¹ This term, cited as ‘muscle’ comes from Lanham (1955). Amaral et al. (2007) offer the possibly related tsʰiꞵa ‘power’ (cited as class 15) and siꞵa ‘hernia’ (class unclear, but likely 3). ² The Tswa-Ronga reflexes here contain an /s/ instead of an expected /f/. BLR3 lists numerous derivatives of the root *jípí ‘short’, including *kúpí cited above. It is possible that Tswa-Ronga reflects a form phonologically influenced by other derivatives of the same root. ³ Lanham (1955, p. 55) states that the regular outcome of the prefix mu- and root initial /h/ in Tonga is /ɦʷ/, though he notes that that pattern was disappearing from the speech of younger speakers. It is therefore unsurprising that such forms are not found in Amaral et al. (2007).

, almost all the languages have glottal fricatives as reflexes of PB *p, but Tonga retains a bilabial sound. Furthermore, both Tonga and Copi reflexes are voiced, along with those of Tsonga and Ronga. The chain from PB into PTC would therefore likely be *p > *p^h > *b^ɦ. The aspiration of the voiceless stop to *p^h is well attested for Southern Bantu as a whole (Janson, 1991–1992), but the subsequent voicing development would represent an innovation within Tsonga-Copi. The intermediate voiced stop is lent support by synchronic pairs in Tsonga like m-b^ɦaɲi ‘living person’, a class 1 agentive derivation from ɦaɲa ‘live’, as well as the cognate set for ‘fruit’.

The complex *mp, which is typically attested as a combination of the class 9 prefix *N- with nominal roots beginning in *p, is reflected by /p^h/ in Copi-Tonga and /m^ɦ/ in Tswa-Ronga as shown in

Table 10. Prenasalized reflexes of PB *p.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
billy-goat	phoŋgo	phoŋgo	poŋgo	mɦoŋgo	mɦoŋgo	mɦoŋgo	mɦoŋgo	*m-phoŋgo	*pòngó
tree hollow	-	phako	-	mɦako	mɦakwa	mɦakwa	mɦakwa	*m-phako	*pàkò
wind	-	pheɦo	puho	mɦeɦo	mɦeho	mɦeho	mɦeɦo	*m-phebɦo	*pép

. This shows a development of *mp > *mp^h, with no further shared development. In Copi-Tonga, the nasal component was then lost, whereas Tswa-Ronga lost the obstruent, with both retaining aspiration. In the latter, this yields an outcome partially overlapping with that of *mʊ-p seen above in the set for ‘fruit’, though Tsonga/Changana is still distinct.

The reflexes of PB *p followed by the close vowels *i and *u trigger spirantization, as discussed in Section 1.6. As seen in

Table 11. Reflexes of PB *p showing spirantization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
conceal	siha	sisa	sisa	fiɬa	fiɬa	fiɬa	fiɬa	*psiɬa	*píc
kidney	ji-tshwa	ji-tsho	i-tʃo	jin-so	jin-so	ji-ʂa	jin-so	*jim-psho	*pígò
resemble	fana	fanana	fana	fana	fana	fanana	fana	*fana	*púan
turn	-	fularela	-	fularɦela	fularɦela	fularela	fulaʐela	*fuladɦela	*púdat

, PB *pi seems to yield /s/ in Copi-Tonga and /f/ in Tswa-Ronga. Isolated examples seem to offer support, such as Tonga simbo ‘branch’ < PB *pímbò, and Tsonga fika ‘arrive’ < PB *pìk. This suggests a complex consonant or affricate at the PTC level, which had both a labial and an alveolar component; this phoneme in PTC will be represented as *ps. This phoneme appears to have further distinct reflexes after a nasal, such as the class 9 prefix in *m-pígò. The Tonga reflex of this lexeme suggests aspiration, which is paralleled by the outcomes and reconstructions of other nasal + voiceless consonant series. It will therefore be reconstructed as *m-ps^h. As for PB *pu, all languages show an outcome of /f/. This is identical to the situation for PB *ku sequences (see Section 3.4), though the reflexes of *ki are not the same as those of *pi. This means that an allophonic alternation inherited from PSB would have been phonemicized due to partial merger, and that the resulting phoneme should be reconstructed as *f.

The vowel *i also causes unique outcomes in certain cases when immediately preceding a consonant. As Wills (2025) extensively details, this occurs when *i would be expected to occur in hiatus. As Proto-Bantu allowed only open syllables, this would mean any word-initial *i would be a trigger. This generally arises as a result of the noun class 5 prefix *i-, or in certain lexical items where initial *ji is traditionally reconstructed (Wills instead reconstructs *i).

The tendency for class 5 words to show initial mutation was already noticed by Lanham (1955). Van der Spuy (1989, p. 136) observed a similar phenomenon in the Southern Bantu languages Northern Sotho, Venda, and Zezuru. He describes the effects in these languages as “to insert an offglide after a following stop consonant,” although that does not seem to describe the situation for Tonga nouns, where the mutation creates an outcome that often resembles what would be expected of an initial nasal. The class 5 prefix seems to stop the chain of PB *p > *p^h > *b^ɦ before the voicing step, as did the nasal prefix.

Table 12. Reflexes of PB *p following *i.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
twin	li-βaha	ma-ɦasa	ma-hasa	haɬa	haɬa	haɬa	ɦaɬa	*-phaɬa	*pácà
cockroach	li-βele	ɗi-pʰele	-	ɦele	hele	hele	ɦele	*-phɬele	*pèdè
cave	li-pʰala	pʰala	pala	mpala	mpala	-	mpala	*-phala	-
cook	-	bika	bika	ʂeka	ʂeka	bika	pʂeka	*phjeka	*jìpɪk
porridge	wu-sa	wu-pswa	vu-swa	ꞵu-ʂa	ꞵu-ʂa	ꞵu-ʂa	bu-pʂa	*ꞵu-phja	-
ask, talk	bula	bula	-	ꞵula	bula, ꞵula	bula	bula	*phjula?	*jípʊd

shows that the effects of class 5 do not apply equally across Tsonga-Copi. They are rarely seen in Tsonga, and only occasionally seen in Copi/Lenge. They are most frequently encountered in Tonga, where class 5 alternation appears to have at one point been relatively robust morphophonological process. However, Lanham (1955, p. 71) notes that such forms were already starting to disappear, which he attributes to the disruptive absorption of class 11 words with no alternation into class 5. Analogy seems to be the major culprit for the lack of expected mutation in many forms; their class 6 plural forms would not show any changes, and analogy with these forms would inevitably lead to some amount of leveling. This would be most apparent in words like *pácà ‘twin’ which appear much more frequently in the plural. Nevertheless, it is clear that at the level of PTC these outcomes were distinct, and they will be reconstructed as *p^h.

While initial *i in nouns has been accounted for, the verbs still pose a problem. Two outcomes are seen, one that generally yields a voiced bilabial stop /b/ and another that yields a labiodental affricate or labialized sibilant. One reviewer suggested that bika ‘cook’ is instead a loanword from Shona, making the reflexes with /(p)ʂ/ the inherited forms. While it may seem dubious for such a basic term to be replaced, Guthrie (1971, p. 188) notes that the meaning of *jìpɪk may have been more specific, something akin to ‘stew’. It is notable that the reflexes of *jip strongly resemble the reflexes of *pɪa seen in

Table 13. Reflexes of PB *p followed by a semivowel.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
burn	βja	ɦja	hja	ʈʂhwa	pʂha	ʂa	pʂa/pja	*bɦja	*pí
new	phja 1	phja	phja	ɳʈʂhwa	mpʂha	ʂa	mpʂa/ntʃa	*mphja	*píà
sweep	βijela	ɦjela	hijela	-	pʂhajela	ʂijela	pʂeɦela	*bɦjajela	*píagɪd
dry up	βwa 2	phwa	hwa	phja	pʂha	pʂa	-	*bɦwa	*pʊ́

¹ The reflexes of PB *píà unanimously indicate the presence of a nasal at the beginning of the root (Tswa appears to have deleted nasals before voiceless consonants). This is unetymological, and perhaps comes from rebracketing with an agreement prefix. ² This entry comes from Lanham (1955), as no translation was given by Amaral et al. (2007).

below, suggesting the presence of a glide, as has been posited by Van der Spuy (1989, p. 136) and Wills (2025, p. 11). In parallel with the nouns, as well as the outcomes of PB *t and *k in similar environments, it seems that PB *p was first aspirated in this environment, at which point the consonant underwent metathesis with the preceding vowel, which then became a glide. This can be schematized as PB *ip > PTC *p^hj. In the daughter languages, the labial consonant underwent further palatalization in a process described in detail by Ohala (1978). As seen below and in Section 3.5, glides do not generally palatalize labials in Copi-Tonga. The lack of voicing does not seem to be a factor, as the PB class 8 prefix *bi- yields Copi-Tonga si-, Tsonga/Changana ʂi-, Ronga pʂi, and Tswa ʐi-. More data is necessary to determine whether there is another conditioning factor present, or if perhaps there were two separate historical palatalizations. Finally, as for the apparent reflexes of *jípʊd, more terms are needed to determine whether there is a unique outcome of PTC *phj before *u, or if the entries here are instead loanwords in one or both language groups.

In PB syllables of type *CɪV and *CʊV, the near-close vowels *ɪ and *ʊ became the glides *j and *w, respectively. While the reflexes across Tswa-Ronga show significant further development of the sequences *pj and *pw, the expected development of *p into /β/ in Tonga and /h/ in Lenge make it clear that at the level of PTC, these stop-glide sequences behaved as two distinct phonemes. Therefore, the development of PB *p described earlier should be applied, and these sequences should be reconstructed as *b^ɦj and *b^ɦw, respectively.

There remains a group of words reflecting PB *p which always yield unaspirated /p/ in the daughter languages with no apparent conditioning element. It is possible that these words, which are listed in

Table 14. Reflexes of PB *p with aberrant synchronic /p/.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
split	pandra	-	banda	paɳɖʐa	pandza	panza	pandʒa	*panda	*pànd
measure	pima	pima	pima	pima	pima	pima	pima	*pima	*pìm
fold	peta	peta	peta	peʈʂa	petsa	petsa	-	*peta	*pèt
skull	-	li-pala	-	ri-pala	ri-pala	ʃi-pala	pala	*lu-pala	*pádá

, reflect late PTC borrowings from other Bantu languages, and there is a typological argument for why they may have been integrated; the unconditioned change from *p to *b^ɦ would give the language a series of voiced and voiced aspirate stops, but only prenasalized voiceless stops. This seems like an implausible system, but the issue is remedied by positing the existence of a voiceless stop series unexplainable through direct inheritance. The identity of the potential donor language is discussed in Section 4.3.

3.2. *t

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC
th	th	t	nɦ	nɦ	nɦ	nɦ	*nth	*N_
tsh	s	s	s	s	s	s	*ts	*_i
pfh/f	pʂh/tsh/f	prh/tsh/f	f	f	f	f	*f	*_u (tentative)
th	th	t	rɦ	rɦ	r	ʐ	*th	*V(#)i_ in nouns
tsh	tsh	ts	s	s	s	s	*tsh	*Vi_i,
th	th	t	tʃh	tʃh	tʃ	tʃ	*thj	*V#i_ in verbs, *_ɪV
rw	rw	hw	rɦw	rɦw	rw	ʐw	*dɦw	*_ʊV
t	t	t	ʈʂ	ts	ts	ʈʂ	*t	early borrowings?
r	r	h	rɦ	rɦ	r	ʐ	*dɦ	elsewhere

Unconditioned outcomes of PB *t are given in

Table 15. Reflexes of PB *t in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
heel	-	tʃi-nrende	tʃi-hende	ʃi-rɦeɳɖʐe	ʃi-rɦendze	ʃi-renze	ʃi-ʐendʒe	*ki-dɦende	*téndé
gravy	mu-ro	in-ro	mu-ho	mu-rɦu	mu-rɦu	mu-ro	mu-ʐu	*mu-dɦo	*tò
buffalo	ɲari	ɲari	ɲahi	ɲarɦi	ɲarɦi	ɲari	ɲaʐi	*ɲadɦi	*játí
saliva	-	ma-ri	ma-hi	ma-rɦi	ma-rɦi	ma-ri	ma-ʐi	*ma-dɦi	*táì
love	-	randa	handa	rɦaɳɖʐa	rɦandza	ranza	ʐandʒa	*dɦanda	-

. In most environments, PB *t is reflected by a rhotic in all the daughter languages except Ronga (where it is spelled <r>, suggesting a recent change from *r > /ʐ/) and Lenge. Interestingly, Tsonga and Changana have a breathy trill /r^ɦ/. This contrasts with /r/, which appears in reflexes of PB *dɪ (see Section 3.6). Tsonga and Changana thus show a retention of aspiration from PTC rather than an internal innovation. If we assume that the voiceless stops underwent changes relatively in parallel, then an unconditioned change of PB *t > pre-PTC *t^h along the lines of *p > *p^h would be expected, with *t^h subsequently developing into *r^ɦ. If one expects total parallelism with *p, then it makes sense to also assume an intermediate voicing step, making the full sequence PB *t > *t^h > *d^ɦ > *r^ɦ. As with PB *p, synchronic support for such a step can be found in the Changana preposition ndz^ɦaku ‘behind’3, which appears to reflect the PB root *tákò ‘buttock’ with a class 18 prefix4. Additionally, the Tsonga and Tswa class 3 words for ‘shadow’, ɳ-ɖʐ^ɦuti and n-z^ɦuti, exist alongside non-class 3 counterparts mi-r^ɦuti (Tsonga) and ʃi-ruti (Tswa). This suggests allophonic variation between *d^ɦ and *r^ɦ in PTC, which was continued into Tswa-Ronga. Additional support from Copi-Tonga is found in Lenge, where the regular outcome is /h/5. Considering this, as well as the likely parallelism between *p and *t as voiceless stops, the reflex of PB *t will be reconstructed as *d^ɦ in PTC, though it was already likely realized as [r^ɦ] in most positions.

When following a nasal, PB *t yields /t^h/ in Copi-Tonga and /n^ɦ/ in Tswa-Ronga, as shown in

Table 16. Prenasalized reflexes of PB *t.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
person	mu-thu	in-thu	mu-ntu	mu-nɦu	mu-nɦu	mu-nɦu	mɦu-nu	*mu-nthu	*ntʊ̀
neck	thamo	thamu	tamu	nɦamu	nɦamu	nɦamo	-	*n-thamo	*tàm
three	-raru	-raru	-raru	-nɦarɦu	-nɦarɦu	-nɦaru	-nɦaʐu	*n-thadɦu	*tátʊ̀
nose	tʰombvu	tʰombvu	tomvu	nɦompfu	nɦompfu	nɦovu	nɦompfu	*n-thomvu	-

. This directly parallels the outcome of *mp, and its phonological history should be reconstructed accordingly. The resultant change was thus PB *nt > PTC *nt^h.

There are also clear spirantized reflexes of *t. PB *t followed by *i,yielded /ts^h/ in Tonga and /s/ in all the other languages. In Tonga, /ts^h/ and /s/ are separate phonemes, the latter being a reflex of PB *p before *i (cf. Section 3.1), and, in some instances, *c (cf. Section 3.3). The Tonga reflex thus cannot be explained as an internal development of /s/ to /ts^h/, and should therefore form the basis of the reconstruction for PTC. The phoneme *ts will be reconstructed, which was ultimately simplified to /s/ in Copi and Tswa-Ronga. The reflex of *tu in PTC is more ambiguous; PB *tu in Tswa-Ronga consistently yields /f/, but the situation is less clear for Copi and Tonga. The reflexes of PB *tù ‘cloud’, the fifth entry in

Table 17. Reflexes of PB *t showing spirantization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
night	wu-tshiɣu	wu-siku	vu-siku	βu-siku	βu-siku	βu-siku	bu-siku	*ꞵu-tsiku	*tíkʊ́
remain	tshala	sala	sala	sala	sala	sala	sala	*tsala	*tígad
wipe/twist	tshula	sula	-	sula	sula	sula	sula	*tsula	*tíʊ̀d
leave	-	suka	-	suka	suka	suka	susa 1	*tsuka	*tíg(-ʊk)
cloud	li-pfhi	ɗi-pʂhi 2	di-pri	ri-fu	ri-fu	re-fu	-	*-fu?	*tù
forge	pfhula	tshula	tshula	fula	fula	fula	fula	*-fula?	*túd
breed	fuja	fuja	fuja	fuja	fuja	fuja	fuja	*fuja	*túg

¹ Only a causative reflex of PB *tíg is attested in Ronga. ² Bailey (1995) gives this form without aspiration.

below, point to a labial affricate of some sort, but the final vowel /i/ is difficult to explain; it is possible that these words are then derived from a different root. It is also possible that the Copi-Tonga forms have been mutated by the class 5 prefix *i-. The reflexes of PB *túd somewhat parallel *tù, but again the Copi-Tonga data is unclear. The Tonga form may reflect pure inheritance, but it may also show influence of an original preceding reflexive pronoun *i, if /pf^h/ is assumed to result from mutation like in *tù. The Copi and Lenge variants are harder to explain, but are perhaps a case of dialectical variation; Bailey (1995, p. 142) mentions that the phoneme that his consultant pronounces as /pʂ/ is pronounced in most other dialects as /tsw/. It is not hard to imagine a parallel situation for the aspirated counterpart of /pʂ/, with the lack of a corresponding glide in ts^hula due to coalescence with the following /u/. It is also possible that the Copi and Lenge forms are due to morphological complexity, with the Copi form only attested in a single verb angatshula ‘he can forge’ in Bailey (1976, p. 70) and the Lenge form being a class 5 noun meaning ‘hearthstone’, again with potential influence from the class 5 prefix. PB *túg consistently shows /f/, though, given the overall unclear data in Copi-Tonga, it is difficult to say whether their terms could be loanwords. For now, PTC *f will be taken as the very tentative reflex of PB *tu, merging with *pu and *ku, as most of the aberrant Copi-Tonga reflexes have plausible alternative morphophonological explanations.

As with other stops, class 5 mutations are found and given in

Table 18. Reflexes of PB *t after *i.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
day	li-tshiɣu	ɗi-tshiku	di-siku	siku	siku	siku	siku	*tshiku	*tíkʊ̀
trunk	li-tshina	ɗi-tshina	n-tsiɲa	n-siɲa	n-siɲa	siɲa	n-siɲa	*tshina	*tínà
buttock	li-thaɣu	ɗi-thaku	taku	rɦaku,	rɦaku	ma-rako	ʐako	*thako	*tákò
cheek	-	ɗi-thama	tama	rɦama	rɦama	rama	li-ʐama	*thama	*támà
bone	li-thambo	ɗi-rambo	r(ɦ)ambu	rɦambu	rɦambu	rambo	ʐambu	*thambo	-
pour	thela	thela	tela	tʃhela	tʃhela	tʃela	tʃela	*thjela	*jìtɪd

. These yield /ts^h/ in Tonga and Copi before *i, and /t^h/ before *a or perhaps elsewhere. These will be projected directly back as *ts^h and *t^h, with the lack of parallel in Tswa-Ronga attributed to analogy or later developments. The Tswa-Ronga reflexes of *jìtɪd support the presence of an off-glide, as was seen for *p. This glide must have been lost in Copi-Tonga.

Unfortunately, there are not many clear examples of *t before glides. PB *t does not appear to have a unique outcome before a labial glide, with the daughter languages pointing simply to *d^ɦw in PTC. The reflexes of *tíab in

Table 19. Reflexes of PB *t followed by a semivowel.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
carry	rwala	rwala	hwala	rɦwala	rɦwala	rwala	ʐwala	*dɦwala	*tʊ́ád
fear	thav’a	thaʋa	tava	tʃhaβa	tʃhaβa	tʃaβa	tʃaba	*thjaꞵa	*tíab

are identical to those of *jìtɪd above, indicating that before a palatal glide PB *t merged with *it into PTC *t^hj. This merger aligns with Wills’ (2025) claims of *ɪ conditioning similar changes to *i.

Table 20. Reflexes of PB *t with aberrant synchronic /t/ and /ts/.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
fold	peta	peta	peta	peʈʂa	petsa	petsa	- 1	*peta	*pèt
price	teŋgo	n-teŋgo	n-teŋgo	ɳ-ʈʂeŋgo	n-tseŋgo	tseŋgo	n-ʈʂeŋgo	*mu-teŋgo	*téngò
feather	teŋga	n-teŋga	n-teŋga	ɳ-ʈʂeŋga	-	tseŋga	n-ʈʂeŋga	*mu-teŋga	*tèngá

¹ The term ʃi-peʈʂwa ‘twisted grass cord’ may reflect this root.

gives three PB lexemes for which the Tsonga-Copi reflexes are aberrant but corresponding. In these sets, PB *t is reflected by Copi-Tonga /t/ and Tswa-Ronga /ts~ʈʂ/, regardless of the presence of a nasal. One of these items, *pèt, also shows diachronically unexplainable /p/ (see Section 3.1). These words appear to represent a class of loanwords which entered Tsonga-Copi after the reflexes of original PB *p and *t had been aspirated or voiced. Critically, the phoneme that these sounds represent, at whatever stage of Copi-Tonga they were borrowed into, was distinct from the phoneme *z yielded by PB *j and *g (see Section 3.7 and Section 3.8), or else one would expect identical reflexes of these sounds in Tswa-Ronga.

3.3. *c

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC
ndz?	tsh	ts	nɬ	nɬ	ɬ	nɬ	*nɬ	*N_
tsh	tsh	ts	ns	ns	s	ns	*nɬ	*N_i, *N_ɪ
tsh	tsh	ts	ɬ	ɬ	ɬ	ɬ	*iɬ	*i_
tsh	tsh	ts	s	s	s	s	*iɬ	*i_i, *i_ɪ
ɦw	ʃw	?	ɬw	ɬw	ɬw	ɬw	*ɬw	*_ʊV, *_oV
h	s	s	s	s	s	s	*ɬ	*_i, *_ɪ
s	s	s	ɬ	ɬ	ɬ	ɬ	*ziɬ	*iji_
h	s	s	ɬ	ɬ	ɬ	ɬ	*ɬ	elsewhere

Table 21. Reflexes of PB *c in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
laugh	heɣa	seka	seka	ɬeka	ɬeka	ɬeka	ɬeka	*ɬeka	*cèk
lose	-	lasa	lasa	laɬa	laɬa	laɬa	laɬa	*laɬa	*dác
hide	siha	sisa	sisa	fiɬa	fiɬa	fiɬa	fiɬa	*psiɬa	*píc

shows that all the daughter languages have voiceless fricatives as unconditioned reflexes of PB *c. Interestingly, the fricatives in Tswa-Ronga are all laterals, which are typologically quite rare sounds (Maddieson, 2013; Van der Vlugt & Gunnink, 2025). The question then remains, which of these situations (if either) is most like the original? To solve this, one must do a bit of bottom-up reconstruction. The reflexes in Tswa-Ronga unambiguously point to *ɬ. For Copi-Tonga, it seems instinctive to posit that Tonga /h/ results from debuccalization of an original *s, reflected directly in Copi. This debuccalization did not seem to occur before Tonga /i/, in which context the reflex of *c effectively merged with that of *p and *k.

The question is now whether *s or *ɬ is original, if either. Kümmel (2007, p. 248) states that lateralization of fricatives is quite rare, and he lists no examples of the unconditioned change of /s/ into /ɬ/. On the other hand, there are numerous attested examples of the opposite change, /ɬ/ becoming /s/ without condition. This would mean that a situation with an *ɬ in PTC is most supported cross-linguistically. At this point, it is important to take into consideration that PB *c has different reflexes when it occurs before *i and *ɪ, given in

Table 22. Reflexes of PB *c before *i and *ɪ.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
feather	-	li-siʋa	-	ri-siβa	ri-siβa	-	li-siba	*li-ɬiꞵa	*cɪba
raw	-mbihu	-mbiso	-mbisi	-mbisi	-mbisi	-mbisi	-mbisi	*-mbiɬi	*bícì
hair	-	in-sisi	n-sisi	n-sisi	n-sisi	sisi	n-sisi	*mu-ɬitsi	*cítì

below. These reflexes have /s/ in all languages except Tonga, which shows subsequent debuccalization. The two scenarios are that an original *s became /ɬ/ in Tswa-Ronga when not before synchronic *i, or that an original *ɬ became /s/ unconditionally in Copi-Tonga and only before synchronic *i in Tswa-Ronga. Given the high typological probability of the change *ɬ > /s/, the latter will be assumed to be correct, and it seems equally likely that the change took place independently in Tswa-Ronga and Copi-Tonga or that it had already begun in PTC, progressing further in Copi-Tonga after it had split from Tswa-Ronga6.

When following a nasal, Tonga and Copi show an aspirated affricate /ts^h/, whereas Tswa-Ronga have an alternation of /s/ and /ɬ/, as seen in

Table 23. Prenasalized reflexes of PB *c.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
genet	tshimba	tshimba	tsimba	n-simba	n-simba	simba	n-simba	*n-ɬimba	*címbá
squirrel	tshindri	-	tʃindi	n-siɳɖʐi,	n-sindzi	ʃinzi	n-sindʒi	*n-ɬindi	*cíndí
underneath	βa-βatshi	ɦa-ɦatsi 1	hansi	ɦansi	hansi	la-hasi	ɦansi	*bɦa-n-ɬi	*cí
spark	tshahe	-	-	n-ɬaɬe	n-ɬaɬe	ɬaɬe	n-ɬaɬe	*n-ɬaɬe	*cácè
python	tsharo	-	- 2	n-ɬarɦu	n-ɬarɦu	n-ɬaru	n-ɬaʐu	*n-ɬadɦu	*cátò
fish	-	tshandzi	tsanzi	n-ɬampfi	n-ɬampfi	-	n-ɬampfi	*n-ɬambzi	*càmb

¹ The lack of aspiration in this form is aberrant, but can perhaps be explained as dissimilation from the /ɦ/ in the preceding syllable. ² The loanword ɬar^ɦu is found in Lenge.

. As in other environments, this appears to be conditioned by the following vowel. For now, it does not seem necessary to reconstruct this sequence as anything besides *nɬ, with the Copi-Tonga reflexes showing development of a cluster *n(t)s < PTC *nɬ.

Reflexes of *c followed by *ʊ are problematic. While the Copi-Tonga cognates generally behave, Tswa-Ronga seems to alternate between /ɬ/ and /s/ or /ʃ/. This happens both when comparing different lexemes, with *cʊ̀ʊj in

Table 24. The reflexes of PB *c before ʊ.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
strain	huta	-	suta	ɬuta	ɬuta	ɬuta	ɬuta	*ɬuza	*cʊ̀ʊj
venom	wu-huŋgu	wu-ʃuŋgu	-	βu-ʃuŋgu	βu-ɬuŋgu	wu-ʃuŋgu	bu-ɬuŋgu	*ꞵu-ɬuŋgu	*cʊ́ngʊ́
leech	tshundrwa	tshundo	-	n-ʃuɳɖʐu	n-sundzu	sunzu	-	*n-ɬundu	*cʊ́ndʊ̀è
mosquito	tshuna	tshuna	tsuna	n-suna	n-suna	suna	n-suna	*n-ɬuna	-
wart	tshuŋgu	-	-	n-ɬuŋgu	n-ɬuŋgu	ɬuŋgu	-	*n-ɬuŋgu	-

yielding /ɬ/ and *cʊ́ndʊ̀è yielding /s/, but also when comparing a single lexeme between different languages, such as with *cʊ́ngʊ́. In light of these facts, it seems that whatever development is occurring here is secondary, perhaps reflecting a still ongoing process of replacing original *ɬ.

The reflexes of *ic in

Table 25. Reflexes of PB *c after *i.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
pestle	mu-tshi	in-tshi	mu-tsi	mu-si	mu-si	mu-si	mu-si	*mu-iɬi	*jìcì
branch	-	ɗi-tshandzu	-	ɬampfu	ɬampfu	ɬavu	ɬampfu	*iɬandzu	*cándú
eye	li-so	ɗi-so	di-so	ti-ɬo	ti-ɬo	ti-ɬo	ti-ɬo	*zi-ɬo	*jícò

are generally the same as *nc. An exception is *jícò, where Copi-Tonga has /s/ instead of /ts^hi/. Comparing the Tonga singular li-so to its plural ma-ho with regular /h/ makes it clear that the class 5 prefix *i is the cause. The preservative effect of *i on *c is noted by Wills (2025, p. 16), but it remains to be explained why the outcomes are inconsistent. At the moment, it seems that the easiest way out is to posit that /s/ is the reflex of the phonological complex *i-jic (or *i-zic), though alternative suggestions are welcome. As for the reconstruction of *ic, an aspirated *ɬʰ would parallel other series, but itself does not seem to be phonetically possible. For this reason, I will use *ⁱɬ with a preceding superscript for this variant of *ɬ as a notational convention.

It is possible that Copi shows a unique development of PB *c followed by a rounded vowel. In these environments, Copi has /ʃ/ with additional labialization if followed by another vowel. It seems that this is a development internal to Copi that is subject to dialectical variation. The Copi verb for ‘set (of the sun)’ is given by Dos Santos (1949) as pʂ^wa and Bailey (1995) as ʂa, with a Lenge equivalent ʃʷa (< PB *cò-a, c.f. Tonga ɦʷa, Tswa ɬʷa). Applicative forms of this verb meaning ‘delay’ are found in most languages and given in

Table 26. Labialized reflexes of PB *c, showing palatalization in Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
termite	mu-ɦwa	in-ʃwa	- 1	mu-ɬwa	mu-ɬwa	mu-ɬwa	-	*mu-ɬwa	*cʊ́á
delay	ɦwela	ʃwela	-	ɬwela	ɬwela	ɬwela	ɬwela 2	*ɬwela	*cò
wild dog	- 3	ɗi-ʃolwa	-	ɬolwa	ɬolwa	ɬolwa	-	*ɬolwa	-

¹ Smyth and Matthews (1902) give a collective vu-sohaha ‘flying termites’, which may be related. ² Baumbach (1969) also has ʃwa, meaning ‘tarry’. The phonetic and semantic overlap suggests that this is at least a borrowed form of the same root, if it is not inherited/dialectical. ³ The Tonga form li-sol^wa must be a borrowing from Copi; otherwise, /dʷ/ would be expected.

. Dos Santos also gives the alternative form s^wela for ‘delay’. Clearly there is substantial synchronic variation within Copi, and at present there is no reason to require this to carry back into PTC or Proto-Copi-Tonga.

A few forms still defy explanation.

Table 27. Aberrant reflexes of PB *c, with sporadic lateral affricates in Tswa-Ronga.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
meet	tshaŋgana	tʃaŋgana	tʃaŋgana	ɬaŋgana	ɬaŋgana	tɬaŋgana 1	tɬaŋgana	-	*càng
five	-	tʃanu	-	n-tɬanu	n-tɬanu	n-tɬanu	n-tɬanu	-	*cáànʊ̀

¹ The phoneme /tɬ/ is described by Ngunga and Faquir (2012) as an alveolar lateral affricate, but is curiously spelled in Persson (1928) as <kl>. Persson’s spelling may point to an original velar lateral affricate which recently shifted to being alveolar. Unfortunately, Persson’s dictionary lacks a robust explanation of this phoneme’s realization.

gives two forms for which at least some of Tswa-Ronga have lateral affricates for PB *c, and Copi has an alveolar affricate. There does not appear to be any obvious conditioning feature here that would make these reflexes different than that of, e.g., *cándú ‘branch’ (see Table 25 earlier). The most likely case is that they are loans. A reviewer pointed out that Lenge li-handi ‘five’ is cognate to Tonga li-βandre, with the Copi term being a loan from Tswa-Ronga. This suggests that the affricates seen in Copi-Tonga are simply a manner of adapting the lateral obstruents from Tswa-Ronga into the native phonology.

3.4. *k

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC
kh	kh	k	ɦ~gɦ	h~gɦ	h	ɦ	*ŋkh	*#N_
tsh	tsh	ts	s	s	s	s	*ns	*N_i
pfh?	?	?	?	?	?	?	?	*N_u
ŋg	ŋg	ŋg	ŋg~ ɦ	ŋg~ h	ŋg~ h	ŋg~ ɦ	*ŋk	*VN_
tsh	tsh	ts	s	s	s	s	*is	*Vi_
kh	kh	k	kh	kh	kh	kh	*kh	*V#i_ in nouns
kh	kh	k	s	s	s	s	*khj	*V#i_ in verbs
s	s	s	s	s	s	s	*s	*_i
s	s	s	ʃ	ʃ	ʃ	ʃ	*s	*ʊ_i, *o_i
f	f	f	f	f	f	f	*f	*_u
ɣj	tʃ	kj~tʃ	ʃ	ʃ	tʃ	ʃ	*kj	*_ɪV,*_eV
f	f	f	f	f	f	f	*f	*_u
g	g	g	k	k	k	k	*k	*mʊ_
ɣ	k	k	k	k	k	k	*k	elsewhere

When it comes to the reflexes of PB *k, the spirantized reflexes shown in

Table 28. Reflexes of PB *k showing spirantization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
fat	ma-fura	ma-fura	ma-fuha	ma-furɦa	furɦa	ma-fura	ma-fuʐa	*ma-fudɦa	*kútà
die	fa	fa	fa	fa	fa	fa	fa	*fa	*kú
rule	fuma	fuma	fuma	fuma	fuma	fuma	fuma	*fuma	*kúm
lung, chest	-	li-fuʋa 1	tʃi-fuva	ʃi-fuβa	ʃi-fuβa	ʃi-fuβa	ʃi-fuba	*ki-fuꞵa	*kúbà
eyelash	ji-tshi	i-tshi	i-tsi	ri-ʃiji	ri-ʃije	li-siji	li-ʃiɦe	*n-sie	*kígè
smoke	wu-tshi	wu-tshi 2	vu-tsi	mu-si	mu-si	mu-si	mu-nsi	*-isi	*jíkì
nerve	mi-siβa	in-siɦa	-	n-siɦa	n-siha	siha	n-siɦa	*mu-sibɦa	*kìpà
bee	ɲosi	-	-	ɲoʃi	ɲoʃi	ɲoʃi	ɲoʃi	*ɲosi	*jókì
blame	sola	sola	sola	sola	sola	sola	sola	*sola	-

¹ The only attestation of the root *kúbà in Copi is this class 11 word meaning ‘tuberculosis’. This is nevertheless paralleled in Tswa-Ronga, e.g., Tsonga ri-fuβa ‘tuberculosis’. ² The aspiration is attested in Bailey (1976).

are the most straightforward. PB *k before *u consistently yields /f/ in all daughter languages, and can directly be reconstructed as *f in PTC. The situation is a bit more complicated before PB *i. In Tswa-Ronga, the reflex is generally /s/, which was sometimes palatalized to /ʃ/. The Copi-Tonga forms point to /s/, with the class 9 reflexes of *kígè ‘eyelash’ yielding /ts^h/. The former can be taken to be the default, and the latter a reflex of the *PB sequence *nki. Unsurprisingly, the reflexes of *i-ki are identical, as shown by *jíkì ‘smoke’. The outcome of *nku, has yet to be found, although the Tonga class 9 word pf^humu ‘chief’ < *N-kúm suggests that it is /pf^h/ within Copi-Tonga.

Before settling on a spirantized PTC reflex of *k, it is worth looking a bit further into palatalization.

Table 29. Non-causative and causative verb pairs with stems ending in PB *k.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
leave	-	suka	-	suka	suka	suka	-	*tsuka	*tíg(ʊk)
remove	-	susa	-	susa	susa	susa	susa	*tsusa	*tíg(ʊk)
roast (itr)	woɣa	joka	oka	oka	-	-	-	*oka	*jòk
roast (tr)	-	-	-	oʃa	woʃa	wotʃa	woʃa	*osa	*jòk
awaken	wuɣa	wuka	vuka	pfuka	pfuka	vuka	pfuka	*ꞵwuka	*bʊ́ʊk
wake	wusa	wusa	-	pfuʃa	pfuʃa	vuʃa	pfuʃa	*ꞵwusa	*bʊ́ʊk

lists roots which have clear causative pairs7; the sequence *ki mostly yields /s/, though in Tswa-Ronga, it occasionally yields /ʃ/. It seems that when preceded by a rounded vowel, /s/ palatalizes further to /ʃ/ in Tswa-Ronga, as is also seen in the reflexes of ‘bee’ above8. The reflexes of *tíg-ʊk-i ‘remove’ are an exception to this, possibly blocked by the spirant in the preceding syllable. Regardless, it seems that *s should be reconstructed as the regular outcome of PB *ki in PTC, with corresponding PTC *ns < PB *nki. As for *ik, this will be represented as *ⁱs for similar phonetic reasons as *ⁱɬ.

For reflexes of *k in other environments, most of the languages have /k/, as shown in

Table 30. Reflexes of PB *k in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
snake	ɲoɣa	ɲoka	-	ɲoka	ɲoka	ɲoka	ɲoka	*ɲoka	*jókà
wife 1	-ɣadzi	-kati	-kati	-kati	-kati	-kati	-kati	*-kadzi	*kádí
arm	li-ɓoɣo	ɗi-woko	poko	βoko	βoko	woko	boko	*boko	*bókò
day	li-tshiɣu	ɗi-tshiku	di-siku	siku	siku	siku	siku	*tshiku	*tíkʊ̀
weave	luɣa	luka	luka	luka	luka	luka	luka	*luka	-

¹ This form is given as a root, as in many of the languages, it is only attested in compounds which refer to various human female entities.

. Tonga is unique in reflecting *k with the voiced velar fricative /ɣ/. If the same series of changes are posited for PB *k as the other voiceless stops, then the chain would be *k > *k^h > *g^ɦ. This would leave an easy explanation for the voicing in Tonga, but would require independent devoicing and deaspiration of the same phoneme in Proto-Tswa-Ronga and Proto-Copi. This seems far more complicated than simply hypothesizing a Tonga internal development of *k > ɣ, and no changes whatsoever in all the other languages. An alternative hypothesis would suggest free variation between a PTC *k~g; since PB *g was typically lost or became *w in PTC, a phonological gap would have existed in the velar series, making voicing noncontrastive if it occurred. Following this, a single phoneme *k will be reconstructed.

The reflexes of *k in initial position in nominal roots show great diversity, but a clear picture can be obtained when these roots are divided up by noun class. The roots in

Table 31. Reflexes of initial PB *k in class 9 roots.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
chicken	khuɣu	khukhu	ŋ-guku	ɦuku	huku	hukwana	ɦuku	*ŋ-khuku	*kʊ́kʊ́
crab	khalala	ɦala, khala	kala	-	hala	-	ɦala	*ŋ-khala	*kádá
fog	huŋguv’a	ɦuŋguva	-	ɦuŋguβa	huŋguβa	huŋguβa	ɦuŋguba	*ŋ-khuŋgu	*kʊ̀ngʊ́
face	khoβe	khoɦe	kohe	n-gɦoɦe	n-gɦohe	n-gɦohe	-	*ŋ-khobɦe	*kópè 1

¹ Given with the meaning ‘eyelash’ in BLR3.

belong to class 9 (prefix *N-) and take plurals in class 10 (prefix *tiN-). The reflexes here are greatly varied, but a narrative can still be constructed. In class 9 and 10 forms, the initial PB sequence *nk became aspirated, just like *mp and *nt. At least in Tswa-Ronga, the resultant sequence *ŋk^h was subsequently voiced to *ŋg^ɦ, and then usually simplified to *ɦ. This was maintained in Tsonga and Ronga, but devoiced in Tswa and Ronga. In Copi-Tonga, it seems most likely that the only development from *ŋk^h was the loss of *ŋ, yielding *k^h and phonologizing an aspiration contrast for velars.

This series of changes in PB *k functions almost as a spectrum, with different languages showing different degrees of adherence. Tonga most consistently reflects simple aspiration, while Copi shows roughly equal amounts of /k^h/ and /ɦ/. Perhaps the former is inherited and the latter is a loan, but the synchronic situation is nevertheless quite mixed. Tswa-Ronga generally has /ɦ/ or /h/, reflecting the full series of sound changes. When it comes to the root *kópè, multiple words reflecting this root appear to exist synchronically in Tsonga: there is n-g^ɦoɦe ‘forehead’, but also ri-koɦe ‘eyebrow’ and ri-ɦoɦe ‘eyelash’. This suggests that as the changes progressed, competing forms which existed simultaneously could be lexicalized and retained in parallel.

The class 11 prefix *dʊ- did not generally trigger any phonological changes, but since its plural counterpart is also class 10, such words were frequently subjected to analogical alteration.

Table 32. Class 11 and 10 reflexes of PB *kʊ́nì ‘firewood’.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
firewood	khuni	li-khuni	kuni	ri-kuɲi	ri-huɲi	-	li-kuɲi	*lu-kuni	*kʊ́nì
	dzi-khuni	ti-khuni	-	ti-ɦuɲi	ti-huɲi	ti-huɲi	ti-ɦuɲi	*dziŋ-khuni

shows both the class 11 and class 10 reflexes of the PB root *kʊ́nì. Tsonga and Ronga have a singular with no sound changes existing alongside a plural with the full shift to /ɦ/. The singular in Changana has been reformed analogically by the plural. In Copi-Tonga, the singular has also been reformed analogically from the plural, carrying over the aspiration derived from the sequence *nk. Within Tonga, there also exists a class 7 reflex of *kʊ́nì, ɣi-ɣuni, which shows the typical unconditioned development of *k into /ɣ/.

In class 3 roots beginning with *k, given in

Table 33. Reflexes of initial PB *k in class 3 roots.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
tail	ŋ-gila	in-kila	n-kila	n-tʃila	n-tʃila	ŋ-kila	ŋ-kila	*mu-kila	*kídà
throat	ŋ-golo	in-kolo	n-kolo	ŋ-kolo	ŋ-kolo	ŋ-kolo	ŋ-kolo	*mu-kolo	*kòdò
banana	ŋ-gombva	in-kombva	n-komva	ŋ-kompfa	kompfa	kova	-	*mu-komva	-
foot	ŋ-gondro	in-kondo	n-kondo	ŋ-kondzo	ŋ-kondzo	ŋ-konzo	ŋ-kondʒo	*ŋ-kondo	-

, the usual unconditioned reflex /k/ is found along with /g/ (with the Tsonga and Changana reflexes of *kídà showing subsequent palatalization). The class 3 prefix in PB was *mʊ-, which often simplified to a homorganic nasal across Tsonga-Copi. This change did not take place early enough for the prefix to merge with that of class 9, and so these roots do not show a merger with the PB sequence *nk. In Tonga, this prefix either fortified the fricative /ɣ/ into /g/, or prevented the lenition of an original Proto-Tonga *g (< PTC *k) into /ɣ/.

Similarly to the situation before *t, class 5 roots beginning with initial PB *k generally have aspirated /k^h/ as an outcome, though the reflexes in

Table 34. Reflexes of PB *k after *i.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
ten	li-khumi	ɗi-gume	kumi	khume	khume	khume	khume	*khume	*kʊ́mè
coal	li-ɣala	ɗi-kala	di-khala	khala	khala	ma-kala	khala	*khala	*kádà
frog	ɲakhele	ɗi-khele,	-	khele	khele	k(h)ele	k(h)ele	*khele	*kédé
froth	li-ɣuv’i	ɗi-khuʋi	kuvi	khuβi	khuβi	khuβi	dubi	*khuꞵi	-
bark	-	ɗi-kamba	-	khamba	kamba	-	kamba	*khamba	-
ash	ŋ-guma	ɗi-khuma	khuma	ŋ-kuma	ŋ-kuma	khuma	-	*-kuma	-
stay	khala	khala	kala	sala	sala	sala	sala	*khjala	*jìkad
satisfy	khura	khura	kuha	ʃurɦa	ʃurɦa	ʃura	ʃuʐa	*khjudɦa	*jíkʊt

are not consistent. Lanham (1955) and Van der Spuy (1989) both contend that this rule seems to have been applied sporadically; to me, it seems like most instances of the rule not applying can be ascribed to analogical influence from their class 6 counterparts. In many languages, this alternation has not been fully leveled; the plurals for ‘ten’ in Tonga and Tsonga are ma-ɣumi and ma-kume, respectively, showing unaffected reflexes of PB *k. Tswa and Ronga have both kele and k^hele as reflexes of *kédé, showing synchronic competition between the inherited and the analogically leveled forms. The single term ɗi-gume in Copi has /g/, but the word ɗi-k^hele for ‘frog’ suggests that this is irregular due to Tonga influence. In the plural, Dos Santos (1949) gives the form ma-kume, which would be the expected inherited form, but Junod (1933) instead gives ma-gume, which would be expected if a paradigm was borrowed from Tonga. The overall picture points to *ik yielding *k^h in PTT, which was largely retained in the daughter languages (remarkably even in Tswa-Ronga, where effects of preceding *i are rarely seen).

In the verbal roots with *ik, the Copi-Tonga reflexes have /k^h/, while Tswa-Ronga has /s~ʃ/. As was the case for PB *ip and *it, these forms seem to reflect aspiration followed by insertion of an off-glide via metathesis. The Copi-Tonga forms lost this glide, as was seen for *it, whereas in Tswa-Ronga, it caused palatalization. The forms with /ʃ/ could be attributed to a secondary palatalization caused by the following *ʊ, though in other such instances, it was a preceding *ʊ, which triggered palatalization.

Certain instances of PB *k before front vowels yield palatalization to /(t)ʃ/ in Tswa-Ronga and Copi. The Tonga reflex of *ké ‘dawn’ in

Table 35. Reflexes of PB *k showing palatalization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
dawn	ɣja	tʃa	kja, tʃa	ʃa	ʃa	tʃa	ʃa	*kja	*ké
deceive	senga	ʃeŋga	-	ʃeŋga	ʃeŋga	ʃeŋga	-	*keŋga	*kéng
buy	-	ʃaʋa	ʃaʋa	ʃaβa	ʃaβa	ʃaβa	ʃaba	*kjaꞵa	-
dance	-	siɲa	kina	tʃina	tʃina	kina	kina	*kina	*kín 1

¹ Also present in BLR3 is the root *bín with the same meaning, which would theoretically yield the same Copi result (aside from the palatalization of *n, which would not be expected from either root).

shows no evidence of palatalization, meaning that this palatalization is a post-PTC development. This fact is supported by the occasional lack of palatalization in Tswa and Ronga, and the typological regularity of /k/ palatalizing before front vowels (Kümmel, 2007, p. 252). The outcomes of the PB class 7 prefix *kɪ- are Tonga ɣi-, Copi tʃi-, and Tswa-Ronga ʃi-, identical to the outcome of *ké. This suggests that these are the regular outcomes of PB *k followed by a non-high front vowel. Perhaps roots like *kín ‘dance’, which show only sporadic palatalization, should be taken as loans, especially given the different forms in Copi and Lenge.

Interestingly, those roots which have PB *nk root-internally show a different outcome compared to initial *nk, as seen in

Table 36. Reflexes of root-internal PB *nk.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
give	niŋga	niŋga	-	ɲika 1	ɲika	ɲika	ɲika	*ɲiŋka	*ɲínk
stink	nuŋga	nuŋga	nuŋga	nuɦa	nuŋɦwa 2	nuha	nuɦa	*nuŋka	*nùnk
answer	-	aŋgula	-	aŋgula	yaŋgula	aŋgula	yaŋgula	*aŋkula	*jánkʊd

¹ The form niŋga is also attested for Tsonga, but the lack of palatalization of /n/ suggests it may be a borrowing. ² The same form is also attested in Tsonga, and a similar form nuŋ^ɦa is found in Ronga. These seem to be re-verbalizations of the original deverbal noun *nùnk-o. Regardless, they seem to show one more potential outcome of the sequence *nk.

. In Copi-Tonga, the outcome of such a sequence is /ŋg/. Following this, it appears that the outcome of PB *nk in Copi-Tonga was simply conditioned by the position of the sequence in the word. As for Tswa-Ronga, the three different attestations show three different outcomes: /k/, /ɦ ~ h/, and /ŋg/. It is possible that these were conditioned by the vowels which occurred around them, with the former two reflexes being associated with the PB close vowels. It is also possible that these forms simply reflect various stages of lexicalization of the reflex of PTC *ŋk^h as the sound underwent further development in Tswa-Ronga.

3.5. *b

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC
mb	mb	mb	mb	mb	mb	mb	*mb	*N
ndz	ndz	nz	mpf	mpf	?	mpf	*mbz	*N_i
ndz	ndz	nz	ɳʈʂhw	mpʂ	ʐ	mpʂ	*mbzj	*N_iV
ɓ	p	p	β	β	β	b	*b	*V(#)i_
s	s	s	pf	pf	v	pf	*bz	*_i
pf	pf	pf	pf	pf	v	pf	*bv	*_u
j?	j~ɓ?	j?	bj	bʐ	bj	bj	*βj	*_ɪV
bw	gw~bjw	gw~bjw	bj	bʐ	bj	bj	*βw	*_ʊV
Ø	Ø	Ø	Ø	Ø	Ø	Ø	*Ø	*mʊ_
w	w	v	β	β	w	b 1	*β	*_ʊ, *_o
v’	ʋ	v	β	β	β	b	*β	elsewhere
1 Transcriptions from Baumbach (1969) suggest that Ronga /b/ may be realized as [v~β] in this environment, but there are not enough forms cited to make this claim definitively.

Among the reflexes of PB *b before non-close vowels, two series can clearly be spotted. The first is found in reflexes with PB *b followed by the vowels *a, *e, and *ɪ (shown by the first six examples in

Table 37. Reflexes of PB *b in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
ear	n-dzev’e	n-dzeʋe	n-zeve	n-dɮeβe	n-dɮeβe	n-dɮeβe	n-dɮebe	*n-ɮeꞵe	*jèbé
boil	v’ila	-	vila	βila	βila	βila	bila	*ꞵila	*bìd
carve	v’ata	ʋata	vata	βatɬa	βatɬa	βatɬa	batɬa	*ꞵazja	*bàij
peel	v’andza	ʋandza	-	ꞵandɮa	ꞵandɮa	ꞵandɮa	bandɮa	*ꞵanɮa	-
hate	v’eŋga	ʋeŋga	venga	ꞵeŋga	ꞵeŋga	ꞵeŋga	beŋga	*ꞵeŋga	-
pay tribute	luv’a	-	luvela	luꞵa	luꞵa	luꞵa	luba	*luꞵa	-
skin	li-ɗowo	ɗi-ɗowo	n-dowo	ɖʐoβo	dzoβo	dzowo	dʒobo	*li-doꞵo	-
mold	wumba	wumba	vumba	βumba	βumba	wumba	bumba	*ꞵumba	*bʊ́mb
hippo	m-bvuwu	m-bvuwo	-	m-pfuβu	m-pfuβu	βuwu	m-pfubu	*m-vuꞵu	*gùbʊ́
pay dowry	lowola	lowola	-	loꞵola	loꞵola	lowola	lobola	*loꞵola	-

), and the second is found in reflexes of *b followed by the rounded vowels *o and *ʊ (shown by the last four examples in Table 37). Tsonga, Changana, and Ronga show the same reflex of *b in these environments, but Tonga, Copi, and Tswa have labial glides instead of their normal reflexes. The question then is whether this distinction existed at the level of PTC, and whether it was phonemic.

A phoneme *w often arises from the PB vowel *ʊ when followed by a vowel (such as after *g in Section 3.8). In Copi, Tonga, and Tswa, the [w] which results from PB *b has merged with this *w into a single phoneme /w/, as seen in the final three entries in Table 37 above. The clear distinction of the reflexes of these phonemes in Lenge, Tsonga, Changana, and Ronga indicates that such a merger cannot have happened at the level of PTC. Therefore, if there was a separate reflex of *b before rounded vowels at the PTC level, it must have been phonetically distinct from *w. Such a sound would have a complimentary distribution with other reflexes of PB *b, and they therefore need only represent a single phoneme. While it is not possible to conclusively determine if this sound had allophony in PTC, the fact that /β/ and /v/ > /w/ around rounded vowels is a well-attested sound change (Kümmel, 2007) means that it could have easily happened independently (or as an areal change) in Copi, Tonga, and Tswa.

Before assigning a representation for the regular PTC reflex of PB *b, it is worth having a look at the forms in

Table 38. Reflexes of PB *b after *i.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
arm	li-ɓoɣo	ɗi-woko	poko	βoko	βoko	woko	boko	*boko	*bókò
liver	li-ɓindri	tʃi-βindi	tʃi-vindi	ʃi-βiɳɖʐi	ʃi-βindzi	ʃi-βinzi	ʃi-bindʒi	*-ꞵindi	*bíndɪ
breast	li-ɓele	-	-	βele	βele	βele	bele	*bele	*béèdè
color	li-v’ala	im-bala	-	βala	βala	mu-βala	bala	*-ꞵala	*bádà
steal	ɓa	pa	pa	jiβa	jiβa	jiβa	jiba	*ba	*jíb

where PB *b follows *i. Tonga has /ɓ/ for PB *b, regardless of the following vowel (additional evidence of a lack of allophony in PTC), and Copi and Lenge have occasional /p/. As with other cases of class 5 influence, there are words which do not seem to show any change, such as the reflex of PB *bádà ‘color’. However, as discussed in other sections (see Section 3.2 and Section 3.4), this could simply be due to analogical replacement from a corresponding class 6 term. I will represent the outcome of PB *ib as PTC *b and PB *b as PTC *β based on these sounds being the closest to the modern reflexes, though *ⁱb and *b or *B and *b could be used just as well.

Words in

Table 39. Reflexes showing elision of PB *b.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
forehead	ɣi-wombo	tʃi-mombo	m-ombo	mombo	mombo	ʃi-mombo	mombo	*m-ombo	*bòmbó
potter	mu-wumbi	mu-wumbi	-	wumbi	-	mu-wumbi	-	*mu-umbi	-
body	mili 1	immiɗi	midi	miri	miri	miri	mi-ʐi	*mili	*bìdì
milk	- 2	ma-si	ma-si	ma-si	ma-si	ma-si	-	*ma-ɬi	*bícì

¹ This form exists in Tonga alongside a class 7 variant ɣi-v’ili, which shows the expected development of PB *b. ² The Tonga term ma-ɦ^we ‘milk’ seems derived from the related PB root *bícù with the addition of a final vowel /e/.

seem to show either deletion of PB *b or fusion with a class prefix. In the reflexes of the class 3 words *bòmbó and *bìdì, it seems that the initial sequence *mʊ-b- yielded /m/ in the daughter languages. The Tonga reflex of *bòmbó is class 7, and therefore shows the expected reflex of *b before *o. The Copi and Tswa class 7 reflexes of *bòmbó still have /m/, but this could potentially reflect either a later formation from an already affected class 3 term, or analogical replacement from an unattested class 3 term. The class 6 reflexes of *bícì suggest derivation from a bare root *cì instead, and it is unclear how the initial *bɪ may have been lost. Guthrie (1971) cites no class 3 examples of this root in Bantu; perhaps the class 6 prefix *ma- occasionally showed similar influence, with other forms lost to analogy, or perhaps there was conflation with the class 11 prefix *lʊ-.

When followed by glides, Tswa-Ronga shows palatalization regardless of the nature of the glide. In Copi-Tonga, there at least seems to have been a distinction maintained between labialization and palatalization, though its ultimate outcome in the daughter languages is unclear. The Copi reflex of *bʊ̀è in

Table 40. Reflexes of PB *b before glides.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
dog	jim-bwa	im-bwa	i-mbjwa	m-bjana	m-bʐana	m-bjana	jim-bja	*jim-ꞵwa	*bʊ́à
stone	-	ɗi-rigwe	di-rigwe	ri-bje	ri-bʐe	ri-bje	ʐi-bje	*li-ꞵwe	*bʊ̀è
plant	jala	jala	jala	bjala	bʐala	bjala	bjala	*ꞵjala	*bíad
belch	-	ɓisa	-	bjisa	bʐisa	bisa	bɦisa 1	*bja	*bì

¹ Baumbach (1969) suggests that this is phonetically [β^ɦisa], but the spelling in Quintão (1951) is ambiguous.

suggests velarization of the stop, but there is no Tonga cognate to say whether this goes back to PTC. It seems that the regular outcome of a sequence *bj may have been /j/ in Copi-Tonga, though a following front vowel seems to have caused the glide to be lost before it could influence the *b (a development apparently also shared with Tswa and Ronga). The lack of sufficient examples of these sequences makes it impossible to make definitive claims about outcomes in the daughter languages, but what has been identified seems to reflect sequences with two distinct phonemes at the level of PTC, *b followed by either *j or *w, with subsequent developments happening after the PTC period.

As demonstrated in

Table 41. Reflexes of PB *b showing spirantization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
close	-	pfalela	-	pfala	pfala	vala	pfala	*bvala	*bùgad
open	-	pfula	pfula	pfula	pfula	vula	pfula	*bvula	*bùgʊd
harvest	-	pfuna	pfuna	pfuna	pfuna	pfuna	-	*bvuna	*bún
white hair	dzim-bvi	-	-	tim-pfi 1	tim-pfi	wu-vu	-	*dzim-bvi	*búì
swell	simba	simba	simba	pfimba	pfimba	vimba	pfimba	*bzimba	*bímb

¹ Exists alongside tim-pfu.

, the spirantized outcome of PB *b in most of Tswa-Ronga is a voiceless labiodental affricate regardless of whether the following glide is *i or *u, with Tswa having a fricative instead. Copi-Tonga is less clear, with the one reflex of *bi having /s/, and the reflexes of *bu having either /v/ or /pf/. If one assumes that the sequences here show parallel development to the spirantized outcomes of PB *p, it would point to *bz < *bi and *v < *bu. However, the Copi reflexes pfuna < *bún and ʋuna < *gún suggest an affricate outcome *bv < *bu and *v < *gu instead9. PTC *bz was then merged with *bv in Tswa-Ronga, but in Copi-Tonga, it was devoiced and deaffricated to /s/. PTC *bv was devoiced in Copi as well as Tsonga, Changana, and Ronga, but in Tswa, it was only deaffricated. It is possible that the normal outcome of PB *bu in Tonga is /pf/, with the voicing in dzim-bvi being conditioned by the preceding nasal, but a lack of clear reflexes precludes anything more than speculation10.

Both root-internally and in class 9 terms with initial PB *b, the daughter languages have /mb/ in most environments; examples are given in

Table 42. Prenasalized reflexes of PB *b.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
bathe	hamba	samba	samba	ɬamba	ɬamba	ɬamba	ɬamba	*ɬamba	*càmb
seed	m-bewu	m-bewu	m-beu	m-bewu	m-bewu	m-bewu	m-bewu	*mu-ꞵewu	*bégʊ́
raw	-mbihu	-mbiso	-mbisi	-mbisi	-mbisi	-mbisi	-mbisi	*-mbiɬi	*bícì
beehive	-	mw-amba	i- ŋwambu	ŋw-ambu	-	ŋw-amba	ŋw-amba	*mw-ambu	-

. For PTC, this can simply be reconstructed as *mb.

The sequence *mb also shows unique outcomes when subject to spirantization, though they are not consistent. The relevant roots in

Table 43. Prenasalized reflexes of PB *b showing spirantization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
wash	handza	-	-	ɬaɳʈʂhwa	ɬampʂa	ɬaʐa	ɬampʂa	*ɬambzja	*càmb
lick	dandza	dandza	danza	laɳʈʂhwa	nampʂa	laʐa	nampʂa	*lambzja	*dámb
fish	n-dzandzi	tshandzi	tsanzi	n-ɬampfi	n-ɬampfi	-	n-ɬampfi	*n-ɬambzi	*càmb

are *càmb and *dámb with a following causative suffix *i, and *càmb again followed by an agentive suffix *i. Copi-Tonga consistently shows /n(d)z/, but some words in Tswa-Ronga show labiodental affrication and others labioalveolar affrication. The most likely explanation for this is the presence of final *a in the verbal forms, which caused additional glide formation.

3.6. *d

Before most PB vowels (*a, *e, *o, *ʊ), PB *d is reflected in all languages as /l/, shown in

Table 44. Reflexes of PB *d in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
heap	li-tala	ɗi-tala	tala	tala	tala	tala	taleni 1	*zala	*jàdà
breast	li-ɓele	-	-	βele	βele	βele	bele	*bele	*béèdè
dream	lora	lora	loha	lorɦa	lorɦa	lora	loʐa	*lodɦa	*dóot
flower	ɣi-luv’a	tʃi-luʋa	tʃi-luʋa	ʃi-luβa	ʃi-luβa	-	ʃi-luba	*ki-luꞵa	*dʊ̀bà
weave	luɣa	luka	luka	luka	luka	luka	luka	*luka	-

¹ Attested as ‘in the pile’.

below.

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC
ndr	nd	nd	ɳɖʐ	ndz	nz	ndʒ	*nd	*N_
?	ndz	?	mpf	mpf	v	mpf	*ndzi	*N_i
ɗ	ɗ	d	ɖʐ	dz	z	dʒ	*d	*V(#)i_
dz	t	ts	t	t	t	t	*dz	*_i
dz	t	ts	pf	pf	v	pf	*dz	*_u
l	ɗ	d	r	r	r	dʒ	*l	*V_ɪ
l	ɗ	d	r	r	r	ʐ	*l	*V_ɪ
dz?	ɟ	g(j)	dj	dʒ	g	d	*dj	*_ɪV
dw	lw	dw~lw	lw	lw	lw	lw	*lw	*_ʊV
l	l	l	l	l	l	l	*l	elsewhere

Table 45. Reflexes of PB *d before *ɪ.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
two	-v’ili	-mbiɗi	-mbidi	-	-	-mbiri	-mbiʐi	*-mbili	*bìdí
mortar	-	tʃi-thuɗi	-	tʃhuri	tʃhuri	tʃuri	tʃuʐi	*tshuli	*tʊ́dí
body	mili	immiɗi	midi	miri	miri	miri	miʐi	*mili	*bìdì
plant	lima	ɗima	dima	rima	rima	rima	dʒima	*lima	*dìm
pay fine	liβa	ɗiɦa	diha	riɦa	riha	riha	dʒiɦa	*libɦa	*dìp
cry	lila	ɗila	dila	rila	rila	rila	dʒila	*lila	*dìd

shows that before *ɪ, the situation is quite different. Tonga still yields /l/, but Copi instead yields /ɗ/; Tsonga, Changana, and Tswa yield /r/; and Ronga yields /ʐ/ root-internally and /dʒ/ root-initially. The Tswa-Ronga reflexes are quite similar to those reflexes of PB *t (perhaps not surprising if PB *d is the voiced counterpart of *t), but critically, the Tsonga and Changana reflexes do not have aspiration. The most that can be projected back into PTC is allophony, which was then eliminated or phonologized independently in the daughter languages, either through merger with other phonemes or introduction of loanwords with the conditioned allophones outside of predictable environments. The singular encompassing phoneme at the level of PTC will be reconstructed as *l, which would have been realized as /d/ in certain environments, such as before front high vowels.

In addition to the reflexes above, a series of correspondences for the sequence *di is quite robustly attested and given in

Table 46. Reflexes of PB *d showing spirantization before *i.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
star	ɲeledzi	ɲeleti	ɲeleti	ɲeleti	ɲeleti	ɲeleti	ɲeleti	*ɲeledzi	*jédèdí
take	dzeɣa	teka	tseka	teka	teka	teka	teka	*dzeka	*diek
extinguish	-	tima	tsima	tima	tima	tima	tima	*dzima	*dím
lake	li-dziv’a	tʃi-tiʋa	-	tiβa	tiβa	tiβa	tiba	*dziꞵa	*dìbà
wife	-ɣadzi	-kati	-kati	-kati	-kati	-kati	-kati	*-kadzi	*kádí
finger	-	li-tiɦo	li-tsiho	ri-tiɦo	ri-tiɦo	li-tiho	li-tiɦo	*li-dzibɦo	-
sharpen	lodzwa 1	lota	lotsa	lota	lota	lota	lota	*lodza	-

¹ Meaning ‘sharp’, which appears to be a passivization of an unrecorded form *lodza.

. While it is reflected as /ti/ in most the languages under consideration, it is reflected as /dzi/ in Tonga and /tsi/ in most Lenge forms. Given that *d is voiced, it makes sense to reconstruct a voiced affricate *dz here. This sound would have been the voiced counterpart of the affricate yielded by PB *ti. Tonga and Lenge then show a retention, whereas in the other languages, it has been deaffricated.

When PB *d is followed by *u, as in

Table 47. Reflexes of PB *d showing spirantization before *u.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
beard 1	ɣi-ledzu	tʃi-leto	tʃi-letsu	ma-lepfu	ma-lepfu	xi-levu	ma-lepfu	*-ledzu	*dèdù
drip	- 2	tuta	tsuta	pfuta	pfuta	vuta	pfuta	*dzudza	*dùɪdi
turtle	-	futu	-	m-fuʈʂu	m-futsu	futsu	m-fuʈʂu	*m-fudzu	*kúdù

¹ Also accepted here was the word ‘chin’. ² Tonga has the form pfuta, meaning ‘to ooze pus’. Despite the seemingly related meaning, it was taken as a loanword since it does not show *di > /dzi/. There is also the form dzudza ‘agitate, mix’ which shows the expected sound correspondences, though I am uncertain if the shift from ‘drip’ > ‘mix’ is tenable.

, the outcome of spirantization is less clear. Tswa-Ronga has the same reflexes as spirantized *b, *j, and *g, except in the reflexes of *kúdù, where the languages instead have voiceless alveolar or retroflex affricates. This is perhaps due to influence from a preceding *u, but more examples would be necessary to confirm this. Copi and Tonga yield /t/ and /ts/, respectively, in parallel with the sequence *di. Tonga seems to mirror this, but only one clear cognate could be found. In any case, it is clear that the outcomes of *du are not the same as any of the other voiced stops (or other phonemes), so the phoneme must have been distinct from those. Synchronically in PTC, this ancestral phoneme can be analyzed as an allophone of *dz before a close back vowel *u.

Class 5 roots with initial PB *d, listed in

Table 48. Reflexes of PB *d after *i.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
tongue	li-ɗimi	li-ɗimi	li-dimi	ri-rimi	ri-rimi	li-rimi	li-dʒimi	*li-dimi	*dímì
sun	li-ɗambo	ɗi-tambo	dambu	djambu	dʒambu	gambo	dambo	*li-djambo	-
skin	li-ɗowo	ɗi-ɗowo	n-dowo	ɖʐoβo	dzoβo	dzowo	dʒobo	*li-doꞵo	-

, show clear mutation in Copi-Tonga. In Tonga, this yields an implosive /ɗ/ just as *ib yielded /ɓ/. In parallel, I reconstruct PTC *d in this environment. In Tswa-Ronga, the reflexes are what would be expected for initial *nd, though there is no other reason to suspect a nasal was present. It is therefore assumed that this outcome is due to mutation from the class 5 prefix.

Before labial glides, the outcome of PB *d is /l/, except in Tonga and one word in Lenge, where it is /ɗ/ and /d/, respectively. Before a palatal glide, it seems that *d was palatalized, though the exact outcome of this differs in the daughter languages, as shown in

Table 49. Reflexes of PB *d before glides.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
fight	dwana	lwa	lwa	lwa	lwa	lwa	lwa	*lwa	*dʊ̀
beer	wadwa	-	vw-adwa	bjalwa	bzala	bjala	bjala	*ꞵw-alwa	*jàdʊ́á
eat	hodza	ɟa	gja	dja	dʒa	ga	da	*dja	*dí
sin	- 1	ɟoɦa	goha	djoɦa	dʒoha	goha	doɦa	*djobɦa	-
learn	-	ɟonda	gonda	djoɳɖʐa	dʒondza	gonza	dondʒa	*djonda	-

¹ The term goha is attested, but the rare /g/ and aberrant /h/ suggest that it is a borrowing from Copi.

. As with other phonemes followed by glides, this need not reflect any phonemic change in PTC, so these can be projected back to the sequences *lw and *dj.

The PB sequence *nd, either as a component of the root or as a result of root-initial *d with a nasal prefix, seems to show some sort of affrication or palatalization in the various daughter languages except Copi and Lenge. These changes seem to be unconditioned, so the reflexes in

Table 50. Reflexes of prenasalized PB *d.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
liver	li-ɓindri	tʃi-βindi	tʃi-vindi	ʃi-βiɳɖʐi	ʃi-βindzi	ʃi-βinzi	ʃi-bindʒi	*-ꞵindi	*bíndɪ
follow	landra	landa	landa	laɳɖʐa	landza	lanza	landʒa	*landa	*dànd
travel	endra	enda	enda	eɳɖʐa	endza	enza	jendʒa	*enda	*gènd
fire	-	n-dilo	n-dilo	ɳ-ɖʐilo	n-dzilo	n-zilo	n-dʒilo	*n-dilo	*dìdò
inside	n-drani	n-dani	n-dani	ɳ-ɖʐeni	n-dzeni	n-zeni	n-dʒen	*n-dani	*dà

reflect a continuation of a single PTC sequence *nd.

Spirantized reflexes of the sequence *nd are few, but possibly shown by the root *cándú in

Table 51. The reflex of PB * cándú ‘branch’, showing spirantization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
branch	-	tshandzu	-	ɬampfu	ɬampfu	ɬavu	ɬampfu	*iɬandzu	*cándú

. Interestingly, the outcome seems to be identical to that of *mb with the agentive suffix *i as discussed in Section 3.5.

3.7. *j

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC
ndz	ndz	ndz	ndɮ	ndɮ	ndɮ	ndɮ	*nɮ	*N_
ɲ	ɲ	ɲ	ɲ	ɲ	ɲ	ɲ	*ɲ	*#N_V[+H], N_VN
t	t	t	t	t	t	t	*z	*i_, *_i
t	t	t	tɬ	tɬ	tɬ	tɬ	*zj	*CVi_
pw	pf	pf	tw	tw	zw	tw	*zw	*#_iu
ɓ	p~b	p	tw	tw	zw	tw	*zw	*mʊ_iu
j	j	j	j	j	j	j	*j	*V_a
Ø	Ø	Ø	Ø	Ø	Ø	Ø	*Ø	elsewhere

The precise nature of Proto-Bantu *j has been the subject of much commentary and disagreement, and recent studies have called into question whether it was ever truly an extant phoneme at all. Wills’ (2022) recent article on PB *j examines a variety of environments in which *j has traditionally been reconstructed, and concludes that in the vast majority of them, no sound need be reconstructed at all. For the sake of visualization and reference to existing materials, *j will be discussed as a PB phoneme, but Wills’ well-founded criticisms will be discussed where relevant.

To begin, let us examine *j in the medial position, where Wills suggests that no sound was present in PB, and that the glides which arise in many daughter languages were later developments to resolve hiatus. This indeed matches with what is seen for *j Tsonga-Copi. The question is then whether or not to reconstruct a glide at the level of Proto-Tsonga-Copi. Based on the examples in

Table 52. Internal reflexes of PB *j in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
trap	rija	rija	rija	rɦeya	rɦiya	-	ʐeja	*dɦeja	*téj
jaw	-	li-saja	-	ri-ɬaja	ri-ɬaja	ɬaja	li-ɬaja	*lu-ɬaja	*càjá
return	wuja	wuja	vuja	βuja	βuja	wuja	buja	*ꞵuja	*bʊ́j

, it seems that at least before a following *a, glides had been well established within PTC and should be reconstructed. This constitutes a merger with PTC *j, the reflex of PB *ɪ before vowels.

Where *j has been reconstructed in initial position in verbs, such as the forms in

Table 53. Reflexes of initial PB *j in verbs, showing loss.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
build	-	aka	aka	aka	aka	aka	jaka	*aka	*ják
sing	embelela	embelela	embelela	jimbelela	jimbelela	jimbelela	jimbelela	*imbelela	*jímb
deceive	-	woŋga	onga	-	(w)oŋga	-	woŋga	*oŋga	*jóng
grow thin	wondra	wonda	onda	ondza	(w)ondza	onza	wondʒa	*onda	*jónd

, Wills suggests that a vowel-initial reconstruction is virtually always sufficient. This is appropriate for Tsonga-Copi, where glides only occur sporadically in such stems and can potentially be a feature of speakers’ idiolects.

In nominal roots where initial PB *j has been reconstructed, there is also little evidence to support any consonant in PTC or PB. The majority of these words, given in

Table 54. Reflexes of initial PB *j in class 1 and 3, and 14 roots.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
year	mw-aɣa	mw-akani	-	naŋwaka	naŋwaka	ŋw-akani	ŋw-aka	*mw-aka	*jákà
child	ɣj-anana	mw-anana	mw-anana	ŋw-ana	ŋw-ana	ŋw-ana	ŋw-ana	*mw-ana	*jánà
soul	-	moja	moya	moja	moja	moja	moja	*mw-oja	*jòjà
salt	mu-ɲu	mu-ɲu	mu-ɲu	mu-ɲu	mu-ɲu	mu-ɲu	mu-ɲu	*mu-ɲu	*jínyʊ̀
fur	li-woja	ɗi-toja	ma-woja	βo-ja	βo-ja	wo-ja	bo-ja	*-oja	*jòjá
brain	wo-ŋgo	wo-ŋgo	woŋga	bj-oŋgo	βo-ŋgo	wo-ŋgo	bo-ŋgwe	*βw-ongo	*jòngó
beer	w-adwa	-	vw-adwa	bj-alwa	bʐ-ala	bj-ala	bj-ala	*βw-alwa	*jàdʊ́á

, show allomorphs of a class prefix depending on the quality of the vowel which follows. Note the differing behavior of each *j in *jòjà ‘soul’ and *jòjá ‘fur’; were there to be no consonant in either position in PTC, one would be left with forms with three consecutive vowels. One might expect such a strange phonetic environment to be resolved in numerous ways between the daughter languages, yet all show a glide /j/ between the final two vowels. To me this lends credence to the idea that this glide was already present in PTC.

In class 5 roots such as those in

Table 55. Reflexes of initial PB *j in class 5 roots.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
heap	li-tala	ɗi-tala	tala	tala	tala	tala	taleni	*zala	*jàdà
pigeon	li-tuv’a	tʃi-tua	tʃi-duvane	tuβa	tuβa	tuβa	tuba	*zuꞵa	*jʊ́bà
country	li-tiɣo	ɗi-tiko	tiko, di-jiko	tiko	tiko	tiko	tiko	*ziko	*jíkò
eye	li-so	ɗi-so	di-so	ti-ɬo	ti-ɬo	ti-ɬo	ti-ɬo	*zi-ɬo	*jícò
tooth	-	ɗi-no	di-no	ti-ɲo	ti-ɲo	ti-no	ti-ɲo	*zi-no	*jínò
egg	li-tandra	ɗ-anda	d-anda	tandza	tandza	tanza	tandʒa	*zanda	-

, the typical outcome of PB *j is /t/. While unique outcomes for class 5 forms are quite common in Copi-Tonga, they are notably rare in Tswa-Ronga except here. It therefore seems that this shows a separate phonological process. Wills (2025, pp. 16–20) proposes that these forms reflect the combination of the PB class prefix *i with a prothetic glide in a vowel-initial root. He denotes the resultant phoneme in Proto-Southern-Bantu as *Z, and the terms containing it “Z-reflexes”. The outcomes of PSB *Z are similar to that of PB *di, but notably distinct in Tonga, where the former yields *t and the latter *dz.

The reflexes of *jícò and *jínò (along with the cognate set meaning ‘egg’) show slightly different behavior. In Tswa-Ronga, an initial ti- is treated as a prefix, being replaced by the class 6 prefix ma- or variant me- in the plural. In Copi-Tonga, no form with /t/ is found, and the standard class 5/6 prefixes are used on the root. The Tswa-Ronga forms reflect two possible situations; first, initial /t/ may derive from PSB *Z, and the first syllable was treated as a prefix, likely due to analogy with the plural. Alternatively, the /t/ may have resulted as a result of the augment, with PB *dɪ-i-ino coalescing into *di-ino, with subsequent spirantization of *d into /t/ and further vowel contraction. In Copi-Tonga, the outcomes generally do not show any changes, with bare roots -so, -no, and -and(r)a being prefixed with the regular class 5 prefix. The only exception is the Tonga word for ‘egg’, li-tandra, which must be derived from a PSB form with *Z. In general, it seems the most economic option is to posit Z-reflexes at the level of PTC, which underwent significant changes in the daughter languages due to analogy.

This is not the only environment where /t/ appears as a reflex of PB *j; it also appears as the regular outcome of the sequence often reconstructed as *jij. Wills (2022) suggests that these be reconstructed as *iy, with many languages turning this sequence into a strong form such as /iz/ or simply /z/. The causative reflexes of *cʊ̀ʊj ‘strain’ in

Table 56. Additional reflexes of PB *j yielding /t/.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
water	ma-ti	ma-ti	ma-ti	ma-ti	ma-ti	ma-ti	ma-ti	*ma-zi	*jíjì
be full	tala	tala	tala	tala	tala	tala	tala	*zala	*jíjad
strain	huta	-	suta	ɬuta	ɬuta	ɬuta	ɬuta	*ɬuza	*cʊ̀ʊj

might reflect a different story, where the sequence *ji (or perhaps *yi following Wills’ notation) yields /t/. It is possible that *jij should instead be reconstructed as *ji or *yi, but it is also possible that both forms existed and yielded /t/ in Tsonga-Copi.

The identical outcomes of *ij, *ji, and the Z-reflexes suggests a merger in PTC. The universal outcome of /t/ suggests an origin in something phonetically similar, though it would have to be distinct from the *t which yielded /ts/ in Tswa-Ronga (see Section 3.2). If a development parallel to PB *c before front vowels is supposed, then PB *j in this environment would yield *z in PTC, which may continue PSB *Z directly. This would rectify the relative chronology, but the phonetics of the change *z > *t require further cross-linguistic analysis. For the time, *z will be accepted as a reflex of PB *ji, and *ij.

One of the few environments where Wills consistently reconstructs a consonant is in the initial position of class 9 roots. He reconstructs *ny for *nj in ‘weak’ reflexes, and *nj/*nz in ‘strong’ reflexes, with the type of reflex present varying between and within languages. In Tsonga-Copi, two sets of outcomes of *nj are indeed found and given in

Table 57. Reflexes of initial PB *j in class 9 roots.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
star	ɲeledzi	ɲeleti	ɲeleti	ɲeleti	ɲeleti	ɲeleti	ɲeleti	*ɲeledzi	*jédèdí
hip	ɲoŋga	-	-	riɲoŋga	ɲoŋga	ɲoŋga	ɲoŋga	*ɲoŋga	*jóngà
buffalo	ɲari	ɲari	ɲahi	ɲarɦi	ɲarɦi	ɲari	ɲaʐi	*ɲadɦi	*játí
snake	ɲoɣa	ɲoka	-	ɲoka	ɲoka	ɲoka	ɲoka	*ɲoka	*jókà
bee	ɲosi	-	-	ɲoʃi	ɲoʃi	ɲoʃi	ɲoʃi	*ɲosi	*jókì
path	n-dzila	n-dzila	n-zila	n-dɮela	n-dɮela	n-dɮela	n-dɮela	*n-ɮela	*jìdà
famine	n-dzala	n-dzala	n-zala	n-dɮala	n-dɮala	n-dɮala	n-dɮala	*n-ɮala	*jàdà
ear	n-dzev’e	n-dzeʋe	n-zeve	n-dɮeβe	n-dɮeβe	n-dɮeβe	n-dɮebe	*n-ɮeꞵe	*jèbé
elephant	n-dzofu	n-dzofu	n-zofu	n-dɮopfu	n-dɮopfu	n-dɮovu	n-dɮopfu	*n-ɮovu	*jògù

; the first of these consistently shows /ɲ/ as a reflex of the sequence *nj, indicating a merger with PB *ɲ at the level of PTC. The other set of cognates shows /ndz/ in Copi-Tonga and /ndɮ/ in Tswa-Ronga. In figuring out the phonemic state of things, it is again worth supposing a parallel between *j and its voiceless counterpart *c. If the reflex of *c was PTC *ɬ, then it would not be strange to posit *ɮ as a parallel reflex of *j in early PTC. This original *ɮ was perhaps preserved as a post-nasal allophone of other phonetic reflexes of PB *j. As these other reflexes were lost or changed, this variation became phonologized in synchronic PTC.

As for what conditions the outcome of *nj, a comparison of the outcomes of roots such as *jédèdí vs. *jèbé, *jóngà vs. jògù, and *játí vs. *jàdà makes it clear that it is not the quality of the following vowel which is responsible here. Instead, it seems to be the tone of the following vowel which conditions the outcome, with *ɲ occurring before high-tone vowels and *nɮ before low-tone vowels. This distribution did not apply to *nj when it occurred word-internally, as in the examples in

Table 58. Reflexes of root-internal prenasalized PB *j.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
hand	ɣj-andza	ɗ-andza	d-anza	ʃ-andɮa	ʃi-jandɮa	tʃ-andʒa	m-andɮa	*-anɮa	*gànjà
cricket	ɲendze	ɲendza	-	ɲendɮe	ʃi-jendɮwa	-	ʃi-jendɮwa	*-enɮe	*jénjé
outside	βaβandze	ɦaɦandze	hanze	ɦandɮe	handɮe	lahandɮe	ɦandɮe	*bɦa-nɮe	*njáì
peel	v’andza	ʋandza	-	βandɮa	βandɮa	βandɮa	bandɮa	*βanɮa	-

. In this environment *nj is consistently reflected as *nɮ, though the Tswa data is unclear. The development of *nj is particularly remarkable, and is discussed further in Section 4.2.

Several forms are reconstructed with the sequence *ji, which formed a phonological complex with the following consonant, and had a unique outcome in the daughter language. Wills treats these as reflexes of verbs with initial *i, possibly a relic of an earlier reflexive marker. This analysis appears to be correct for Tsonga-Copi, and therefore these forms will be treated in the sections for the relevant phoneme following *i. They are presented together in

Table 59. Reflexes of PB *ji preceding other consonants.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
strike	ɓajisa?		-	ꞵa	ba	ba	ba	*phja?	*jìp?
ask	bula	bula	-	ꞵula	bula	bula	bula	*phjula?	*jípʊd
cook	-	bika	bika	ʂeka	ʂeka	bika	pʂeka	*phjeka	*jìpɪk
pour	thela	thela	tela	tʃhela	tʃhela	tʃela	tʃela	*thjela	*jìtɪd
stay	khala	khala	kala	sala	sala	sala	sala	*khjala	*jìkad
satisfy	khura	khura	kuha	ʃurɦa	ʃurɦa	ʃura	ʃuʐa	*-khjudɦa	*jíkʊt
steal	ɓa	pa	pa	jiβa	jiβa	jiβa	jiba	*ba	*jíb
hear	pwa	pfa	pfa	twa	twa	zwa	twa	*zwa	*jígu
thorn	mu-ɓa	im-pa, im-ba	mu-pa	mu-twa	mu-twa	mu-zwa	mu-twa	*mu-zwa	*jígùà

for convenient reference. The only ones pertinent to discuss here are the final two, *jígu ‘hear’ and *jígùà ‘thorn’. In these words, PB *g must have been lost, yielding a sequence *jiua, which became *zua. In the absence of any true spirantized or labialized reflexes of *j/z, this phoneme will be represented in PTC as *zw. The different further developments seen in Copi-Tonga will be assumed to be secondary, perhaps further influenced by the presence of a noun class prefix.

Finally, a few more reflexes of *j deserve attention; these are the reflexes of PB *bàij, *jʊ́ì, and *junja in

Table 60. Aberrant reflexes of PB *j.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
carve	v’ata	ʋata	vata	βatɬa	βatɬa	βatɬa	batɬa	*ꞵazja	*bàij
word	li-to	ɗi-pʂi 1	di-prji	ri-to	ri-to	ge-zu, re-vu	ʐi-to	*li-zwi	*jʊ́ì
hare	-	tʃifundza, tʃiʋundza	-	m-pfundɮa	m-pfundɮa	vundɮa	m-pfundɮa	*-vunɮa	*junja

¹ The form attested in Lanham (1955) is ɗi-pʂ^wi, which shows additional labialization.

. The first of these unexpectedly yields a lateral affricate in Tswa-Ronga, but a stop in Copi-Tonga. Perhaps this is conditioned by the preceding *i. The second example may show the outcome of *j followed by glides, though the different reflexes do not make it clear what a single ancestral form was. It is possible that the forms outside of Copi and Lenge instead reflect deverbal nouns from the root *jígu, and only Copi and Lenge (and perhaps Tswa) truly continue *jʊ́ì. The final example appears to show some sort of spirantization of the sequence *ju, the only instance of such a sequence in the dataset. The root *junja is confined to Southern Bantu, so it is possible that this root entered Proto-Tsonga-Copi late in its existence, and therefore displays distinct developments in comparison with vocabulary inherited from Proto-Bantu. It is also worth pointing out that these outcomes are identical to what is expected for PB *gu, so it is yet another possibility that the root should be reconstructed with initial *g instead.

3.8. *g

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTT
ŋg	ŋg	ŋg	ŋg	ŋg	ŋg	ŋg	*ŋg	*N_
ŋ	ŋ	ŋ	ŋ	ŋ	ŋ	ŋ	*ŋ	*N_VN
mbv	mbv	mv	mpf	mpf	v	mpf	*mv	*N_u
?	t	t	t	t	t	t	*z?	*_i
t	t	t	t	t	t	t	*z	*i_
?	ʋ	bv	pf	pf	v	pf	*v	*#_u
f	f	f	pf	pf	v	pf	*v	*V_u
Ø	Ø	Ø	Ø	Ø	Ø	Ø	*Ø	elsewhere

The examples in

Table 61. Reflexes of PB *g in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
remain	tshala	sala	sala	sala	sala	sala	sala	*tsala	*tígad
close	-	pfalela	-	pfala	pfala	vala	pfala	*bvala	*bùgad
hand	ɣ-jandza	ɗ-andza	d-anza	ʃ-andɮa	ʃi-jandɮa	tʃ-andʒa	m-andɮa	*-anɮa	*gànjà
fingernail	-	tʃ-ala	tʃ-ala	ŋɦw-ala	-	-	ŋw-ala	*-ala	*gádà

show that PB *g is lost before *a root-initially and root-internally, and so no phoneme should be reconstructed in PTC here.

In most other environments, PB *g is reflected in the daughter languages by a glide. This glide is /w/ before rounded vowels (the final two entries in

Table 62. Reflexes of PB *g as glides.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
bewitch	loja	lojisa	loja	loja	loja	loja	loja	*loja	*dòg
abandon	-	sija	sija	sija	sija	sija	sija	*tsija	*tíg
breed	fuja	fuja	fuja	fuja	fuja	fuja	fuja	*fuja	*túg
go	ja	ja	ja	ja	ja	ja	ja	*ja	*gì
offend	-	ila	-	jila	jila	jila	jila	*ila	*gíd
travel	endra	enda	enda	eɳɖʐa	endza	enza	jendʒa	*enda	*gènd
journey	li-endra	li-endo	-	r-eɳɖʐo	ri-jendza	lw-enzo	ri-jendʒo	*lu-endo	*gèndo
fall	wa	wa	wa	wa	wa	wa	wa	*wa	*gʊ̀
seed	m-bewu	m-bewu	m-beu	m-bewu	m-bewu	m-bewu	m-bewu	*m-bewu	*bégʊ́

) and /j/ elsewhere. It seems that rather than reconstruct either *w or *j in these environments, it is most economical to assume that PB *g was lost just as it was before *a. The presence and realization of a glide can be explained as a phonological process to avoid hiatus, as found in reflexes of *j. One datum might appear to complicate this hypothesis: the Tswa reflex of the root *gèndo ‘journey’, a deverbal noun from *gènd ‘travel’, has a labial glide in the position where a reflex of *g would be expected, despite it being followed by the front vowel /e/. Though it is the only attestation of a labial phoneme in this position, it has the distinction of being unexplainable if not etymological. This however does not seem to be a reflex of *g, but in fact, a reflex of the original class 11 prefix *dʊ-, which was replaced in most of the daughter languages by the prefex /li-/ or /ri-/. Finally, the reflexes of *gì and *gʊ̀ show that consonantal glides are reconstructable for PTC as reflexes of semivowels after the loss of a preceding *g, rather than as reflexes of PB *g itself.

When followed by the PB close vowels *u and *i, spirantization occurred. Reflexes of the sequence *gu are well attested, and point to a labial affricate as a common source. The voicing of this phoneme is in question; Tsonga, Changana, and Ronga have voiceless reflexes, while Tswa shows voicing. The Copi-Tonga shows two separate reflexes with different voicing (compare the first five entries in

Table 63. Reflexes of PB *g showing spirantization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
help	-	ʋuna	bvuna	pfuna	pfuna	vuna	pfuna	*vuna	*gún
hippo	m-bvuwu	m-bvuwo	-	m-pfuβu	m-pfuβu	vuwu	m-pfubu	*m-vuꞵu	*gùbʊ́
sheep	m-bvuta 1	m-bvuta	m-vuta	ɲimpfu	ɲimpfu	ji-vu	-	*jim-vu	*gú
banana	ŋ-gombva	in-kombva	n-komva	n-kompfa	kompfa	kova	-	*mu-komva	-
nose	thombvu	thombvu	tomvu	nɦompfu	nɦompfu	nɦovu	nɦompfu	*n-thomvu	-
wasp	mi-fi	in-fi	mu-fi	mi-pfi	mu-mpfi	mi-vi	mu-pfi	*mu-vi	-
elephant	n-dzofu	n-dzofu	n-zofu	n-dɮopfu	n-dɮopfu	n-dɮovu	n-dɮopfu	*n-ɮovu	*jògù
village	-	in-ti	n-ti	mu-ti	mu-ti	mu-ti	mu-ti	*mu-(d)zi	*gìì

¹ The /-ta/ suffix in Copi-Tonga is anomalous, but the root appears to be the same as the others.

below to the subsequent two). The conditioner could be root-initial vs. root-internal position, or it could be a progressive voicing assimilation triggered by the class 9 prefix. However, only the affrication and labialization must necessarily have occurred at the level of PTC, and therefore the voicing of *g can be maintained, and a phoneme *v will be reconstructed. This would merge with PB *j in the same environment if the cognate set for *junja ‘hare’ is valid. The voicing developments could have happened in later PTC or independently in the daughter languages.

The only potential reflex of PB *gi is the root for ‘village’, which is -ti in all languages. It is therefore not clear if this is a valid series. If it is, the absence of a Tonga reflex makes it impossible to say whether this represents a merger with PB *di or *ji. The tentative merger of *gu and *ju certainly supports a merger with *ji if parallelism is assumed. It is worth noting that while /ti/ has often been cited as the Tonga outcome of PB *gi, I cannot find data here or elsewhere to support this. Gunnink et al. (2022) cite Janson (2007), who in turn cites Guthrie. Guthrie (1971, p. 64) describes the phonological developments of Tonga as “broadly similar” to Copi, but does not explicitly give /ti/ as a reflex of *gi, and none of the 69 Tonga lexemes he cites exemplify a root with unambiguous *gi. For now, PTC *z will be taken as the likely reflex of PB *gi, merging with *ji, but the absence of Tonga data remains problematic.

There is less ambiguous evidence to support *z as the outcome of PB *ig; the class 5 root *i-gʊ̀dò ‘evening, yesterday’ in

Table 64. The reflexes of PB *i-gʊ̀dò.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
yesterday	-tulo	-tulo	-tulo	tolo	tolo	-tolo	tolo	*zulo	*gʊ̀dò

is reflected either as an adverb meaning ‘yesterday’ or as a suffix on another adverb meaning ‘the day before yesterday’. Wills (2025, p. 17) observes that Z-reflexes are often seen for class 5 roots with initial *g, as the glides it yields (through lenition or as a hiatus-filler) behave as those of *j. The merger of PB *ig and *ij lends further support to the merger of *gi and *ji discussed above, but again the lack of Tonga data should not be disregarded.

In the PB complex *ng, whether a component of the root or a class 9 prefix with an initial *g, the daughter languages are unanimous in reflecting /ŋg/, and this phoneme should be reconstructed directly into PTC as *ŋg. Examples are given below in

Table 65. Prenasalized reflexes of PB *g.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
hip	ɲoŋga	-	-	ri-ɲoŋga	ɲoŋga	ɲoŋga	ɲoŋga	*ɲoŋga	*jóngà
fog	huŋguv’a	ɦuŋguʋa	-	ɦuŋguβa	huŋguβa	huŋguβa	ɦuŋguba	*ŋ-kuŋgu	*kʊ̀ngʊ́
enter	-	ŋgena	-	ŋgɦena	ŋgɦena	eŋgena	ɦiŋgena	*iŋgena	*jíngɪn
leopard	jiŋ-gwe	iŋ-gwe	-	ɲiŋgwe	ɲiŋgwe	ɲiŋgwenzala	jiŋ-gwe	*jin-gwe	*gòì
crocodile	ŋgona	ŋgona	ŋgona	ŋgweɲa	ŋgweɲa	ŋgweɲa	ŋgweɲa	*ŋ-gwina	*gòìnà
drum	ŋgoma	ŋgoma	ŋgoma	ŋgoma	ŋgoma	ŋgoma	ŋgoma	*ŋ-goma	*gòmà

.

The PB class 9 root *gàngà ‘witchdoctor’ is attested in all languages under consideration as ŋaŋga. Van der Spuy (1989, p. 94) suggests that this may be an instance of Meinhof’s rule, which causes nasal stop sequences in consecutive syllables to dissimilate. Van der Spuy notes that the evidence for Meinhof’s rule being active within Southern Bantu is marginal, with no reconstructable stems having the shape NCVN-. I contend that the reflex of *gàngà is indeed a product of Meinhof’s rule, though not necessarily phonologically active within PTC. The rationale for this is discussed in Section 4.2.

3.9. *m

As seen in

Table 66. Reflexes of PB *m in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
sprout	-	mila	mila	mila	mila	mila	mila	*mila	*mìd
meat	ɲama	ɲama	ɲama	ɲama	ɲama	ɲama	ɲama	*ɲama	*nyàmà
plant	lima	ɗima	dima	rima	rima	rima	dʒima	*lima	*dìm
be wide	anama	anama	anama	anama	anama	anama	janama	*anama	-

, PB *m remains largely unchanged across Tsonga-Copi, being reflected as /m/ in most positions in all the languages under consideration.

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC
mw	mw	mw	ŋw	ŋw	ŋw	ŋw	*mw	*_ʊV
m	m	m	m	m	m	m	*m	elsewhere

When followed by a labial glide, *m is instead reflected by /ŋ/ in Tswa-Ronga, as seen in

Table 67. Reflexes of PB *m followed by glides.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
one	-mwe-	-mwe-	-mwe-	-ŋwe	-ŋwe	-ŋwe	-ŋwe	*mwe	*mòì
year	mw-aɣa	mw-akani	-	naŋwaka	naŋwaka	ŋw-akani	ŋw-aka	*mw-aka	*jákà
child	ɣj-anana	mw-anana	mw-anana	ŋw-ana	ŋw-ana	ŋw-ana	ŋw-ana	*mw-ana	*jánà

. Most words that demonstrate this are vowel-initial roots of class 1 or 3 traditionally reconstructed with initial *j in PB. As Copi-Tonga shows, this resulted in the PTC sequence *mw, with development into /ŋ^w/ a Tswa-Ronga internal development.

3.10. *n

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTT
n	n	n	ɲ	ɲ	ɲ	ɲ	*n	*_i, *i_
n	n	n	n	n	n	n	*n	elsewhere

Table 68. Reflexes of PB *n in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
see	wona	wona	vona	βona	βona	wona	bona	*ꞵona	*bón
stink	nuŋga	nuŋga	nuŋga	nuɦa	nuŋɦwa	nuha	nuɦa	*nuŋka	*nùnk
help	-	ʋuna	bvuna	pfuna	pfuna	vuna	pfuna	*vuna	*gún
enter	-	ŋgena	-	ŋgɦena	ŋgɦena	eŋgena	ɦiŋgena	*iŋgena	*jíngɪn
good	-	-nene	-nene	-nene	-nene	-nene	-nene	*-nene	-

shows that in most environments in the daughter languages, PB *n is reflected as /n/.

However, in Tswa-Ronga, *n is reflected as palatal /ɲ/ when adjacent to PB *i. It is quite clear that a following *i conditions this change (shown by the first three entries in

Table 69. Reflexes of PB *n showing palatalization.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
give	niŋga	niŋga 1	-	ɲika	ɲika	ɲika	ɲika	*ɲiŋka	*nínk
firewood	khuni	li-khuni	kuni	ti-ɦuɲi	ri-huɲi	ti-huɲi	ti-ɦuɲi	*-kuni	*kʊ́nì
trunk	li-tshina	ɗ-itshina	n-tsiɲa	n-siɲa	n-siɲa	tsiɲa	n-siɲa	*tshina	*tínà
crocodile	ŋ-gona	ŋ-gona	ŋ-gona	ŋ-gweɲa	ŋ-gweɲa	ŋ-gweɲa	ŋ-gweɲa	*ŋ-gwina	*gòìnà
tooth	-	ɗi-no	di-no	ti-ɲo	ti-ɲo	ti-no	ti-ɲo	*zi-no	*jínò

¹ Also attested for Copi and Lenge is the form nika, which still does not show palatalization.

below), but the situation is less clear for a preceding *i. Of the reflexes of *jínò in Tswa-Ronga, three of the four show palatalization. This situation is reversed for the plural, for which Ronga has meɲo and the rest meno. This situation could be explained by postulating that the *a in the PB class 6 prefix and the *i in the root had already coalesced to *e (after prior loss of intervocalic *j) before the palatalization rule took effect. However, that would preclude the explanation for the reflexes of *gòìnà, unless for some reason *a and *i coalesced before *o and *i. Any hypothesis which requires the palatalization rule before vowel coalescence would require this alternation to be reconstructed at the time of PTC. At the moment, there does not yet seem to be enough data to come to a distinct conclusion, so I reconstruct *n in all cases.

3.11. *ɲ

Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTT
ɲ	ɲ	ɲ	ɲ	ɲ	ɲ	ɲ	ɲ	everywhere

There are only three cognates which reflect PB *ɲ, given in

Table 70. Reflexes of PB *ɲ in Tsonga-Copi.

English	Tonga	Copi	Lenge	Tsonga	Chang.	Tswa	Ronga	PTC	BLR3
meat	ɲama	ɲama	ɲama	ɲama	ɲama	ɲama	ɲama	*ɲama	*nyàmà 1
salt	mu-ɲu	mu-ɲu	mu-ɲu	mu-ɲu	mu-ɲu	mu-ɲu	mu-ɲu	*mu-ɲu	*jínyʊ̀
door	ɲaŋgwa	-	-	ɲaŋgwa	ɲaŋgwa	ɲaŋgwa	ɲaŋgwa	*ɲaŋgwa	*nyàngò

¹ BLR3 uses <ny> to represent PB *ɲ.

, but all of the present languages reflect it identically as /ɲ/. This phoneme should therefore be reconstructed as *ɲ for PTC.

3.12. The Consonants of Proto-Tsonga-Copi

Following all the material discussed earlier in this section, the reconstructed consonant inventory of Proto-Tsonga-Copi is given in

Table 71. The consonant inventory of Proto-Tsonga-Copi.

		Bilabial	Labiodental	Labio-Alveolar	Alveolar	Lateral	Palatal	Velar
Nasal		*m			*n		*ɲ	*ŋ
Stop	Voiceless	(*p)			(*t)			*k
	Voiceless Aspirate	*ph			*th			*kh
	Voiced	*b			*d
	Voiced Aspirate	*bɦ			*dɦ
Affricate	Voiceless			*ps	*ts
	Voiceless Aspirate			*psh	*tsh
	Voiced			*bz	*dz
Fricative	Voiceless		*f		*s	*ɬ
	“Strong”				*is	*iɬ
	Voiced	*ꞵ	*v		*z	*ɮ
Liquid					*l
Glide		*w					*j

.

Based on the above system, it is likely that there was a phonological repairing of certain sounds and their prenasalized counterparts. The etymological prenasalized counterparts of the voiced aspirates *b^ɦ and *d^ɦ would be *mp^h and *nt^h, the pairs deriving from PB *p and *t, respectively. Synchronically, in PTC, these must have been analyzed as counterparts to *p^h and *t^h. The borrowed PTC phonemes *p and *t would have had no prenasalized counterparts, but new ones may have been created via the introduction of loanwords in class 9, beginning in [p] and [t] or via the reduction of the noun class 1/3 prefix.

As for the voiced stops, the system suggested here might explain the synchronic irregularity of Tsonga/Changana aspiration discussed in Janson (2001). While PTC *b^ɦ and *d^ɦ generally yielded /ɦ/ and /r^ɦ/ in Tsonga, it seems that their value as stops was preserved in certain class 1, 3, and 18 forms. These sequences yielded the prenasalized aspirated consonants /mb^ɦ/ and /ndz^ɦ/, which would be synchronically contrastive with /mb/ < PB *mb and /ndz/ < PB *nd. This explains the presence of contrasting prenasalized voiced stops and prenasalized voiced aspirate stops in the absence of a corresponding set of voiced aspirate stops without prenasalization, but it does not explain the frequent coexistence of aspirated and unaspirated forms of the same lexeme which Janson (2001) notes. Perhaps this is due to a relative lack of minimal pairs between the aspirates and voiced aspirates, and so the system is in a state of entropy.

The system as a whole displays several typological regularities. In general, each obstruent phoneme occurs as part of a pair, with each voiced obstruent having a corresponding voiceless obstruent aside from /ꞵ/. Among the voiced stops, a gap at /g/ like that shown here is a typical occurrence. Per the voiceless stops, the expected gap would be at /p/. This supports the notion of *p as a marginal phoneme in PTC. It is interesting to note that the space of *t, another marginal phoneme, was likely filled by *z at the end of PTC as part of a chain shift, but no such development occurred with *p.

The most difficult issue still remaining is the phonetic nature of the sounds I have labeled ’strong’. These are the fricative reflexes of PB *c and j following *i. Elsewhere in this environment, voiceless stops were aspirated and voiced stops were unchanged (though their Tonga-Copi reflexes had implosives). It is very tempting to unify these within a single phonological class. It is possible that these sounds were at some stage all geminated; as long vowels and diphthongs were eliminated across Southern Bantu, the creation of geminates would serve to preserve an existing moraic structure. The question is whether geminates could yield outcomes similar to aspirates, or even aspirates themselves at the intermediate stage. A systematic change of (voiceless) geminates to aspirates is attested in some New Caledonian languages (Ozanne-Rivierre, 1992), but comparisons within Bantu would be welcome.

4. Discussion

4.1. Proto-Tsonga-Copi Within Southern Bantu

Up until now, this paper has demonstrated an array of sound changes which are shared across the Tsonga-Copi clade. Critically, the demonstration of these shared innovations is not sufficient do demonstrate the validity of the clade; to do this, one would have to demonstrate that the clade shows innovations which are not shared by the rest of Southern Bantu. Therefore, a detailed systematic comparison of this reconstruction with other reconstructed Southern Bantu protoforms would be necessary. This goes outside the scope of the present paper, but it is still possible to offer a few initial observations that could direct future research.

Figure 1 below shows the phylogeny of Southern Bantu generated by Gunnink et al. (2025). As discussed earlier, this phylogeny groups Tswa-Ronga and Copi-Tonga together under a single parent node, implying common descent from a single ancestral language which the present paper sought to reconstruct. One node above is a trinary node which links Tsonga-Copi with Nguni (S40) and Sotho-Tswana (S30); this node has support value of 1, meaning that this node appeared in 100% of the models generated by the program. This is in contrast to the same authors’ earlier (2022) study, which grouped only Tsonga-Copi and Nguni together in a node with a support value of 0.91.

It makes sense then to focus comparison within Southern Bantu towards Nguni and Sotho-Tswana. Unfortunately, the necessary data is hard to come by. Ownby (1985) makes use of Nguni language data primarily for reconstruction of the cultural lexicon, but his work focuses little on the particulars of phonology and is therefore not very useful for the purpose necessary here. Msimang (1989) has a more robust phonological study of the Tekela subgroup of Nguni languages, but this group only represents around half of the Nguni languages. Gunnink (2022) reconstructed the inventory of clicks in Proto-Nguni, but as clicks are marginal in Tsonga-Copi, they are not an ideal point of comparison. The most recent reconstruction of Proto-Nguni phonology is by Van der Vlugt and Gunnink (2025), who focus specifically on lateral obstruents. This provides some initial ground for comparison, though a detailed and thorough reconstruction of the phonology of Proto-Nguni as a whole would be welcome. As for the body of historical linguistic work in Sotho-Tswana languages, Dickens (1986) seems to offer a fairly robust reconstruction of Proto-Sotho-Tswana based on seven language varieties. It is well received by Creissels (1999), who provides historical phonological data specific to Tswana. Future research may well begin here, though it is unlikely that such research would prove useful without the corresponding Nguni data.

Based on the available data, an ideal point to begin comparison is with the lateral obstruents. The cross-linguistic rarity of these sounds was noted earlier, yet they can be found in multiple Southern Bantu clades: Shona (S10), Sotho-Tswana, Nguni, and Tsonga-Copi. If a single point of origin can be traced for these sounds, it would be a strong diagnostic criterion for a genetic family grouping. Van der Vlugt and Gunnink reconstruct both *ɬ and *nɮ in Proto-Nguni as reflexes of Proto-Bantu *c and *nj. This is entirely identical to the situation I reconstruct for Proto-Tsonga-Copi, which itself is a strong argument for a Nguni-Tsonga-Copi subclade. Further research into the lateral obstruents in Sotho-Tswana would be welcome to see if they too are reconstructable for the protolanguage in the same environment.

Another set of potentially diagnostic sound changes are the outcomes of Proto-Bantu spirantization. Gunnink et al. (2022) note that while spirantization is not present in Sotho-Tswana, and therefore cannot be reconstructed for Proto-Southern Bantu, similar outcomes of spirantization in the daughter languages could still identify lower-order groupings. They observe a particular degree of shared innovation between Nguni, Tsonga, and Copi on one hand and Shona and Venda on the other. They offer a speculative reconstruction of the outcome of spirantization in Proto-Nguni-Tsonga-Copi, which is given in

Table 72. Comparison of proposed spirantization outcomes for various Bantu stages.

Proto-Bantu

*pu

*tu

*ku

*bu

*du

*gu

*pi

*ti

*ki

*bi

*di

*gi

Proto-NTC

°fu

°fu

°fu

°bvu

°dzu

°bvu

°fi

°si

°si

°vi

°zi

°zi

Proto-TC

*fu

*fu?

*fu

*bvu

*dzu

*bvu

*psi

*tsi

*si

*bzi

*dzi

*dzi

alongside the corresponding reconstructions for Proto-Tsonga-Copi identified in this study.

Overall, there is a large degree of overlap between the speculative reconstruction and what is actually seen in Tsonga-Copi, but there are also a few points where there are differences. I reconstruct affricate reflexes of PB *ti, *di, and *gi for Proto-Tsonga-Copi, but there is no reason that the tentative Proto-Nguni-Tsonga-Copi reflexes could not be reconstructed as affricates as well.

A more substantial difference is in the outcomes of PB *pi and *bi, where Gunnink et al. posit *fi and *vi but I reconstruct *psi and *bzi. This was due to /s/ being the expected outcome of both series in Copi-Tonga, a development more similar to spirantization in Shona/Venda. Gunnink (p.c.) proposes a solution of original *fi > *si in Copi-Tonga, which is certainly a possibility. In Copi-Tonga, etymological /f/ always occurs as a reflex of PTC *fu, so the existence of *fi would not have any consequences for the diachronic phonology. If a parallel Copi-Tonga development of *vi > *si is also assumed, the phonemes *ps and *bz need not be reconstructed for Proto-Tsonga-Copi at all11. In sum, the outcomes of spirantization in Nguni, Tsonga, and Copi provide more evidence for their shared relationship than otherwise.

As for what might identify Tsonga-Copi as a distinct clade, the most significant systematic change seems to be in the voicing of the reflexes of PB *p and *t. Based on Guthrie (1971), it seems that Sotho-Tswana and Venda each have /r/ < *t, but Nguni does not, nor does any Southern Bantu language show *p > /b/. However, a single change is not enough to demonstrate the validity of a clade, so more lexical phonological reconstructions from other Bantu clades would be welcome in approaching this issue further.

4.2. Double Reflexes of *nj

Of the changes laid out in Section 3, one of the most remarkable to me is the double reflexes of *nj, where initial PB *nj yields *nɮ in PTC before low-tone vowels and *ɲ before high-tone vowels. The potential of tone to condition double reflexes of the Proto-Bantu voiced stops *b and *d has already been observed in Northwestern Bantu, as detailed in Philippson (2022), though similar observations for *j were not noted. Regardless, none of the reconstructions given as evidence are controversial, and the reflexes in the daughter languages are all perfect cognates, obeying all relevant sound laws identified throughout this work. Given the peculiarity of this development, it is worth investigating further to ensure that it is not conditioned by more opaque segmental phonology or simply a rather spectacular coincidence.

To do this, I returned to BLR3 and examined all class 9 reconstructions with initial *j. Of the 30 such terms, 21 were labeled as attested in Southern Bantu. Using the established sound laws, I attempted to find cognates in the S50 and S60 languages. Excluding the nine items already presented in Table 57, the remaining 12 PB reconstructions are given below along with apparent cognates where found.

(1)	*jàdí	‘lightning’	>	Tsonga ndɮati, Tonga ndzadzi
(2)	*jʊ̀gʊ́	‘groundnut’	>	Tsonga ndɮuwu
(3)	*jʊ̀ndò	‘hammer/anvil’	>	Tsonga ɲundzu, Copi ɲundo
(4)	*jʊ̀mbá	‘house’	>	Copi ɲumba
(5)	*jàngé	‘heron/egret’	>	Tsonga muɲaŋgana (Class 1)
(6)	*jìkà	‘grassland/desert’	>	?
(7)	*jànjá	‘lake’	>	?
(8)	*jàtí	‘grass’	>	Copi ndzari ‘porch/balcony’?
(9)	*jʊ̀	‘house’	>	Tsonga jindɮu, but intervocalic
(10)	*jénjé	‘cricket’	>	Tsonga ɲendɮe
(11)	*jádà	‘fingernail/claw’	>	Tonga ɲala
(12)	*jóngò	‘bile’	>	Tsonga ɲoŋgwa, Copi ɲoŋgwa

As can be seen from the data, the sound law appears to hold for all entries except (3), (4), and (5), where an unexpected /ɲ/ occurs in both the Tsonga and Copi languages. The question is then whether these forms have anything in common which would allow the sound law to be reformulated. As it turns out, they do. Each of (3–5) contains a root-internal prenasalized (voiced) stop. It appears that the presence of the sequence always yields the high-tone form. While the tones of the relevant terms in the daughter languages are not always present in the data, the Tsonga forms in (3) and (5) each maintain a low tone on the conditioning vowel, so it does not appear that the presence of a prenasalized stop affected the development of the autosegmental phonology. It is also unclear whether the internal PB sequence *nj triggers the same development as PB *mb, *nd, and *ng, as none of the daughter languages show a clear reflex of (7).

This exception to the sound law appears to be less of an exception, and more of a case of Meinhof’s rule eliminating the conditioning environment. As was noted in Section 3.8, the reflex of PB *gàngà (with class 9 prefix *N-) was ŋaŋga in all the languages in this study. If Meinhof’s rule for PTC is formulated as PB *NC[+voice] > PTC *N / _VNC[+voice], it would explain both the outcome of *gàngà as well as those of *jʊ̀ndò, *jʊ̀mbá and *jàngé. It is clear that this rule operated before the tone law in the chronology, but whether it was still a phonologically active process in PTC is uncertain. PB *N-pòngó ‘billy-goat’ yielded PTC *m-p^hoŋgo, per Section 3.1. This suggests that Meinhof’s rule did not apply to prenasalized voiceless stops in PTC, but it is also possible that the aspiration is relevant for the formulation of the rule itself, which further research could clarify.

It seems clear then that within Proto-Tsonga-Copi, the tonally conditioned double reflexes of *nj are expected. But how far back must this phenomenon have developed? Van der Vlugt and Gunnink (2025, p. 13) demonstrate /ⁿdɮ/ to be the regular outcome of *nj in Nguni languages, citing five class 9 lexemes: *jàdà ‘starvation, hunger’, *jèbé ‘ear, lobe’, *jògù ‘elephant’, jìdà ‘path’, and *jʊ̀ ‘house’. Interestingly, all of these have low tone, making Nguni appear to follow the same rule as Tsonga-Copi here. Unfortunately, there are no lexemes included which derive from a PB root with initial high tone, so it is unclear if double reflexes are present. Creissels (1999) presents a list of *nj correspondences for the Sotho-Tswana language Tswana (S31) which includes roots with both initial high- and low-tone vowels. Of these, nearly all conform to the tone rule as it operates in Tsonga-Copi, with a few apparent exceptions: *jàmà ‘meat’, *jàmbí ‘God’, *jʊ̀ní ‘bird’, and *jʊ̀ndò ‘hammer’. The first of these is often reconstructed with initial *ɲ at the level of Proto-Bantu. Besides this, all these exceptions have a low tone with a nasal consonant in the second syllable. If the treatment of Meinhof’s rule described above is expanded to refer to any nasal consonant in the second syllable, rather than just a prenasalized stop, then all exceptions here are covered. The only instance where a high-tone root appeared to show a low-tone reflex was *jáì ‘outside’. In Tsonga-Copi, this root was generally prefixed with a locative class marker, causing the *nj sequence to appear word-internally. It seems perfectly reasonable to postulate that at some point in the historical chronology, the Tswana root might have had a similar shape. In any case, the double reflexes of PB *nj appear to arise quite regularly according to tone within both Tsonga-Copi and its reported closest relatives within Southern Bantu.

Expanding the scope to Bantu as a whole, Wills (2022) identifies several class 9 nouns traditionally reconstructed with *j and divides them into groups appearing to reflect original *nj/*nz and original *ny. Of the roots pointing to *nj/*nz, all have an initial low-tone vowel. Of those which point to *ny, most have an initial high tone, with *jùmá ‘rear, behind’ and *jʊ̀ní ‘bird’ being apparent exceptions. As was the case for exceptions in Tswana, these both have a nasal in the second syllable. Wills (2022, p. 92) concludes that “neither the influence of tone nor a subsequent vowel would give us a phonological rule to generate the strong [*nz] reflexes”, and goes on to offer influence from the subsequent consonant as a possible explanation. I propose that a combination of the two—both the presence of a low-tone vowel in the first syllable and the absence of a nasal consonant in the second syllable—is almost certainly sufficient to explain the situation in Southern Bantu, and perhaps can also be extended to Bantu as a whole.

4.3. Evidence for Language Contact

If one reexamines the phonological inventories of the S50 and S60 languages and marks what can be explained by sound law, an interesting pattern with great implications for the history of Southern Bantu languages can be seen.

Table 73. Consonant inventory of Tsonga indicating uninherited phonemes.

		Bilabial	Labiodental	Alveolar	Lateral	Postalveolar	Retroflex	Palatal	Velar	Glottal
Nasal	Plain	m		n				ɲ	ŋ
	Aspirated	mɦ		nɦ					ŋɦ
Stop	Voiceless	p		t					k
	Voiceless Aspirated	ph		th					kh
	Voiced	b		d					g
	Voiced Aspirated	bɦ		dɦ					gɦ
Affricate	Voiceless		pf	ts	tɬ	tʃ	ʈʂ
	Voiceless Aspirated		pfh	tsh	tɬh	tʃh	ʈʂh
	Voiced		bv	dz	dɮ	dʒ	ɖʐ
	Voiced Aspirated		bvɦ	dzɦ	dɮɦ	dʒɦ
Fricative	Voiceless		f	s	ɬ	ʃ	ʂ
	Voiced	β	v	z			ʐ			ɦ
Liquid	Plain			r	l
	Aspirated			rɦ
Glide		w						j

reproduces the consonant inventory of Tsonga given in Section 1.3. Phonemes which cannot be derived directly from Proto-Bantu via sound law are bolded and underlined.

It is striking that 20 of the 53 Tsonga phonemes (38%), almost half of the inventory, do not appear to be inherited. Compare that to Tonga, for which a consonant inventory is given in

Table 74. Consonant inventory of Tonga indicating uninherited phonemes.

		Bilabial	Labio-Dental	Alveolar	Lateral	Palatal	Velar	Glottal
Nasal		m		n		ɲ	(ŋ)
Stop	Voiceless	p		t			k
	Voiceless Aspirated	ph		th			kh
	Voiced	b		d
	Implosive	ɓ	v’	ɗ
Affricate	Voiceless		pf	ts
	Voiceless Aspirated		pfh	tsh
	Voiced		bv	dz
Fricative	Voiceless		f	s	(ɬ)			h
	Voiced	β		(z)			ɣ
Liquid				r	l
Glide		w				j

. In Tonga, only 2 of the 29 phonemes (7%) not claimed to be loans cannot be explained via sound law. Adding in the data from Copi (7/37, 19%), Tswa (10/37, 27%), and Ronga (19/47, 39%), it is clear that much more of the synchronic inventories of S50 languages are unexplainable in comparison to S60. So, what could the origin of this discrepancy be?

If a phoneme’s origin cannot be explained through sound law, it may instead have arrived via language contact, which begs the question of which language the contact was with. The mark that Khoisan languages have left on Tsonga-Copi languages is clear: all of them contain clicks, which must have been borrowed either directly into Tsonga-Copi or indirectly by means of another language. But clicks are not the only way that Khoisan may have influenced the development of Copi-Tonga. Lanham (1964) describes the impact of Khoisan languages on the expansion of phonemic systems in Bantu languages, and a particular target for expansion was the system of affricates. This is consistent with the data present, in which the more westerly (and therefore more proximal to Khoisan) S50 languages show a greater number of unexplainable phonemes, particularly affricates. Beyond this, Van der Vlugt and Gunnink (2025) outline the role that Khoisan has played in the development of lateral obstruents within Nguni. In addition to the velar lateral obstruents which have been borrowed directly from Khoisan into Nguni, the phonological shift from PB *c and *j to lateral obstruents may have been brought about or aided by substrate influence from Khoisan speakers shifting to Proto-Nguni. If the innovation of lateral obstruents for PB *c and *j in Southern Bantu was a singular occurrence, then this language shift is also a part of the prehistory of Proto-Tsonga-Copi.

It also cannot be ruled out that a source of contact-induced change was another variety of Bantu. As noted in Section 3.1 and Section 3.2, PTC had several terms with Proto-Bantu etymologies, but aberrant reflexes of PB *p and *t. If these terms were interpreted as borrowings from another Bantu variety, it would explain why these apparently unexpected reflexes still appear to reflect valid Bantu etymologies. While these borrowings could be dated to the time of PTC, they could also be borrowed at the level of PSB, which would have had a series of voiceless aspirates but no unaspirated voiceless stops. This explanation aligns well with the spread-over-spread hypothesis for Southern Bantu given in Gunnink et al. (2022), in which Southern Bantu speakers migrated into territory that had already been previously occupied by Eastern Bantu-speaking populations. However, as one reviewer notes, the data could instead point to later contact with Shona speakers.

One instance of likely Shona borrowing was already noted earlier in the apparent loanword bika ‘cook’. Besides this, Guthrie (1971, p. 62) finds that the regular outcomes of PB *p and *t in Shona are /p/ and /t/, which aligns perfectly with the reconstructions of the aberrant reflexes in Proto-Tsonga-Copi.

Table 75. Comparison of Shona to PTC reconstructions with *p and *t.

English	Shona 1	PTC	BLR3
split	panda	*panda	*pànd
measure	pima	*pima	*pìm
fold	peta	*peta	*pèt
baldness, skull	ru-para	*lu-pala	*pádá
price, quantity	mu-teŋgo	*mu-teŋgo	*téngò
feather	mu-nɦeŋga	*mu-teŋga	*tèngá

¹ These terms are collectively attested in all dialects surveyed by Hannan, though only the Manyika and Zezuru dialects attest all of these terms.

compares Shona vocabulary from Hannan’s (1974) dictionary with all PTC reconstructions containing *p and *t. Four of the reconstructions are identical to the modern Shona terms. The reflexes of *pádá ‘baldness, skull’ have regular *l in PTC and /r/ in Shona from PB *d. This suggests borrowing of this term either preceding the change of *d (or *l) > /r/ or from a dialect which did not undergo this change. One could also suggest that *l was an adaptation of /r/ to the PTC phonology which lacked it, but then one would have to explain why *p was not similarly adapted. The Shona reflex of PC *tèngá ‘feather’ suggests an initial nasal, with PB *nt > Shona /nɦ/, though no class 9 form is cited by Hannan. In any case, the overall high degree of similarity implies two possibilities: either PTC borrowed a reasonably large quantity of Shona vocabulary, or both PTC and Shona borrowed from a third unidentified language. The latter cannot be disproven, though there is substantial historical evidence to support the former.

The Shona-speaking population of Great Zimbabwe, which reached its peak in size and influence between the 13th and 15th centuries CE, was extensively connected to the Indian Ocean trade network (Mitchell, 2024). This would have put them in direct contact with early Tsonga-Copi speakers, whose historical and modern range sits just to the southeast, centering on the mouth of the Limpopo River. The presence of potential trade words such as ‘split’, ‘measure’ and ‘price/quantity’, noted by one reviewer, lends support to a commercial nature of contact. Besides this, multiple migrations of Shona speakers into territory occupied by Tsonga-Copi speakers are present in the historical record. Smith (1973 p. 572) reports Ronga accounts of waves of Shona or Sotho migrations into the Delagoa Bay (now Maputo Bay) between the first half of the second millennium and the 16th century. The chiefdoms founded by these migrants would go on to be some of the largest found among the Tsonga peoples. Another wave of Shona migrants in the 15th century arrived at the coast of Mozambique roughly 90 miles south of Inhambane in an area called ‘Wutonga’ which Smith (1973, p. 572) claims was populated by the ancestors of the Tonga. These migrants, fleeing civil war, would establish a dynasty which proliferated over the following centuries, giving rise to an array of southern chiefdoms whose mixed Shona-Tonga culture differed from the Tonga populations around them. The powerful chiefdoms founded by Shona-speaking migrants across the Tsonga-Copi territory would be ideal sources for lexical borrowing in early Tsonga-Copi; the only question then remains the overall time-depth of Tsonga-Copi’s split. This question will be left to future research, though the ubiquity of the Shona superstrate across Tsonga-Copi suggests a break-up towards the end of or after the period of Shona interactions and migrations between the 13^th and 16^th centuries.

Even in modern times, language continues to shift. As noted in Section 1, speakers of virtually all language varieties of Tsonga-Copi have a high level of bilingualism, often speaking at least one or two other Bantu varieties. This is to say nothing about non-Bantu languages such as English, Portuguese, and Afrikaans, which have undoubtedly altered the lexicon and perhaps phonologies of Bantu languages across Southern Africa. Returning to Bantu, it is often unclear where the boundaries between a language and a dialect lie, making it difficult to identify whether forms are representative of a particular variety, or if a given variety exists as a homogenous entity at all. The high degree of mobility that speakers have both in present and past, compared with the high level of relatedness among the languages they speak to begin with, makes it difficult to tease apart the effects of language contact in the present just as much as in ancient times. For all languages encompassed in this study, a detailed and systematic analysis of regional dialectology and sociolinguistics would be welcome.

5. Conclusions

For every consonant phoneme in Proto-Bantu, there emerged clear and regular phonological outcomes in the Tswa-Ronga (S50) and Copi-Tonga (S60) languages. From these, it was possible to derive a relatively complete consonantal inventory of the hypothetical intermediate between Tswa-Ronga, Copi-Tonga, and Proto-Bantu, which was dubbed Proto-Tsonga-Copi. This inventory was both typologically plausible, and sufficiently distinct from both Proto-Bantu and attempted reconstructions of Proto-Southern Bantu to suggest a discrete subclade.

However, as noted earlier, the scope of this study is not sufficient to fully demonstrate the validity of Tsonga-Copi as a genetic unit. In order to do this, it would have to be shown that the languages of Tsonga-Copi shared innovations which were unique to them, to the exclusion of the other groups of Southern Bantu. Initial avenues for comparison were already presented, with the apparently shared innovation of lateral obstruents and outcomes of spirantization suggesting a close affinity between Nguni and Tsonga-Copi. The next step would be a thorough reconstruction of Proto-Nguni to serve as a point of comparison, so that developments uniquely shared by Tsonga-Copi languages could be more easily identified. Following this, comparison to other Southern Bantu varieties is welcome, particularly to Sotho-Tswana, for which Gunnink et al. (2025) report a high level of affinity for Tsonga-Copi and Nguni.

At the level of PTC itself, there is still a large amount of work to be done. The reflexes of Bantu spirantization in Copi-Tonga are still unclear for a few phonemes, which makes it difficult to draw strong conclusions for PTC. Spirantized prenasalized consonants are hardly attested, marking yet another gap in the data. Finally, more examples of consonants preceding semivowels would be useful, especially the labials, as their outcomes are quite opaque on the surface, and there seems to be much dialectical variation within the daughter languages. Nonetheless, a vast quantity of useful correspondences have been established. Some of these have potential implications for the phonology of Proto-Bantu as a whole, such as the tonally conditioned double reflexes of PB *nj.

A major issue at present is the very limited data which exists on these languages, particularly those of S60. There are only so many lexemes given by Amaral et al. (2007) and Dos Santos (1949). But at this point, enough sound laws should be evident for PTC that theoretical correspondent forms could be deduced and searched for. This would be aided if consultant speakers could be found and asked for these hypothetical forms.

It is clear that despite the limitations which this study faced, the results yield substantial new information about the prehistory of certain Southern Bantu language groups. The linguistic evidence presented here includes substantial data from the S60 languages which was missing in prior comparative studies. The findings are consistent with existing and emerging hypotheses on the prehistory of Southern Africa and provide new argumentation and avenues for further exploration on that front. There nevertheless remains room for more work on historical reconstruction in Southern Bantu to gradually paint a clearer and clearer picture of the people who lived there and the languages that they spoke.