Preverbal clitic clusters in the Tanzanian Rift Valley revisited

2.1 Context of the study

KMN (2008) “The Tanzanian Rift Valley Area”, a chapter in the larger title A Linguistic Geography of Africa (Heine and Nurse eds.) is the first work to examine preverbal clitic complexes of the TRV together. In broad terms, the chapter section (pages 197–206) observes that, although preverbal clitic complexes of this type [i.e. those which have extended the complex through fusion with conjunctions and adpositions etc.] are characteristic of Southern Cushitic, they are not typical in either Bantu or Nilotic. However, within the TRV, both Bantu (esp. Nyaturu) and Nilotic (esp. Gisamjanga Datooga) languages “display incipient preverbal clitic clustering” (page 198). From a historical viewpoint (page 199), preverbal clitic complexes in both Nyaturu and Gisamjanga Datooga are said to have arisen as a result of long-term contact between speakers of pre-Nyaturu and speakers of pre-Gisamjanga with speakers of Proto West Rift (the hypothetical ancestor of the current South Cushitic languages) (page 199). These are observations from KMN’s (2008) paper to which we will return over the course of the paper as we explore these constructions in more detail. The Bantu languages Nyilamba, as well as Kimbu (though the data for this latter language is sparse) do have preverbal clitic clusters, but not to the same degree as Nyaturu (page 201). The same historical dynamic as proposed for Nyaturu is also proposed for these two languages (page 199). Sandawe (possibly either a language isolate or a distant member of Khoe-Kwadi) has a series of person-gender-number (PGN) markers, which are taken as a close analogue of preverbal clitic complexes (page 205). Historically (and because of similarities with forms in Khoe-Kwadi), it is judged that the Sandawe and South Cushitic forms are due to chance rather than contact, though the two language groups being in contact may have reinforced the structures present in each (page 205). The Hadza language (isolate) is seen as having similar structures (page 203), but it was decided that the available data were not sufficient for further analysis (page 206). The other languages in the sample, Rangi and Mbugwe, are judged to not have preverbal clitic complexes, and are not discussed in the section.

KMN (2008) is the model on which we have based this paper, as such, we will be considering all the languages included in their chapter. We also focus on the structures identified therein as PCCs. Following this brief summary, a few statements can now be made about the limits of KMN (2008), and how this paper will respond to them.

First, and as a result of the fact that their work could only dedicate one subsection of one book chapter to the topic, examples of the preverbal clitic clusters of each language of the area were not provided. In the present paper, we have space to provide examples and analysis of examples for every language for which we have data. Second, KMN (2008) takes the South Cushitic PCC to be the prototype for examinations of the structures under assessment. Under this assumption, South Cushitic is taken both to be the ‘source’ of the prevalence and concentration of this structure in the region, and to constitute the structure on which the criteria for what constitutes PCC is based. The present paper critically examines this position, giving a finer-grained examination of the preverbal clitic clusters mentioned, both in terms of their similarities and differences to the South Cushitic prototype, as well as their proposed historical origins. The picture that emerges shows a contact situation which is rather more complex. Third, several languages are considered homogeneous in terms of how their PCCs behave.^[1] The South Cushitic languages (Alagwa, Burunge, Gorwaa, and Iraqw) are all considered as a single group. Nyilamba and Kimbu are considered another group. Rangi and Mbugwe are considered another, and the Datooga varieties yet another. The present paper examines these assumed groupings, and indicates where forms differ.

Since the publication of KMN (2008), Anderson (2011) has also commented on the preverbal clitic complexes of the TRV, this time from the point of view of auxiliary verb constructions (AVCs). Anderson does not aim to demonstrate that the constructions under discussion constitute AVCs, but rather operates under the assumption that they do, and uses a typology of AVCs to compare the constructions based on their morphosyntactic and semantic features. Anderson finds no areal AVC profile for the TRV and reports that many of the languages reflect the typological profiles of their genetic groupings. Anderson proposes that synchronic AVCs in Sandawe and Hadza indicate that the two languages previously had Verb-Auxiliary order.

2.2 The languages of the study

The Tanzanian Rift Valley Area has been the site of extensive, sustained contact between languages of different linguistic stocks for a very long time indeed. What follows is a brief introduction to the languages of our sample, with languages presented alphabetically within their larger family (if applicable), and arranged from the family with the most numerous members in the sample (Bantu) to the convincingly-argued isolate, Hadza (Figure 1).

Figure 1:

Map of the Tanzanian Rift Valley Area (from KMN 2008: 187).

2.2.1 The South Cushitic languages

Forming part of the larger Afroasiatic phylum, the South Cushitic branch numbers 8 spoken (or formerly-spoken) languages, mainly spoken in Tanzania, but with Dahalo (dal) spoken in Kenya. The South Cushitic languages of our sample are: Alagwa (wbj); Burunge (bds); Gorwaa (gow); and Iraqw (irk). Internal classification of South Cushitic is difficult (Kießling and Mous 2003: 2–3): Qwadza (wka) and Aasax (aas) are dormant, with only small amounts of extant lexical data; Ma’a (mhd) is a mixed language (Mous 1994) with the Cushitic component limited to the roots of the ‘inner’ register only; and the status of Dahalo is unclear. With that said, within the largest branch of South Cushitic (named West Rift in Kießling and Mous 2003), there has been considerable success in representing genetic affiliation (Figure 2).

Figure 2:

Internal classification of West-Rift (adapted from Kießling and Mous 2003: 2).

Alagwa, also known as Uasi, is spoken in and around the Kondoa district by approximately 52,800 people (LOT 2009: 2). Neighbouring languages include Rangi, Datooga, Gogo [gog], and Sandawe. All Alagwa data used in this paper are from Mous (2016).

Burunge is spoken by approximately 23,000 people (LOT 2009: 2) in and around the Kondoa and Chemba districts of Dodoma, central Tanzania. Neighbouring languages include Rangi, Sandawe, Datooga, Gogo, and Maasai [mas]. All Burunge data used in this paper are from Kießling (1994).

Gorwaa, also known as Fiomi, is spoken by approximately 133,000 people in the lowlands surrounding Lake Babati (Harvey 2019a: 139–141), and has a high degree of mutual intelligibility with Iraqw. Contemporary contact with speakers of Iraqw and Mbugwe to the north, and speakers of Rangi and Alagwa to the south and east is strong and frequent. Evidence from barely two generations ago also shows close contact between speakers of Gorwaa and the Datooga varieties (for a more detailed treatment, see Harvey (2019a: 137–138)). Gorwaa data used here come from Harvey (2017).

Iraqw, also known as Mbulu, is spoken by approximately 602,600 people (LOT 2009: 2), primarily in the highlands of north-central Tanzania. Mutual intelligibility with Gorwaa is high. Neighbouring languages include Maasai, Datooga, Hadza, Gorwaa, Mbugwe, and Nyilamba. All Iraqw data used in this paper are from Mous (1993, 2007. In this paper, Iraqw and Gorwaa data are treated together.

Typologically speaking, the South Cushitic languages Iraqw, Gorwaa, and Alagwa exhibit predominantly SOV word order, with other orderings possible in both Gorwaa and Alagwa for pragmatic effects. Burunge is predominantly SVO, with similar flexibility as Gorwaa and Alagwa. The languages all employ a system of grammatical gender, which is conflated with number in a series of nominal suffixes, triggering agreement on dependents including adjectives and verbs. The languages are primarily head-marking, fusional, and employ a series of verbal extensions which alter the lexical semantics of the verb, and sometimes also its valency.

2.2.2 The Bantu languages

The Bantu languages, part of the larger Niger-Congo phylum, are widely spoken by people throughout central, eastern and southern Africa. The Bantu languages of our sample are Ihanzu [isn], Mbugwe [mgz], Nyaturu [rim], Nyilamba [nim], and Rangi [lag]. All of these languages form part of Guthrie’s F30 group (Maho 2009: 45),^[2] but when attempting a genetic grouping, the membership of Rangi and Mbugwe (though convincingly shown to be related to each other) is generally contested (c.f. Dunham 2005; Mous 2021; Stegen 2003). Save these two languages, below is a genetic tree placing the rest of the Bantu languages within a subgroup referred to as Takama (Nurse and Philippson 2003). Nodes labelled X represent unnamed predecessor languages of lower branches (Figure 3).

Figure 3:

The Takama genetic grouping (adapted from Masele 2001: 274).

Ihanzu, also known as Isanzu, is spoken by approximately 26,000 people (LOT 2009: 2) in Mkalama district in northern Singida region. Neighboring languages include Nyilamba, Hadza, Datooga, and Sukuma [suk]. All Ihanzu data in this paper are taken from Harvey (2019b). In this paper, the Ihanzu and Nyilamba data are treated together.

Mbugwe (also known as Buwe) is spoken by approximately 37,000 people (LOT 2009: 2) in Manyara region. Neighboring languages include Maasai, Iraqw, and Gorwaa. Mbugwe data in this paper are taken from Mous (2000) and Wilhelmsen (2014, 2018.

Rangi (also known as Langi) is spoken by approximately 371,000 people (LOT 2009: 2) in and around Kondoa district of Dodoma region. Two distinct but mutually intelligible varieties – Highland Rangi, and Lowland Rangi – are identified. Neighbouring languages include Alagwa, Gorwaa, Burunge, Sandawe, and Nyaturu. All data in this paper will draw from the Highland variety (Gibson 2012, 2019). In this paper, Rangi and Mbugwe data are treated together.

Nyaturu (also known as Rimi), is spoken by approximately 552,000 people (LOT 2009: 2) across a vast area of central and southern Singida region. Distinct, but mutually intelligible, varieties of Nyaturu include Munyiganyi, Ahi, and Rwana (Masele 2001). Neighbouring languages include Sandawe, Rangi, Nyilamba, Sukuma, Gogo, Nyamwezi, and Kimbu. Olson (1964) describes Rwana as the variety spoken by the most people, and, unless otherwise specified, it is on this variety that the grammar is based. In turn, it is this grammar from which all of our examples are drawn.

Nyilamba (also known as Iramba) is spoken by approximately 386,000 people (LOT 2009: 2) across much of northern Singida region. Ʉshoola is identified as a distinct but mutually intelligible variety of Nyilamba (Masele 2001). Neighbouring languages include Nyaturu, Ihanzu, Sukuma, and Nyamwezi. Johnson (1923) does not indicate on which variety his grammar notes are based, but it is from this work that all Nyilamba examples will be drawn.

As an aside, Kimbu (kiv) is another Bantu language featured in KMN (2008)’s sample, though at that time, the data on Kimbu were sparse. Unfortunately, that state of affairs remains, and as such, Kimbu is not discussed in the present paper.^[3]

The Bantu languages exhibit a broad degree of typological similarity. This is particularly true of the Eastern Bantu languages which have a dominant SVO word order which often allows some flexibility for pragmatic purposes. The languages employ a system of noun classes which behave like grammatical genders and trigger agreement across a range of dependents, including in both the verbal and nominal domain. The languages are primarily head-marking, agglutinative, and make extensive use of a system of verbal suffixes or ‘extensions’ which perform a range of functions, including in some instances being valency-altering (Figure 4).

Figure 4:

Southern Nilotic family tree (without sub-categorization of Kalenjin).

3.1 Iraqw-Gorwaa (South Cushitic)

The analyses for Gorwaa (Harvey 2018) and Iraqw (Mous 1993, 2005) differ, but the forms are virtually the same. As such, data for these languages will be treated together in this paper, with roughly as per Harvey (2018).^[4] Every finite clause in Gorwaa and Iraqw contains a preverbal clitic cluster^[5] (see (8)).

Phonologically, preverbal clitic clusters can bear no tone, which distinguishes them from lexical categories, but, syntactically, these forms are independent, with constituents including patient arguments, oblique arguments, and adverbs able to occur between the preverbal clitic cluster and the verb (see (9)).

Morphosyntactically, the preverbal clitic cluster is a series of clitics either procliticised or encliticised to a semantically null auxiliary. When the auxiliary has no phonologically-realised argument markers, it is realised as a. Schematically, the preverbal clitic cluster selector may be illustrated as seen in Table 1 (where elements within the same column are mutually exclusive of each other).

Gorwaa and Iraqw index all core arguments as proclitics to the auxiliary (see Table 2 below). Morphosyntactic alignment is split, depending on whether the argument is third person, or a speech-act participant (i.e. 1st or 2nd person). For third person arguments, alignment is (superficially) tripartite: the (S)ole argument of an intransitive clause, the (A)gent of a transitive clause, and the (P)atient of a transitive are all realised differently. For arguments which are speech-act participants, alignment is accusative: the (S)ole argument of an intransitive clause and the (A)gent of a transitive clause are marked in one way, and the (P)atient of a transitive clause are realised differently.

When a direct object intervenes between the preverbal clitic cluster and the lexical verb, it is no longer marked as an argument within the clitic cluster. This phenomenon is sometimes called encapsulation (Kießling 2007; Whiteley 1958), and fits Mithun’s (1984) description of Type III incorporation. As can be seen in (10) below, when the noun dó ’ Ngawdá ’ ‘the house of Ngawdá’ precedes the clitic cluster, it is marked as a (P)atient argument (u-). In a similar sentence, (10), a similar argument do’ oo Qutadu ‘the house of Qutadu’ occurs between the clitic cluster and the lexical verb. In this case, the encapsulated noun phrase is no longer marked on the clitic cluster, the clitic cluster instead showing marking as if it only had an (S) argument – that is, as if it were an intransitive verb.

Finally, if the object is postverbal in Gorwaa and Iraqw, it is still obligatorily marked in the clitic cluster, as in (11).

This pattern seems to differentiate Gorwaa-Iraqw from the other South Cushitic languages in the sample.

3.2 Alagwa (South Cushitic)

Finite clauses in Alagwa contain a preverbal clitic cluster (12), though clitic clusters which mark only the arguments may be omitted if the subject is phonologically overt (13) (Mous 2016: 173). As for Iraqw and Gorwaa above, the analysis of Alagwa clitic clusters in this paper is roughly as per Harvey (2018).

Phonologically, and different from Gorwaa and Iraqw, at least one of the enclitics of the Alagwa clitic cluster -- general past enclitic -áa -- is marked as having high tone. As with Gorwaa and Iraqw, these forms are syntactically independent, with constituents including direct and oblique object arguments, and adverbs able to occur between the preverbal clitic cluster and the verb, as in (14) below.

Morphosyntactically and similar to Gorwaa and Iraqw, the preverbal clitic cluster is a series of clitics either procliticised or encliticised to a semantically null auxiliary. Unlike Gorwaa and Iraqw, however, this auxiliary is always phonologically null.^[6] Schematically, the preverbal clitic cluster may be illustrated as shown in Table 3 (where elements within the same column are mutually exclusive of each other).

Alagwa indexes all core arguments as proclitics to the auxiliary (see Table 4). Morphosyntactic alignment is accusative, allowing a division between (S)ubject and (O)bject.

As with Gorwaa and Iraqw, if the direct object is encapsulated between the clitic complex and the verb, it is unmarked on the clitic complex, and the clitic complex behaves as if it is encoding an intransitive sentence (see (15)).

Alagwa has a freer word order than Iraqw: typically, new information is introduced following the lexical verb, and given information is introduced before the clitic complex (Mous 2016: 222). A postverbal object is given in (16) below. If the object is postverbal in Alagwa, it is not marked in the clitic complex.

3.3 Burunge (South Cushitic)

Finite clauses in Burunge contain a preverbal clitic cluster (17). However, unlike the other South Cushitic languages, the most common constituent-ordering in Burunge is Subject-Verb-Object, and in this configuration, marking the object in the preverbal clitic complex is optional (compare (18a) in which the object is unmarked, and (18b), in which the object is optionally marked). Analysis of Burunge clitic complexes in this paper is roughly as per Harvey (2018).

Morphosyntactically and similar to Gorwaa and Iraqw, the preverbal clitic cluster is a series of clitics either procliticised or encliticised to a semantically-null auxiliary. When the auxiliary has no phonologically-realised argument markers, it is realised as i.^[7] Schematically, the preverbal clitic cluster may be illustrated as shown in Table 5.^[8]

As Alagwa, Burunge indexes all core arguments as proclitics to the auxiliary (see Table 6). Morphosyntactic alignment is accusative, allowing a division between (S)ubject and (O)bject.

As with the other South Cushitic languages, if the direct object is encapsulated between the clitic complex and the verb, it is unmarked on the clitic complex, and the clitic complex behaves as if it is encoding an intransitive sentence (19).

3.4 Nyaturu (Bantu)

Preverbal clitics do not occur in every clause in Nyaturu (20), but are used to express certain tenses-aspects (sequential, persistive, near and far past, and near and far future) (21), as well as for subordinate clauses (22).

When these tense-aspects and clause types are expressed together, these preverbal clitics occur in what KMN (2008) identify as the clitic cluster for this language. This exemplified in (23) below, where subordination and far past are marked in the clitic cluster.

Phonologically, these preverbal clitic complexes can bear tone. Syntactically, these forms are independent, with constituents such as the subject of an intransitive verb (both unergative (24) and unaccusative (25) able to intervene between what has been described as the preverbal clitic cluster and the verb.

The Nyaturu preverbal clitic cluster can be analysed as analogous to a series of additional verbal ‘positions’ which occur before the initial part of the verb, traditionally construed. The structure of the verbal template, following Güldemann (1999: 546) is shown in (26) below.

Under the system found in Nyaturu the first element in the verb phrase is occupied by the relative marker nɨ́, which can also be considered as a proclitic (cf (3) vs. (6) above).^[9] The tense-indicating auxiliaries (e.g. the far past náa) may occur next, in a position which is most probably the pre-initial position of (26) immediately above. These so-called tense-indicating auxiliaries are used to encode sequential senses, connecting one clause to the next (cf. similar systems that are found in South Cushitic in example (14) above). In terms of linear ordering, the auxiliaries which appear after those encoding tense indicate aspect or clause-combining. And it is only after this that auxiliaries optionally marking the subject of the verb occur. But notably they use a different inflectional paradigm than that employed for subject marking on the main verb, reflecting a widely documented morphosyntactic difference between dependent and independent clauses in Bantu languages (Güldemann 1999). Schematically, the preverbal clitic complex can be illustrated as shown in Table 7.

Note that, of the (tense) auxiliaries, a future auxiliary may occur simultaneously with a past auxiliary. The result is an event which happened in the past, but subsequent to something else (27). This does not seem to be the case for slot 3 (aspect/clause-combining) auxiliaries.

The slot 2 (tense) auxiliaries may also occur without a main verb in copular constructions (28).^[10] The same pattern has not been described for the aspect/clause-combining auxiliaries.

3.5 Datooga (Southern Nilotic)

Verbs in all of the Datooga varieties include both preverbal and post-verbal morphology (Kießling 2007; Rottland 1982). Griscom (2019) identifies two sets of preverbal morphology in Asimjeeg Datooga associated with two distinct categories of verbal constructions: simplex-stem and complex-stem verbal constructions. Simplex-stem constructions maximally feature three prefixes (see Table 8 while complex-stem constructions may feature up to five (see Table 9). The first four slots of the corresponding preverbal morphology of complex-stem constructions in Gisamjanga Datooga are identified together by KMN (2008) as a preverbal clitic cluster. These include the 1) conditional, 2) polarity, 3) future tense, and 4) aspect slots. The morphemes in these slots are represented here as prefixes here (see Section 4.2).

The simplex- and complex-stem constructions feature slightly different subject indexation paradigms and are utilized in two mutually exclusive sets of tense-aspect-polarity constructions (see Table 10). Within each of the complex-stem constructions, at least one morpheme from the first four preverbal morphology slots must be present.

The difference between the simplex- and complex-stem subject indexation paradigms are exemplified in (29) and (30). In (29) the verb root ŋɛ́ːd ‘wake, start’ occurs in a non-future tense construction with simplex-stem morphology, including the affirmative prefix q- and the first person subject indexation marker áː-. In (30), by contrast, the verb root tʃàg ‘send’, occurs in a future tense construction with complex-stem morphology, including the affirmative prefix g-, the future prefix ì-, and the first person subject indexation marker dà-.

KMN (2008) analyze the future tense morpheme i(dʒ)(a)- as a clitic based on examples of two different constructions occurring in the Berger corpus (recorded by Paul Berger in 1935/1936). The first of these is analyzed as a future relative construction (see (31). For Asimjeeg Datooga (see (32), Griscom (2019) offers an alternative analysis as a distinct but etymologically related morpheme due to semantic similarities but morphosyntactic differences, such as the lack of a polarity marker or subject indexation on the future relative, which set the verbal future morpheme and the future relative apart.

The second construction KMN (2008) cites in support of the analysis of the future construction as a clitic is also a construction found in the Berger corpus, and one in which the future is separated from the lexical verb by the subject argument (see (33)). Unlike the relative construction, this separated future construction has not been found in any other Datooga data.

3.6 Ihanzu-Nyilamba (Bantu)

Ihanzu and Nyilamba show some significant differences (for lexical differences, see Masele 2001), but in terms of the preverbal clitic cluster, we feel the forms are sufficiently similar such that the data from these two languages may be treated together here. As with Nyaturu, preverbal clitics are not obligatory elements of every clause in Ihanzu or Nylamba (see the Ihanzu example (34)), but are used to express certain tenses (35), negation (36), as well as for subordinate clauses (37).

When these notions are expressed together, we may posit the resulting string as the preverbal clitic complex in these languages (38).

Phonologically, these preverbal clitic complexes can bear tone. Syntactically, and unlike Nyaturu, material cannot intervene between the clitic complexes and the lexical verb (compare (39)). With that said, the preverbal clitics in Ihanzu do show some degree of flexibility in terms of their relative ordering (compare examples (a) and (b) in (40)), and are therefore not as tightly integrated into the verb as, say, the subject prefixes.

Morphosyntactically, and within the Bantu convention, the preverbal clitic cluster in Ihanzu and Nyilamba can be analysed as a series of additional verbal ‘slots’ which occur before the initial part of the verb. Slot 1 is occupied by the relative marker, which is a clear example of a proclitic. Slot 2 is occupied by a marker of negation. Slot 3 is occupied by a tense-marking auxiliary. Schematically, the preverbal clitic complex may be illustrated as shown in Table 11.

3.7 Sandawe (Khoe-Kwadi or isolate)

In Sandawe, the person, gender, and number (PGN) of the subject is coded in realis clauses through verbal enclitics and/or preverbal enclitics that attach to other clause constituents (Eaton 2008). In realis negative and irrealis clauses, there are additional paradigms of verbal suffixes which are distinct from those used in the PGN clitics. In the case of the realis negative clauses, the realis PGN clitic may be present elsewhere in the same clause as a realis negative PGN verbal suffix.^[11] The three PGN clitic and suffix paradigms of Sandawe are listed in Table 12.

The PGN clitics can attach to non-subject clause constituents such as objects and adverbs in addition to or rather than the verb, as seen in (41) and (42). The choice of which constituent is marked with the clitic depends on the information structure of the clause (Eaton 2008: 127).

The subject NP or pronominal in a realis clause is optionally marked with a subject focus (SF) marker -áː. Generally, a verb without a PGN clitic cannot precede the first PGN clitic or subject-focus (SF) marker of a clause, and a verb with a PGN clitic cannot be preceded by another PGN clitic or SF marker in the same clause (Eaton 2008: 128). In (43), for example, the SF marker precedes a verb without a PC, and in (44) the verb ‘go’ occurs with a PC prior to all other instances of PCs.

Furthermore, it is possible that the SF marker and the PGN clitic paradigm may have developed from the same diachronic source, as there are formal similarities between the SF marker and the 3rd person PGN clitics, they share semantic and pragmatic properties, and their distribution is mutually exclusive (Eaton personal communication).

3.8 Hadza (isolate)

In all non-imperative verbal-predicate clauses in Hadza, the person, gender, and number (PGN) of the subject argument is obligatorily coded together with the tense and aspect of the clause. These PGN-TAM morphs occur in four sets of fusional paradigms, identified by Miller et al. (2016) as anterior (non-past or recent past tense), posterior (past or remote past tense), potential (some certainty and/or non-past tense), and veridical (less certainty and/or counterfactual). The segmental forms of one of the PGN paradigms, the anterior, is presented in Table 13. Other PGN paradigms are presented in Section 4.3 below.

Each paradigm may occur either as verbal enclitics as in (45), preverbal enclitics that attach to an auxiliary or adverb as in (46), or as a preverbal syntactically independent constituent as in (47). Enclitic forms exhibit varying levels of phonological dependence and reduction, such as vowel-copying (e.g. haka + ephee → haka:phee) and elision (e.g. kwa + heso → kweso).

Sands (2013c: 267) argues that the clitics most commonly attach to an auxiliary (46), and somewhat less frequently attach to the end of the verb (45). The clitics infrequently attach to adverbials or occur as independent constituents. Two clauses with different clitic patterns may be linked together, as in (47). Object indexation is always coded directly as a suffix on the verb, regardless of where or how subject indexation is coded. The Hadza auxiliaries, all of which may occur with PGN clitics, are listed in Table 14, as identified by Miller et al. (2016). The subjunctive occurs with its own paradigm, and for some auxiliaries there are one or more paradigms that have additional habitual forms.

Auxiliaries almost always precede the lexical verb of the same clause, and often directly precede it. Other syntactic units may occur between the auxiliary and the verb, however. In (48) the adverb kenena ‘early’ occurs between the auxiliary and the verb, in (49) the negative auxiliary akhwe occurs between the auxiliary and the verb, and in (50) the subject NP zzutchibisa sanzako ‘the North Wind’ occurs between the auxiliary and the verb.

There is some evidence indicating possible historical origins of the PGN-TAM clitics and auxiliaries. The PGN-TAM clitics can often be subdivided into distinct subject PGN and TAM components, with the PGN component preceding the TAM component in 1st and 2nd person, and the PGN component following the TAM component in 3rd person. This indicates the possibility that Hadza may have previously had Verb-Auxiliary word order, with the TAM morphs coming from verbal auxiliaries and two distinct post-verbal word order patterns for 1st/2nd person and 3rd person subjects. The origins of the synchronic auxiliary system are less clear, but there is evidence of possible connections to semantically and formally similar copula and conjunction constructions.

3.9 Rangi-Mbugwe (Bantu)

Despite their present-day geographical separation, Rangi and Mbugwe are considered sufficiently similar in their clitic clusters for the data from these two languages to be treated together in this paper. For discussion of the preverbal clitics in Rangi-Mbugwe, the relevant construction has been described as a complex auxiliary-based construction. These are used to encode a variety of TAM distinctions, with the specific tense-aspect combinations also showing variation between the two languages. Rangi and Mbugwe were not included in the original study of KMN (2008). We include them here due to the further work that has been done on the languages in the intervening years, as well as to provide a point of reference and comparison for the other Bantu languages included in the current paper. Furthermore, since one of the goals of the current paper is to consider the Rift Valley as a linguistic area, the inclusion of further languages contributes to a better understanding of the presence (or absence) of regional (i.e. areal) features.

In Rangi, the verb-auxiliary construction is used to encode the immediate future tense and the general future tense, formed using the auxiliaries =íise and =rɪ respectively. In Mbugwe, the construction is found in the present imperfective, habitual, future perfective, and the past progressive. While Rangi also exhibits auxiliary-verb order in certain tense-aspect combinations, in Mbugwe all auxiliary constructions exhibit verb-auxiliary order in declarative main clauses. A summary of the relevant structures is shown below in Tables 15 and 16, along with examples from each language (see (50)–(54)). We follow the Bantu tradition and mark the auxiliaries with a hyphen (rather than explicitly as a clitic using =). However, for the purposes of the current study, we consider these constructions as analogous to those found in the other languages under examination in that they attract subject and tense-aspect information and form a sort of ‘complex’, albeit one that behaves somewhat differently from those from the other language families reported here.

In both Rangi and Mbugwe, the PCC is obligatory but only in certain tenses and also under certain syntactic constraints. In Rangi, the verb-auxiliary construction is obligatorily found in the immediate and general future tense (Dunham 2005; Gibson 2013, 2019; Stegen 2002). In Mbugwe, this is found in the present imperfective, habitual, future perfective, past progressive (53), future imperfective (54) (Mous 2000; Wilhelmsen 2014). For the current purposes we consider this to be obligatory since an attempt at a different word or omission of the cluster results in ungrammaticality. In terms of phonological characteristics, these comply with the broader phonotactics of the language. In other words, these units can be tone-bearing and in the case of the past-tense construction in Mbugwe for example, the auxiliary hosts the past tense prefix á- (cf. Wilhelmsen 2014: 3).

4.1 Nyaturu-Proto-West Rift

In KMN (2008), the preverbal clitic cluster in Nyaturu is described as arising from sustained contact with Proto-West Rift, the predecessor language to the current South Cushitic languages. Of the Nyaturu preverbal clitic cluster, it is written that “[i]t looks as if Bantu material has been used to build a system of preverbal clitics, encoding Bantu categories in a Southern Cushitic frame” (page 199). The possibility of the Nyaturu preverbal clitic cluster developing through entirely internal grammaticalisation processes (and therefore resembling the South Cushitic forms by chance) is ruled out because it cannot be traced to any auxiliary structures in nearby related languages, such as Sukuma and Nyamwezi.

What follows attempts to provide a bit more precision on KMN (2008)’s claim. Each morpheme of the Nyaturu preverbal clitic cluster will be considered, along with a plausible source (or in some cases, two competing sources). The Nyaturu subordinate clause marker nɨ́= (55) has two possible origins. The first is that it was borrowed into Nyaturu from Proto-West Rift or one of its predecessor languages – both Gorwaa and Iraqw use a preverbal clitic ni= to indicate a relative clause lacking an internal patient whose object is 1st person singular.

This morpheme appears to be of good Cushitic origin, in that it appears with the same function in the geographically distant, but genetically-related Lowland East Cushitic language Harar Oromo (orm) as -`n (Owens 1985).

Alternatively, and perhaps more plausibly, the Nyaturu subordinate clause marker may have its origins in a Bantu focus-marker, commonly of the form ni (cf. Schwarz 2003; Gibson 2019).

The near past tense clitic ájà has many analogues in Bantu languages (c.f. -zà for perfective past disjoint in Sambaa (ksb) (Riedel 2009: 29), for example).

The Nyaturu sequential morpheme qàá is quite possibly Bantu in origin as well. Masele indicates that the [ɣ] sound may have developed from a [g] sound (2001: 122) in Nyaturu. Though we do not have a morpheme gàá in any nearby language, Batibo (1985) identifies the Sukuma morpheme -ka- as “circumstantial” (page 259) and a “non-definitive or unaccomplished” (page 263) morpheme. Nurse (2000: 523) points out the ka of subsequent action, widespread in Bantu (but whose connection to the Nyaturu qàá would require some further phonological explanation (Nurse 2000: 533, ff11)).

The Nyaturu persistive morpheme kɨ̀ɨ (56) also seems uncontroversially Bantu in origin, compare the Ihanzu auxiliary verb kɨlɨ, also used to convey a persistive meaning. Nurse (2000: 523) also draws attention to the widespread Bantu persistive aspect ki, but again notes that to connect this ki to the Nyaturu kɨ̀ɨ would require some further phonological explanation (Nurse 2000: 533, ff11).

The Nyaturu far past tense morpheme náa appears to be a good candidate for borrowing from Proto West Rift, or one of its predecessor languages (c.f. Iraqw/Gorwaa “perfective” =(g)a, Alagwa “general past” =áa, and Burunge “preterite” =áa). For an alternative source, many Bantu languages employ a verb-internal morpheme a- to express past notions. For example, Sukuma employs a morpheme a- for “accomplished and immediative” (Batibo 1985: 263).

The Nyaturu near future tense morpheme nàa has also been proposed as a borrowing from Proto West Rift (c.f. Burunge “future1” =aa). With that said, because this form is only employed in Burunge, the direction of the transfer is still unclear (Roland Kießling, p.c. 08.09.2021).

The Nyaturu far future tense morpheme ìkwɨ́ was identified in KMN (2008: 199) as having no obvious origin either in Bantu nor in West Rift. We would submit that a possible Bantu source for this morpheme is the lexical verb “to stand” (Nyaturu ɣw-ɨmɨ́ka c.f. Jinakɨ̀ɨ̀ya Sukuma gwɨ̀ɨ̀ma (Masele 2001: 594) and Rangi kwɨ̀ɨ̀mà (Masele 2001: 709)). With that said, an anonymous reviewer points out that phonological form, as well as the semantics of “stand” or “stop” would not relate to future in any straightforward way, and overall, a closer analysis would need to be done to convincingly establish this link.

4.2 Datooga-PWR

KMN (2008) propose that speakers shifting from West-Rift to Pre-Datooga influenced the grammaticalization of Pre-Datooga constructions by incorporating Datooga morphology into West-Rift-like structures. KMN (2008) specifically mention three Datooga constructions, the future, persistive, and sequential, which resemble the South Cushitic clitic clusters in that they appear to be separable from lexical verbs to which they attach.

According to Datooga oral history (Wilson 1952), contact between speakers of Datooga and Iraqw is believed to have occurred in the TRV for at least the past 300 years, but little is known about contact between South Cushitic and Datooga groups further back in time. There is linguistic evidence of earlier periods of contact between West-Rift and Southern Nilotic speakers (Kießling 2002), which most likely occurred in areas north of the TRV. If the patterns exhibited by the Datooga future, persistive, and sequential constructions are due to contact which occurred within the TRV and did not involve speakers of other Southern Nilotic languages, then we would expect to see patterns which are not found in other Southern Nilotic languages.

There is evidence from the Kalenjin languages of the Southern Nilotic language which indicate that the persistive and sequential constructions have either been inherited or have developed independently of contact with South Cushitic languages. This evidence comes in the form of cognate constructions (see Table 18), which resemble the Datooga constructions in both form and function (Creider and Creider 2001; Griscom 2019; König et al. 2015; Mietzner 2016).

The Kalenjin constructions which are cognate with the Datooga persistive construction are also used to code persistive aspect (often translated as ‘still’). The persistive construction in Nandi (niq) and Cherang’any (enb) are analyzed as verbal prefixes, whereas the Akie (oki) persistive construction is analyzed as an auxiliary. The Nandi and Akie constructions which are cognate with the Datooga sequential constructions are analyzed as coordinating conjunctions, and the cognate construction in Cherang’any is analyzed as a verbal enclitic. The similarities between these constructions indicates that the Datooga persistive and sequential constructions were inherited from Proto-Southern Nilotic.

The diachronic source of the Datooga future construction is unknown (Roland Kießling p.c.). There are multiple cognate future constructions in the Kalenjin languages which can be linked to the verb mac(-ey) ‘want’ (e.g. Nandi mâ- and Akie mach-), and, although the verb gas ‘want’ in Datooga shares similar semantics with these Kalenjin verbs which serve as the source for the Kalenjin future constructions, it is not the source of the Datooga future construction. The development of a future tense construction from a verb meaning ‘want’ has been documented in many languages before (Heine and Kuteva 2005: 18), but there is no clear evidence linking the Datooga future construction specifically to a verb meaning ‘want’.

There is, however, internal evidence indicating that the Datooga future and persistive constructions developed from multi-verbal constructions, in a parallel fashion to the future and persistive constructions of the Kalenjin languages. Kießling (2019) suggests that these Datooga constructions developed out of multi-verbal clauses and were simply inherited rather than borrowed or created as a result of contact.

One key feature of Datooga verbs which supports the putative verbal origins of the complex-stem constructions is the affirmative prefix g- ∼ q-. Within the subject indexation paradigm of complex-stem constructions, third-person subject indexation mandatorily co-occurs with a form that is identical to the affirmative prefix, as shown in (57). This form has been analyzed as either an instance of the affirmative prefix (Griscom 2019), as indicated in (57), or as part of the subject indexation prefix (Kießling 2000). Anderson (2011: 173) writes that the Datooga future construction may have developed from the fusion of a multi-verbal, doubly-inflected construction.

The development of the doubling pattern could be described as having occurred in three stages (see Table 19). In the first stage, the two fully inflected verbs co-occur in a multi-verbal bi-clausal construction. In the second stage, the first verb grammaticalizes and in the process loses the capacity to co-occur with standard inflectional suffixes and a full subject indexation paradigm. The diachronic development of the second verb is less transparent, as it is tied to the wider distribution of the dependent stem structure in imperative and subordinate clause constructions, but it is clear that the second verb loses its capacity to co-occur with the affirmative prefix for non-3rd person subjects. In the third and final stage, the first verb becomes syntactically bound to the second verb and constitutes a cluster of preverbal morphology.

Kießling (2019) proposes a potential historical source for complex-stem constructions in Gisamjanga Datooga that consists of a verbal auxiliary followed by a lexical verb, corresponding to Stage II of the diachronic pathway. The co-occurrence of future + bound auxiliaries could be due to the merger of multiple verbs through the same pathway (i.e. FUT + AUX + VERB -> FUT=AUX=VERB), and some examples from the 1930s Berger corpus indicate that the future may have been more syntactically independent in the past (Kießling 2019). There is also evidence from Asimjeeg Datooga that the persistive may function independently as a lexical verb (58).

4.3 Sandawe (coincidence)

KMN (2008) propose that any similarities between the Sandawe PGN clitics and the clitic constructions of the other TRV languages under discussion are most likely due to chance, and that the Sandawe PGN clitics are more than likely a retention, with similar structures found in languages of the Khoe-Kwadi family (with which Sandawe may have a distant genetic relationship). We find this claim to be supported by a number of pieces of evidence, some of which mentioned in KMN (2008), and some of which we will explore for the first time here.

KMN (2008) establishes three principal ways in which the Sandawe PGN clitics resemble preverbal clitic clusters of neighbouring languages: 1) Sandawe PGN clitics are non-verbal elements which mark for the subject, similar to preverbal clitic complexes in Southern Cushitic and Nyaturu; 2) “cliticization is to the left” (page 205); and 3) the Sandawe PGN clitic has a focus function, similar to preverbal clitic complexes in, for example, Burunge. Points (1) and (3) above are accurate, but the observation made in point (2) must surely be a mistake: Sandawe clitics are enclitic in nature, thus rendering these PCCs different to those found in South Cushitic and Ihanzu, Nyaturu, Nyilamba, but similar to those found in Hadza and (probably) Rangi-Mbugwe.

In terms of how the Sandawe PGN clitic differs from the preverbal clitic clusters of neighbouring languages, KMN (2008) notes that: 1) Sandawe PGN clitics do not mark tense-aspect, and 2) Sandawe PGN clitics do not have a fixed position. To this, we would add that 3) there exist no morphemes in the Sandawe PGN clitic paradigm which can be readily linked to any other language in the TRV, and that 4) Sandawe PGN clitics are monomorphemic, and are therefore not clusters of clitics at all.

Finally, it is worth revisiting a detail in KMN (2008: 205) noting that the Sandawe PGN clitics are similar in many respects to the PGN markers in the Khoe languages of southwest Africa. Syntactically, and as mentioned above, the PGN clitics of Sandawe are either enclitics to verbs (as in (42) and (44) above), or preverbal clitics encliticising to other constituents in the clause (as in (41) and (43) above). In the Khoe-Kwadi (and depending on the language), the PGN markers appear as nominal suffixes or as clitics on noun phrases, among other constituents (Witzlack-Makarevich and Nakagawa 2019: 397–398). In (59), an example from Namibian Khoekhoe (naq), the PGN marker =b, itself coding subject, encliticises to both the subject noun, the boy but also to the object noun the girl, which has its own PGN marker -s.

Formally, KMN (2008: 205) note that the PGN markers “seem cognate” with those of the Khoe-Kwadi languages, and though there do not seem to be many resemblances between the PGN clitics of Sandawe and those reconstructed for Proto-Khoe (compare columns 2 and 3 in Table 20), Sandawe PGN clitics and personal pronouns certainly seem cognate (compare columns 1 and 2 in Table 20), a pattern which also obtains for the languages of Khoe-Kwadi.

As such, and with some further proof, it seems reasonable to assume that the PGN markers of Sandawe are not a result of areal contact, and that most (if not all) resemblance is due to chance. Whether the construction represents a retention specifically from Khoe-Kwadi, there is some evidence both syntactically and in terms of a similar grammaticalisation pathway to support this, though cognacy via any phonetically identifiable segments is not, at least on the surface, evident.

4.4 Hadza (unknown)

At the time during which KMN (2008) was written, the documentary and descriptive state of Hadza sufficed only for the article to confirm that Hadza featured a preverbal clitic cluster, but nothing more could be said about its historical development. Advances in Hadza studies now allow us to make some observations. As mentioned above, the Hadza preverbal clitic complex is formed of a PGN-TAM marking element, as well as, in some cases, a set of auxiliaries expressing polarity, mood, and clause combining functions such as subordination and sequentiality. Our first observation deals with the auxiliaries, and the rest treat the PGN-TAM marking element.

Of the auxiliaries, the two marking sequentiality, (h)a and ka, are both candidates for cognacy with the “-ka type” sequentials mentioned elsewhere, such as the qàá sequential of Nyaturu (or even the ka- verbal morpheme of Sukuma) both described in Section 4.1 above.

Of the subject PGN-TAM marking element, there are several observations to make: the first will treat the TAM component, and the latter will treat the PGN component. Of the TAM component, the =aa enclitic for posterior tense (see Table 22) seems close enough to the West Rift =aa past enclitic to be worthy of note. With that said, it ought to be noted in Hadza that this form is phonetically [aʔa], whereas the West Rift form is phonetically [a:].

PGN-TAM marking in the veridical (expressing less certainty or counterfactual, see Table 21 below) is striking in that it is characterised by the morpheme ikwi.

Here there exist two possible explanations. The first suggests contact with Bantu (c.f. the form ìkwɨ́ identified in Nyaturu above, with an origin in the lexical verb “to stand”). The second suggests an internal origin: Hadza features a verb ikha defined in Miller et al. (2016) as “to stop (doing something)”, or “to stand or to be standing”. In fact, the meaning of the Hadza form is so close to the Bantu form, the Hadza verb itself may be a borrowing from Bantu. Whatever the case, the ikwi -future connection seems to be a common pattern across the languages in our sample.

Of the PGN component, the pronominal elements are almost always marked by /n/ in the 1st person, /t/ in the 2nd person, and /s/ in the 3rd person (see Table 22). A clear Hadza-internal origin for these person marking consonants is not entirely clear. The personal pronouns of Hadza feature /n/ in the 1st person (e.g. ono ‘I’) and /t/ in the 2nd person (e.g. te ‘you’), but do not mark 3rd person with /s/. With this in mind, a Hadza-external origin for these forms could be entertained. In fact, this /n-t-s/ paradigm is strikingly similar to widespread patterns observed in the languages of the Afroasiatic phylum (c.f. Tucker 1967: 22). When looking for parallels within West Rift (Cushitic) – indeed the most likely group of languages to serve as the donor for Afroasiatic pronominal material to Hadza – the evidence for this /n-t-s/ paradigm is not so convincing if we examine only the independent pronouns (Kießling and Mous (2003) reconstruct these as *ʔana ‘I’, *kii ‘you (f.sg.)”, *kuu ‘you (m.sg.), and *ina ‘he, she’, with the 2nd person form lacking a /t/ morpheme, and the 3rd person form lacking an /s/ morpheme). Roland Kießling (p.c. 21/09/2022) notes that more promising evidence of the /n-t-s/ paradigm in West Rift can be found in the preverbal clitics marking subordinate clauses in Gorwaa and Iraqw, where the forms are ni ‘that I’, and ta ‘that you’ (see esp. Mous 1993: 125). Further, Kießling (2002: 358–360) reconstructs Proto-West Rift pronominal roots *ʔani (1.Sg) and *ʔata (2Sg) as the sources of these preverbal subordinate markers, which may themselves link to the Proto-East Cushitic pronominal roots *ani (1.Sg) and *ata (2Sg), reconstructed in Sasse (1981: 144). Kießling (p.c. 21/09/2022) further points out the Proto-West Rift 3rd person possessive pronominal suffix *-s.

Syntactically, it is worth noting that Hadza PGN-TAM marking clitics can not only occur before the verb (as enclitics to preverbal auxiliaries), but also occur as enclitics to the lexical verb. In this way, they are similar to the PNG-markers of Sandawe in that they are enclitic in nature.

Further, and as mentioned above, the subject marking of Hadza PGN enclitics can more-or-less be seen as deriving from pronominal material (in this case, material which seems Afroasiatic in origin). In this way too they are then similar to Sandawe PGN-TAM clitics, in that these are also historically linked to pronouns.

4.5 Rangi-Mubgwe

The compound auxiliary constructions found in Rangi and Mbugwe which we consider here to be part of the larger PCC constructions under examination in the present paper were not addressed by KMN (2008). The analysis developed here adopts however the same categories used in this section so far. There are no formal similarities between the Rangi-Mbugwe auxiliary forms and no reason to think that this represents a borrowing of form. However, there are functional similarities in that these constructions are associated with the encoding of specific tense-aspect-mood distinctions and also exhibit a sensitivity to clause type. The relevant socio-historical context involves a sustained history of contact between Rangi-Mbugwe speakers and Cushitic-speaking communities over an extended period of time. Well-established pathways of grammaticalisation can be identified which sees the development of auxiliary forms from (lexical) verbs, a process which is widespread across Bantu and cross-linguistically. The combination of auxiliary forms and main verbs to encode a wide range of tense-aspect distinctions has also been described across the language family.

5.1 Assessing the pre-verbal clitic cluster as an areal feature

In KMN (2008), positing the pre-verbal clitic cluster as one of the 19 areal features makes two major underlying assumptions. The first is that the preverbal clitic clusters in the languages of the sample could be compared (which we will comment on in Section 5.1.1 below). The second was that so doing would help develop insights into i) the nature and mechanics of language contact (which we will comment on in Section 5.1.2 below), ii) the character of linguistic areas as a linguistic phenomenon, and iii) the history of the encounters between the peoples of the area (both of which we will address in Section 5.2 below).

5.2 Whither the Tanzanian Rift Valley (area)?

KMN (2008)’s presentation of preverbal clitic clusters in the languages of the Tanzanian Rift Valley was a subsection of a chapter with a larger goal: that is, the establishment of a ‘Tanzanian Rift Valley Area’ – the languages within which having come to resemble each other in at least 19 ways, due primarily to a situation of long-term, dynamic language contact. Subsection 5.2.1 provides an alternate approach, focusing on the history of the individual encounters between the peoples of the area. Subsection 5.2.2 reflects on what the analysis of preverbal clitic clusters put forth in this paper means in terms of the purported Tanzanian Rift Valley linguistic area and represents the conclusion of the paper.