Middle voice in Bantu: in- and detransitivizing morphology in Kagulu

1 Introduction

In this paper, we assess the middle voice system (MVS) of the Bantu language Kagulu (ISO 639-3 code: kki, Guthrie code: G12), spoken in the Morogoro region of Tanzania. Our analysis of Kagulu’s MVS focuses on eight verbal affixes, presented in (1).

Because verbs with these affixes generally have reduced transitivity patterns, we refer to them as in- or detransitivizing. We argue that, according to the middle voice definition provided below, Kagulu’s MVS includes only two of the affixes in (1), namely -ik and ki-.

We follow a specific comparative-typological tradition in using the term middle voice system to refer to verbal coding strategies that (i) are used with (at least) bivalent verbs to encode minimally two of the following voices (in alphabetical order): anticausative, antipassive, autobenefactive, conversive, passive, reciprocal, reflexive; and (ii) occur with a set of verb forms that do not exist without one of the MVS coding strategies (see Bahrt 2021; Grestenberger 2014; Inglese 2020, 2022; Kulikov 2011, 2013; Shibatani 2004; Zúñiga and Kittilä 2019). We use the term voice to mean “the way a specific diathesis is formally marked on functional or lexical verbs” (Zúñiga and Kittilä 2019: 4), with diathesis being “any specific mapping of semantic roles […] onto grammatical roles […]” (Zúñiga and Kittilä 2019: 4).^[2] The coding strategies of a MVS are called middle marking. Middle verbs are verbal predicates that have middle marking, and non-middle verbs are verbs without middle marking. Adopting terminology from Grestenberger (2014) and Inglese (2020), we use the term oppositional middle verb to refer to a middle verb that alternates with a non-middle verb and encodes one of the voices listed above, as in (2) below, and non-oppositional middle verb to refer to a middle verb that lacks a non-middle counterpart, as in (3).

The definition of MVS adopted in this paper differs from other comparative-typological work, notably that of Kemmer (1993), likely the most well-known cross-linguistic account of middle voice. Kemmer (1993) defines a semantic middle as:

a coherent but relatively diffuse category that comprises a set of loosely linked semantic sub-domains centering roughly around the direct reflexive. Although lacking precise boundaries, the area of middle semantics can be delimited with reference to two semantic properties. These two properties, which are characteristic of every middle system during the period in which it has a functioning middle marker are 1) Initiator as affected entity (Endpoint) and 2) low degree of elaboration of events. (Kemmer 1993: 238)

The three central differences between Kemmer’s conceptualization of middle voice and that adopted in this paper are that (i) Kemmer distinguishes middle voice from other voice categories such as the reflexive, reciprocal or passive, while in our approach the co-expression of two or more of these voice categories by the same verbal coding device is an essential feature of a MVS (see also Bahrt 2021); (ii) Kemmer’s middle voice category consists exclusively of a set of semantic classes of situations (middle situation types) which are recurrently expressed by middle verbs cross-linguistically, while our definition incorporates a broader set of phenomena; and (iii) our definition does not attempt to capture a unifying semantic feature underlying the multiple heterogeneous (types of) meanings covered by a MVS in a language, in contrast to Kemmer’s “low degree of elaboration of events”.

This paper is organized as follows: We sketch an overview of the history of research on middle voice in Bantu linguistics in Section 2; we describe the main aspects of Kagulu verbal morphology and elaborate briefly on the data collection practices in Section 3; the analysis of Kagulu’s MVS is presented in Section 4, focusing first on those in- and detransitivizing affixes that we consider as middle markers in Section 4.1, followed by an assessment of the affixes that we do not analyze as middle markers (Section 4.2), with a summary in Section 4.3. We conclude the paper in Section 5.

2 Middle voice in Bantu linguistics

The category ‘middle (voice/diathesis)’ has its origins in the study of older Indo-European languages such as Vedic, Sanskrit and Ancient Greek (Kulikov 2011: 368, 2013: 261; Zúñiga and Kittilä 2019: 168). In these studies, middle refers to a specific inflectional form or paradigm of the verb contrasting with an active inflectional form or paradigm. Table 1 illustrates the well-known case of Ancient Greek active versus middle verb forms.

The formal contrast between active and middle forms of a verb lies in the different inflectional suffixes, bolded in Table 1. Middle verb forms in those older Indo-European languages are typically attested in a wide range of different grammatical contexts such as reflexive, autobenefactive, reciprocal, passive and anticausative constructions (see, e.g., Grestenberger 2014: 20–22; Inglese 2020; Kulikov 2013: 273; Zúñiga and Kittilä 2019: 168–171). In addition, some verbs do not alternate between active and middle paradigms, but instead only appear in one or the other. Those verbs that only take the active inflections are called activa tantum (Latin for ‘active only’) and those that only appear in the middle form are called media tantum (Latin for ‘middle only’; for these terms, see also Delbrück 1897; Grestenberger 2014; Kulikov 2013: 275).

Indo-Europeanists at the beginning of the 20th century attempted to formulate an underlying semantic feature that could account for the eclectic usage of middle verb forms (for a detailed overview, see Gonda 1961: 30–43). For example, Meillet (1908: 213) states that “les désinences moyennes indiquent que le sujet est intéressé d’une manière personelle à l’action” [‘the middle inflections indicate that the subject is personally interested in the action’; translation added]. Drawing from this tradition of classical Indo-European linguistics, some older grammars of Bantu languages, written by European missionaries during the colonial era, mention the category of middle voice. For example, in his description of the Kikongo variety spoken in Mbanza Kongo in the late-19th century, Bentley (1887: 621–627) writes:

There is a third voice to nearly all Kongo verbs, which is neither active, transitive nor passive, but between the two, since it conveys the idea of action without the need of an object to complete the idea; as:

O nlele ubakuka: The cloth tears.

At the same time, it expresses the idea of an active condition, or state, which is attributed to the subject itself, and is not regarded as being suffered or caused by anything exterior to the subject. It is therefore neither active transitive nor passive; but possessing an idea half way between the two, has been called by grammarians the Middle voice. A verb in that voice, or of that nature or form, is Active Intransitive. [bold in original]

Dammann (1957: 77) describes a middle marker -ar in Kwangali (ISO 639-3 code: kwn, Guthrie code: K33) with the following semantic definition: “Die ursprüngliche Bedeutung dieser Spezies ist, daß die Handlungen oder Vorgänge im Interesse des Subjektes liegen” [‘the original meaning of these types is that the actions or processes are in the interests of the subject’; translation added]. The examples provided in the grammar are reproduced in (4).

However, in general, middle voice has not received much attention in the descriptive literature on Bantu languages. This absence could well be due to the typological difference in grammatical structure between Bantu languages and those Indo-European languages in which the category was first identified. For example, the specific type of inflectional active/middle voice system found in Ancient Greek has no equivalent in Bantu languages. Furthermore, the reflexive, reciprocal, passive and anticausative voices for which middle verb forms were used in older Indo-European languages are more typically expressed in Bantu languages by verb forms with specialized derivational morphology. This is illustrated with examples from Chewa (ISO 639-3 code: nya, Guthrie code: N31b) in (5)–(8).

As with many language-particular grammatical categories, middle (voice/diathesis) has been integrated into the typological-comparative linguistics tradition as a comparative concept (among others, see Inglese 2022; Kazenin 2001; Kemmer 1993; Klaiman 1991; Kulikov 2013; Zúñiga and Kittilä 2019). Still, middle voice as defined in frameworks within this tradition has received little attention in Bantu linguistics (see, e.g., Bernander 2018; Bostoen and Nzang-Bi 2010; Chavula 2018; Creissels 2002: 399–404; Guérois and Bostoen 2018; Jerro 2018; Ngwasi 2021).

The first general discussion of middle voice in Bantu from a typologically-informed perspective is presented by Dom et al. (2016). The authors assess the middle status of the following five verbal affixes reconstructed to the most recent common ancestor of the Bantu languages, Proto-Bantu (see Schadeberg and Bostoen 2019: 173): *-am, *-an, *-ɪk, *ji- and *-ʊk.^[6] The analysis by Dom et al. (2016: 129–130) draws heavily on Kemmer’s (1993) semantic middle situation types. They illustrate how reflexes for each of the five reconstructed affixes in various individual Bantu languages conform to their definition of middle voice. For example, the suffix -am in Mongo (Hulstaert 1965: 247–250) occurs with verbs that denote middle situation types, such as kukam ‘become attached or stuck against something’. The lexical meaning corresponds to Kemmer’s ‘positional’ and ‘spontaneous’ middle situation types.^[7] In addition, -am is used to derive passive verb forms in Mongo, e.g. tómbam ‘be carried’ from tómb ‘carry’ (Dom et al. 2016: 137).

Dom et al. observe that in many Bantu languages these affixes are limited in their (co-)expression of voice alternations. They thus analyze these affixes as ‘quasi-middles’, distinguishing them from ‘canonical’ middle markers in other languages that encode the anticausative, antipassive, passive, reciprocal and reflexive voices. These five voices are taken by Dom et al. as the ‘functional core’ that make up the ‘canonical middle’. In Kagulu, as in the Bantu languages examined by Dom et al., reflexes of these Proto-Bantu affixes are limited in their (co-)expression of voice alternations. Unlike Dom et al. (2016), in this paper we do not distinguish between a canonical and quasi-middle, but instead consider any single coding strategy used for the co-expression of at least two detransitivizing voices to belong to the MVS of that particular language. This is in line with cross-linguistic observations that there are very few languages that have a single coding strategy co-expressing four or five voice alternations involving the anticausative, antipassive, passive, reciprocal and reflexive. In a typological sample of 222 languages, Bahrt (2021: 158–159) describes how patterns of two or three of these voices co-expressed by a single marker are much more frequent. A similar tendency has been observed by Inglese (2022) in a typological sample of 129 middle-marking languages.

3 Kagulu

Kagulu is a Bantu language spoken in the Morogoro region of Tanzania. The language is under-described: there is one older grammar (Last 1886) and a grammatical description for present-day Kagulu (Petzell 2008). Like all Bantu languages, Kagulu has a noun class system (for an overview of the nominal morphology of Kagulu, see Petzell 2008: 47–72). Unlike many Bantu languages, Kagulu is not tonal (Petzell 2008: 40) and has a reduced set of tense and aspect constructions (Bar-el and Petzell 2021: 534).

In this section, we briefly discuss the morphological structure of Kagulu verb stems, an understanding of which is essential for the analysis of the affixes under investigation in this study. As is the case in all Bantu languages, the core building block of a verb form is the root. Kagulu verb roots can have various phonological shapes, given in (9).

Not every root exists independently. For example, the verb stem lakal ‘become burned’ can be formally analyzed as consisting of a CVC root lak and a suffix -al (see also Section 4.2.5), but the root lak by itself is not attested in Kagulu.

There are 11 functionally distinct morpheme slots other than the root (see Petzell 2008: 98–99), including those for subject and object indices, relativizers, tense-aspect-mood-polarity affixes, and derivational affixes influencing the lexical meaning, aspectual semantics or valency of the stem. This study focuses on derivational affixes. These occur immediately before and after the verb root. There are only two derivational prefixes in Kagulu, namely i- and ki-. All other derivational morphemes are suffixes, most of which have the phonological shape -V(N)C, as illustrated in (10).

Multiple derivational affixes can co-occur on a single verb root, e.g. ki- and -an in kifan ‘resemble’.

Derivational affixes in Kagulu vary in productivity. We define productivity as “the ability of word-forming elements to be used to form new linguistic expressions” (Brinton and Traugott 2005: 92).^[8] Of the eight affixes that we examine in this study, only three are used productively, namely -igw, -ik and ki-. The remaining five, i.e., -al, -am, -an, i- and -uk, are unproductive.

The data for this study stem from three sources: (i) Petzell’s (2008) grammatical description and word list, (ii) a corpus of five stories (Petzell 2008, 2015, and (iii) fieldwork data collected in 2020 and 2021. We compiled a database of all verbs with any one of the eight affixes examined for this study, drawn from the grammar, stories and fieldwork data. An overview of the number of attestations of each affix is given in Table 2.

For each verb, we queried the published sources on Kagulu for roots or semantically related derived verb forms in correspondence with the coded in- or detranitivized verbs of the database. We re-checked all verb forms and filled gaps in the data with a language consultant. To obtain the etymology of the derived verbs or their corresponding roots, we linked lexical reconstructions from the online ‘Bantu lexical reconstructions’ (BLR) database (Bastin et al. 2002) to the Kagulu verb forms. In total, we found lexical reconstructions with varying time depths for 48 of the 68 verbs in the database.

4 Kagulu in- and detransitivizing affixes

In this section, we analyze the middle marking status of eight affixes in Kagulu that attach to verb roots forming stems with reduced transitivity patterns. Our assessment is based on the two criteria for MVS that we adopt in this paper: (i) co-expression of two or more detransitivizing voices, and (ii) the attestation of non-oppositional verbs. Out of the eight affixes, we analyze only two as middle markers: -ik and ki- (Section 4.1). We argue that the remaining six affixes, namely -al, -am, -an, i-, -igw and -uk, do not qualify as middle markers (Section 4.2).

5 On the role of middle voice systems in Bantu languages

The middle voice system, as adopted in this paper, is defined as a cluster of non-oppositional and oppositional verb forms, the latter of which exhibit voice syncretism, that is, marked by a single coding strategy that expresses at least two of the following voice alternations: anticausative, antipassive, autobenefactive, conversive, passive, reciprocal, and reflexive (Kulikov 2013). The range of voice alternations encoded by individual middle markers varies from language to language. Some languages have a well-developed middle marker whose voice syncretism covers up most or all of the detransitivizing voice categories, such as the suffix -śk in Udmurt (ISO 639-3: udm; see Bahrt 2021: 263). Kagulu is not such a language. It does not have one dedicated middle marker, but rather two affixes that can be analyzed as middle markers, each encoding only two voice alternations.

The underdeveloped status of middle markers in Kagulu has implications for our understanding of MVSs in Bantu languages. We suggest that middle voice is not a salient grammatical category in most Bantu languages, if any at all. In Chewa, those voices that are typically subsumed under the middle voice cluster are all encoded by different verbal affixes (see Section 2). This situation goes back to Proto-Bantu, as different verbal affixes are reconstructed to the ancestral language for almost every voice that is part of the middle system as proposed in Kulikov (2013): *-(ib)ʊ for passive, *i- for reflexive, *-an for reciprocal, *-ik and *-ʊk for anticausative (Polak 1983; Schadeberg and Bostoen 2019). Middle markers in a particular Bantu language thus develop through changes in a language’s derivational system and functional innovations of one or more derivational affixes.

In one such diachronic situation, the functional range of a given affix gradually expands to include the coding of voices previously encoded by other affixes. This type of development has been described for the Kagulu prefix ki- in Section 4.1.2, which developed a reflexive-reciprocal syncretism resulting in the suffix -an, a reflex of the Proto-Bantu reciprocal marker *-an, becoming unproductive. Other examples of this type of development in Bantu are the reflexes of Proto-Bantu *-ɪk in Tonga (ISO 639-3 code: toi; Guthrie code: N15), Tumbuka (ISO 639-3 code: tum; Guthrie code: N21) and Manda (ISO 639-3 code: mgs; Guthrie code: N11) (see Bernander 2018; Chavula 2018). In these languages, the coding of passive voice has shifted from reflexes of Proto-Bantu *-ibʊ to the reflexes of the neuter *-ɪk, with the former suffix still attested but its productivity reduced, and the latter co-expressing not two but three voices, namely anticausative, potential passive and passive.

Further attested developments of middle markers in Bantu languages include affixes innovating new voice functions that did not previously exist in the language or its ancestors. One well-studied example is the development of antipassive voice encoded by reciprocal markers (Bostoen et al. 2015). The diachronic development of Bantu affixes into middle markers along with the multiplicity of dedicated derivational morphology raises important questions as to the viability of cross-linguistic comparisons of middle voice systems.