HAL Id: halshs-03045136
https://shs.hal.science/halshs-03045136
Submitted on 22 Dec 2021
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entic research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diusion de documents
scientiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
The Bantu Relative Agreement Cycle
Mark van de Velde
To cite this version:
Mark van de Velde. The Bantu Relative Agreement Cycle. Linguistics, 2021, 59 (4), pp.981-1015.
�10.1515/ling-2021-0113�. �halshs-03045136�
Mark L. O. Van de Velde*
The Bantu relative agreement cycle
https://doi.org/10.1515/ling-2021-0113
Received April 1, 2019; accepted August 22, 2020; published online June 23, 2021
Abstract: This article presents the Bantu relative agreement (BRA) cycle, a scenario
of recurrent morphosyntactic change that involves the emergence of relativizers,
which are subsequently integrated into the relative verb form, where they can
ultimately replace the original subject agreement prefix. All logical outcomes at
every stage of the cycle are amply attested in the languages of the Bantu family.
The BRA cycle makes sense of many of the puzzling characteristics of relative
clause constructions in the Bantu languages, especially in the domain of
agreement.
Keywords: agreement; Bantu; historical syntax; relative clauses
1 Introduction
Relative clause constructions in the Bantu languages differ from each other along a
bewildering amount of parameters, some of which are correlated and typologically
exceptional. This explains why they have received so much attention in the liter-
ature (e.g., Atindogb
´
e and Grollemund 2017; Demuth and Harford 1999; Diercks
2009; Gu
´
erois and Creissels 2020; Henderson 2007; Meeussen 1971a; Nsuka-Nkutsi
1982; van der Wal 2010; Zeller 2004). Parameters of variation include, among
others, the position of the subject in non-subject relative clauses; the presence or
absence of a relativizer; the position of the relativizer; the relativizer’s morpho-
logical status (word versus clitic versus affix); whether and how relative verb forms
are morphologically different from non-relative verb forms; which paradigm of
agreement markers is used in relative verb forms; whether this paradigm has first
and second person forms; and, finally, which element in the sentence controls
agreement on the relative verb: its subject or the relativized NP.
This paper provides evidence for a recurrent scenario of morphosyntactic
change in the Bantu languages – the Bantu Relative Agreement (BRA) cycle –
*Corresponding author: Mark L. O. Van de Velde, LLACAN (UM R 813 5 du CNRS – INALCO - EPHE),
7, rue Guy Môquet - BP 8, 94801 Vi llejuif, France, E-mail: mark.vandevelde@cnrs.fr. https://
orcid.org/0000-0002-2803-2130
Linguistics 2021; 59(4): 981–1015
which accounts for much of the variation. It focuses on the last parameter of
variation in the preceding enumeration, viz. the controller of agreement on
therelativeverb,whichisarguablythe most remarkable from a typological
point of vie w. We ca n sum marize and label t he ma in poss ibilit ies as follows
below.
(1) Type
SBJ: agreement with the subject only
Type
NP
rel
-SBJ: agreement with the relativized NP and the subject
Type
NP
rel
: agreement with the relativized NP only
These three possibilitie s can be found in constructions throughout the Bantu
area, except for a notable gap in the distribution o f Type
NP
rel
in the North-
East (see Nsuka-Nkutsi (1982) for an overview of the areal distribution). They
have been regularl y reported to e xist for non-subject relatives in the litera-
ture, but I will show that the three exist in subject relatives as well, where the
differences between the types are h arder to notice, because the subject and
the relativized NP are co-referential. Moreover, we will see that this typology
of agreement types known from the Bantuist literature should be enriched
with at least a fourth type: double agreement with the relativized NP,
i.e., Type
NP
rel
-NP
rel
. For the sake of simplicity, I will not systematicall y
mention it i n what follows.
Many Bantuists have used the terms direct and indirect to typologize relative
clause constructions. According to Zeller (2004: 76), the former term has been used
for subject relatives and the latter for non-subject relatives since Doke (1954), but
Doke himself rather uses it to distinguish between constructions on the basis of the
paradigm of agreement markers they use. Since Meeussen (1967), the term direct
relative has often alternatively been used to mean relative clause constructions of
Type
NP
rel
or Type SBJ and indirect relative for those of Type NP
rel
-SBJ. I will avoid this
terminology because its different interpretations by different authors are a source
of confusion and because its binary nature makes it unsuitable for capturing the
rich typological diversity that it is supposed to cover. See Van de Velde (forthc.) for
arguments against reconstructing a direct and indirect relative clause construction
in Proto-Bantu.
The Haya examples in (2) illustrate the typologically widespread agreement of
Type
SBJ, in which the same verbal prefix is used to index the subject in main
clauses (2a) and relative clauses (2b) alike. The relevant morpheme is glossed
IN,
short for Initial, which refers to a position in the slot-filler template of the Bantu
languages that will be presented in 2.1. Using this convention will allow us to keep
982
Van de Velde
track of which kind of agreement marker occurs where across constructions
belonging to different agreement types.
1
(2) Haya; Tanzania
a.
`
e-m-bwá y-à-ly’
´
ebìtòòk
`
e
AUG
9
-9-dog IN
9
-PST-eat AUG
8
-8-banana
‘The dog has eaten the bananas.’
b.
`
e-bì-tòòk’ [
´
eby’
´
e-m-bwá y-á-lyà]
AUG
8
-8-banana DEM
8
AUG
9
-9-dog IN
9
-PST-eat
‘the bananas that the dog has eaten’
(Duranti 1977: 121)
The Orungu examples in (3) illustrate Type
NP
rel
. The verb in the main clause in (3a)
begins with a prefix that indexes the subject. In contrast, the same position in the
corresponding relative verb form in (3b) is used to index the relativized NP, the class
6noun‘mangoes ’. The relative verb form in (3b) does not index its subject ‘woman’.
(3) Orungu; Gabon
a. óꜜŋwánt àgòlín óꜜg ábà
óꜜŋw-ántò à-à-gòl-in-í ó-gà á-bà
1-woman.
DTP IN
1
-RPST-buy-APPL-RPST 1-chief.DTP 6-mango.DTP
‘The woman bought the mangoes for the chief.’
b. ábà [mágòlìn óꜜŋwánt ꜜógà]
á-bà má-à-gòl-in-ì óꜜŋw-ántò ó-gà
6-mango
IN
6
-RPST-buy-APPL-REL.RPST 1-woman 1-chief
‘the mangoes that the woman bought for the chief’
(Van de Velde and Ambouroue 2017: 623)
Example (4b) illustrates agreement of Type
NP
rel
-SBJ in the Kaonde language: both
the relativized NP and the subject are indexed on the relative verb.
(4) Kaonde; Zambia, DRC
a. ba-njimi ba-sa-kú-lal-a mu-nzubo ayo
2-9.farmer
IN
2
-FUT-INF-sleep-FV 18-9.house DEM
9
‘The farmers will sleep in this house.’
1 I added glosses to examples from older sources that lack them. Glossing conventions and
abbreviations are those of the Leipzig Glossing Rules. The glosses
IN Initial, preIN Preinitial and IF
Infix are labels for positions in the comparative Bantu verbal template; and PPr Pronominal prefix,
VPr Verbal prefix and RPr Relative prefix are used to refer to paradigms of agreement markers,
introduced in Section 2. Arabic numbers are used to refer to noun classes, except in combination
with
SG and PL, where they are used to refer to first or second person. Additional glosses: AUG
Augment, DTP definite tone pattern, RPST Remote past, RPT Recent past.
The Bantu relative agreement cycle
983
b. mu-nzubo ayo [mo-ba-sa-kú-lal-á ba-njimi]
18-9.house
DEM
9
preIN
18
-IN
2
-FUT-INF-sleep-FV 2-9.farmer
‘the house where the farmers will sleep’
(Kawasha 2008: 46)
Another parameter of variation that can be observed in the preceding examples is
the position of the NP that expresses the agent of the relative verb.
2
In (2b) the
relative verb form is preceded by its subject, whereas the subject follows the verb in
(3b) and (4b). This is no coincidence. There is a statistical correlation between
preverbal subjects and Type
SBJ agreement on the one hand and postverbal subjects
and Types
NP
rel
and NP
rel
-SBJ agreement on the other. A relatively rare example of a
construction in which a postverbal subject coexists with Type
SBJ agreement is (5b).
(5) Luvale; Angola, Zambia
a. mu-tu w-a-mwangana a-mbata-nga ci-teli (…)
1-person
PPr
1
-CON-chief IN
1
.PST-carry-PST 7-load
‘The chief’ s person carried the load.’
b. vi-ze [mw-a-va-han-a mw-ata va-kw-etu]
8-
DEM FUT-IN
1
-IF
2
-give-FV 3-master PPr
2
-PPr
15
-1PL.POSS
‘those (things) which the master will give to our fellows’
3
(Horton 1949: 21, 177)
A further salient detail in the above examples is that the agreement marker mo- at
the beginning of the relative verb form in (4b) does not belong to one of the
paradigms of agreement markers found on verbs or on other agreement targets. It
belongs to a dedicated paradigm of relative agreement markers and originates in a
pronominal form that historically consisted of an agreement marker of class 18 and
the pronominal stem o.
The three agreement types, their random distribution over the Bantu area, the
statistical correlation with word order and the unexpected shape of certain
agreement markers are beautifully accounted for by a scenario of recurrent lan-
guage change that is here called the Bantu Relative Agreement (BRA) cycle. The
three major stages of the BRA cycle are schematized for non-subject relatives in
Examples (6)–(8). The representations in Examples (6)–(8) are schematic and
2 Whether the NP that expresses the agent of the relative verb fulfills the grammatical relation of
subject depends on the construction and on the definition of subject one chooses to adopt. It is of
no importance for the current topic and I will consistently call it subject for the sake of simplicity.
3 The class 1 agreement marker on the verb is due to semantic agreement, which is systematically
triggered by human subject controllers in Luvale, whatever their morphological class marker.
984 Van de Velde
illustrative. They do not summarize all possible states of affairs at every stage. They
also ignore any typological variation in the morphology of relative verbs that is not
touched by the BRA cycle. As has been said, subjects can be preverbal or post-
verbal in non-subject relative clauses, depending on the language and on whether
they are nominal or pronominal. The difference is relevant in Stage 2, Example (7),
and will also be shown in the schematic representation of Stage 1, Example (6).
In Stage 1, an element emerges in between the relativized NP and the relative
clause of Example (6). This element can either be a pronoun that functions as a
nominalizer, in the sense that it turns a modifying relative clause into a headless,
independent relative clause, or it can emerge as a linker/relativizer. It is common in
the Bantu languages for adnominal modifiers to be nominalized (i.e., turned into
referring expressions) and put in apposition with the head noun (Van de Velde
2019). The pronominal element that fulfils this function with relative clauses can be
either a personal pronoun, a demonstrative or a so-called augment. Such apposed
nominalized modifiers subsequently tend to become reintegrated into an integral
NP, leading to the reinterpretation of the initial nominalizer as a linker, in this case
a relativizer. Alternatively, the element that emerges in between the relativized
noun and the relative clause can be a so-called connective relator, used to mark
genitive relations (see Van de Velde 2013). Crucially, the emerging element tends to
be an agreement target and when it is, it agrees with the relativized NP. It is
represented as
REL in the following schematic representations, and will henceforth
be called relativizer.
(6) example of a pre-Stage 1 situation Stage 1
a.
HEAD
i
[AGR
j
-V SUBJECT
j
(…)] → HEAD
i
REL
i
[AGR
j
-V SUBJECT
j
(…)]
b.
HEAD
i
[SUBJECT
j
AGR
j
-V (…)] → HEAD
i
REL
i
[SUBJECT
j
AGR
j
-V (…)]
Second, the relativizer is reinterpreted as part of the relative verb form and
morphologically integrated to become a bound marker that indexes the head noun
in Example (7a). The result is Type
NP
rel
-SBJ agreement, i.e., agreement with both the
relativized NP and the subject. This step is hampered when the relative verb has a
nominal subject that is in preverbal position, in which case the relativizer tends to
stay in between the head noun and the subject (but see Section 4). It can, but need
not become prosodically integrated into the subject nominal in (7b). Stage 2 is
optional in the BRA cycle, in that constructions can either stay at Stage 1 or move
directly to Stage 3 through haplology (see Example (44c) from Punu and its dis-
cussion for evidence).
(7) Stage 1 Stage 2
a.
HEAD
i
REL
i
[AGR
j
-V SUBJECT
j
(…)] → HEAD
i
[AGR
i
-AGR
j
-V SUBJECT
j
(…)]
b.
HEAD
i
REL
i
[SUBJECT
j
AGR
j
-V (…)] → HEAD
i
[REL
i
(-)SUBJECT
j
AGR
j
-V (…)]
The Bantu relative agreement cycle
985
Finally, the succession of two agreement prefixes at the beginning of the relative
verb in (7a) is reduced. If the second prefix goes, we end up with Type
NP
rel
agreement in (8a). If the first prefix goes, the result is typologically well-behaved
subject agreement, as if nothing has happened (8b).
(8) Stage 2 Stage 3
a.
HEAD
i
[AGR
i
-AGR
j
-V SUBJECT
j
(…)] → HEAD
i
[AGR
i
-V SUBJECT
j
(…)]
b.
HEAD
i
[AGR
i
-AGR
j
-V SUBJECT
j
(…)] → HEAD
i
[AGR
j
-V SUBJECT
j
(…)]
The Bantu Relative Agreement cycle is permanently potentially active in the Bantu
languages. It applies to constructions, rather than to entire languages. Depending
on the language it can touch subject relatives, non-subject relatives, both or
neither. When it touches both, it may play out differently in different construc-
tions, for instance in having different origins of the relativizer in subject and non-
subject relatives. There are also many examples of relative clause constructions
that are in Stage 2 and Stage 1 at the same time, i.e., that have entered a new cycle
while being halfway through a previous cycle. Such constructions have a relative
verb form with double agreement preceded by a morphologically independent
relativizer. Finally, the three stages of the Relative Agreement Cycle are not
discrete. It is often impossible to tell the difference between an independent rel-
ativizer (Stage 1) and an agreement marker (Stage 2) in a non-arbitrary way.
Likewise, the transition from Stage 2 to Stage 3 tends to be incremental, with
successions of agreement markers surviving in some contexts whereas a merger
has taken place in others (see the discussion of Luba type languages in Sections 4
and 5).
The following sections provide evidence for the BRA cycle. Section 2 looks at the
shape of agreement prefixes in Bantu relative verb forms, which often differ from
those found in non-relative verbs. Sections 3 to 5 illustrate the three stages of the
Bantu Relative Agreement cycle, all of which are widely attested in contemporary
Bantu languages. Section 6 adds to the evidence by showing how alternative relative
clause constructions in a single language (Punu) are situated in the BRA cycle.
2 The agreement marker paradigm of the Initial
This section will look at the paradigms of agreement prefixes found in relative verb
forms as an important indication towards explaining the variation in agreement
controllers. I will first introduce the structure of the Bantu verb and then discuss
the different paradigms of agreement prefixes. I will point out a striking correlation
between the paradigm of agreement markers used in a relative verb form and the
type of agreement it has.
986
Van de Velde
2.1 The structure of the Bantu verb
Comparative and historical studies in Bantu linguistics often use a slot-filler model
for the structure of the Bantu verb, introduced by Meeussen (1967). A simplified
version is provided in (9). A number 1 in the second row means that maximally one
morpheme can fill the position, whereas n stands for one or more. Brackets are used
to mean that the position can be left empty, depending on the verb form. I
N is short
for Initial,
FO for Formative, IF for Infix (the name of a prefix position, i.e., not an
actual infix),
EXT for Extension and FIN for Final. These are Bantu-specific names for
positions in Meeussen’s slot-filler model. The Initial position is typically filled by a
form from the paradigm of so-called Verbal prefixes (VPr) and the infix position by
an Object prefix (OPr). These prefix paradigms will be discussed in Section 2.2.
(9) pre
IN-IN-postIN-FO-IF-root-EXT-FIN+postFIN
(1) (1) (1) (n)(n)1 (n) 1 (1)
Verb forms are usually easy to segment into morphemes, but their meaning tends
to be non-compositional. The Final morpheme, for instance, contributes to the
expression of tense, aspect and/or mood, together with the Formative(s), but its
shape can also depend on the presence of an Object index in the Infix position. The
slots that are relevant for this study are the Initial and the Pre-initial, and to a lesser
extent the Infix and the Post-final. The Initial slot is usually occupied by a prefix
that indexes the subject of the verb. It tends to be glossed with a functional label
such as
SM (subject marker) or with a paradigmatic label such as VPr (Verbal prefix).
Since we are interested in explaining why this morpheme is sometimes used to
index the relativized NP and since in those cases it tends to be taken from another
paradigm than that of the Verbal prefixes, the purely positional label Initial is very
useful for our purposes. The Pre-initial, if present, is often filled by a negative
marker or by a prefix indexing the relativized NP in relative verbs that agree with
both their head noun and their subject (Type
NP
rel
-SBJ). The so-called Infix is the last
of a series of prefix positions before the verb root. It hosts a reflexive marker and/or
one or more prefixes indexing objects. Example (10) from Yao contains a relative
verb form in which these three positions are fi lled. The Pre-initial indexes the class
6 relativized NP ‘reeds’. It is followed by the class 2 subject index in Initial position.
The tense/aspect/negation prefix kaga- in Formative position is followed by a class
6 Object index in Infix position, which agrees with the same controller as the Pre-
initial morpheme.
The Bantu relative agreement cycle
987
(10) Yao; Malawi, Mozambique, Tanzania
ma-tete [ga-a-kana-ga-saka]
6-reed pre
IN
6
-IN
2
-NEG.PST-IF
6
-want
‘reeds which they did not want’
(Sanderson 1922: 75)
2.2 Paradigms of class markers
The Bantu languages have several paradigms of class prefixes. Meeussen (1967)
reconstructs five such paradigms in Proto-Bantu, provided in Table 1. The first
paradigm is that of the Nominal prefixes (NPr), used as overt class markers on
nouns and in some languages as noun class agreement markers on adjectives. The
second are Numeral prefixes (EPr), used to mark agreement in noun class on lower
adnominal cardinal numbers.
4
The third are the so-called Pronominal prefixes
(PPr), used to mark agreement in noun class on pronouns and adnominal modifiers
such as demonstratives. The fourth and fifth are the verbal prefixes (VPr) and
Object prefixes (OPr), used respectively to index subjects and objects on the verb.
Contemporary languages can have a higher or lower number of distinct paradigms.
Table : The Proto-Bantu class marker paradigms (Meeussen : ) (abridged).
NPr EPr PPr VPr OPr
SG –– –ǹ- ǹ-
SG –– –ʊ
̀
-kʊ
̀
-
PL –– –tʊ
̀
-tʊ
́
-
PL –– –mʊ
̀
-mʊ
́
-
cl mʊ
̀
-(ʊ
̀
?) jʊ
̀
- ʊ
́
-, á- mʊ
̀
-
cl bà- bá- bá- bá- bá-
cl mʊ
̀
-(ʊ
́
?) gʊ
́
-gʊ
́
-gʊ
́
-
cl mɩ
̀
-(ɩ
́
-?) gɩ
́
-gɩ
́
-gɩ
́
-
cl ì- dɩ
́
-dɩ
́
-dɩ
́
-dɩ
́
-
cl mà- (á-?) gá- gá- gá-
cl kɩ
̀
-kɩ
́
-kɩ
́
-kɩ
́
-kɩ
́
-
cl bì- bí- bí- bí- bí-
cl n- (ɩ
̀
-) jɩ
̀
-jɩ
́
-jɩ
́
-
cl n- í- jí- jí- jí-
…………… …
4 The abbreviation EPr is short for Enumerative prefix, chosen by Meeussen (1967: 81) to avoid
confusion with the Nominal prefix.
988 Van de Velde
As can be seen in Table 1, the formal differences between the reconstructed
paradigms are minimal. The relevant distinction for this study is between the
Pronominal prefixes (PPr) and the Verbal prefixes (VPr). First and foremost, the PPr
paradigm differs from the VPr paradigm in its lack of first and second person forms.
Second, in Meeussen’s reconstruction the Pronominal pre fi xes have a low tone in
class 1 and 9, where Verbal Prefixes are high. And third, there is a segmental
difference in class 1.
2.3 Agreement markers in relative verb forms
It has been known since Meinhof (1948) that the Initial position of Bantu relative
verb forms is often occupied by a Pronominal prefix, instead of the Verbal Prefix
used in other verb forms. In his comparative study of the relative clause con-
structions of the Bantu languages, Nsuka-Nkutsi (1982) determines for each con-
struction in his sample of slightly over 100 languages which paradigm it uses. He
also specifies for each construction whether the verb agrees with its subject or with
the relativized NP in non-subject relatives. When comparing these two parameters
of variation in Nsuka’s data, I found an implicational correlation that appears to be
exceptionless:
(11) If the Initial of a relative verb indexes the subject, it is occupied by a verbal
prefix
5
The inverse implication holds as well, but only as a (strong) tendency:
(12) If the Initial of a relative verb indexes the relativized NP, it tends to be
occupied by a Pronominal prefix
The remainder of this section discusses the four criteria that Nsuka-Nkutsi (1982:
114) used to determine whether the Initial or Pre-initial position of a relative verb
form is occupied by a Pronominal prefix (PPr), namely:
(13) a. the presence of an augment;
b. the shape of the class 1 prefix;
c. the absence of 1st and 2nd person forms;
d. a tone difference between the prefixes of class 1 and 9 (sometimes
also 4 or 6) and those of the other classes.
5 This Verbal prefix may be preceded by an invariable element, as in the Mbagani construction
discussed and illustrated in Section 5, Example (38). Together, such a combination of a Verbal
prefix and an invariable element can be analyzed as a dedicated Relative prefix. The implicational
universals in (11)–(12) are based on Nsuka-Nkutsi’s data and analyses, where Verbal prefixes are
opposed to Pronominal prefixes and dedicated Relative prefixes are ignored.
The Bantu relative agreement cycle
989
Nsuka-Nkutsi encountered problems when applying his criteria. These are pre-
dicted by the BRA cycle, because they are due to the fact that the markers he
identified as Pronominal prefixes are reflexes of agreeing relativizers, rather than
direct reflexes of prefixes from a single Proto-Bantu PPr paradigm. Many of the
paradigms he and others have identified as Pronominal prefixes should therefore
be recognized as separate Relative prefix (RPr) paradigms. Once this is clear, the
implicational universals in (11)–(12) are easily explained as a natural consequence
of the BRA cycle schematized in (6)–(8). I will briefly discuss his criteria in what
follows.
The term augment is used by Bantuists to refer to a bound form that precedes the
class prefix of nouns and some adnominal or nominalized modifiers (de Blois 1970).
Formally, the augment is typically either identical to the PPror it consists of the vowel
of the PPr. Meeussen (1967: 99) reconstructs it into Proto-Bantu as a weak demon-
strative, but there is evidence for multiple occurrences of augment creation (from
demonstratives) and loss in the Bantu languages (Van de Velde 2019b: 247–255).
When prefixed to nouns, the funct ion of the augmen t, if any, differs from language to
language. Often, one can only list the conditions in which it does or does not appear,
and the former tend to be far more numerous than the latter. In Ganda, for instance,
nouns usually have an augment, but it is absent on an object noun after a negative
verb form or when it is under focus as in (14b) (Hyman and Katamba 1993).
(14) Ganda; Uganda
a. y-à-gúl-ìr-à à-bá-ànà
`
e-bí-tábó
IN
1
-PST-buy-APPL-FV AUG-2-child AUG-8-book
‘He bought the children books.’
b. y-à-gúl-ír-á bá-ànà
`
e-bí-tábó
IN
1
-PST-buy-APPL-FV 2-child AUG-8-book
‘He bought the children books.’
(Hyman and Katamba 1993: 228–229)
The augment does have a clear function when it is used as a nominalizer, typically
one that changes an adnominal modifier into a referring expression, as in the
Nande examples in (15b) and (15d), where the nominalizing augment is
underlined.
(15) Nande; DRC
a. ɔ
̀
mʊ
̀
kɩ
̀
rá ɣwáːyɔ
̀
‘its tail’
b.
ɔ
́
-ɣwáːyɔ
̀
‘his one’
c. ɔ
̀
mʊ
̀
tɩ
́
mùkúhí ‘the short tree’
d. ò
-mùkúhí ‘the short one’
(Valinande 1984: 642, 709, 714)
990
Van de Velde
Nsuka-Nkutsi points out that the relative verb forms of several languages begin
with an element that is formally identical to the augment. His somewhat circular
reasoning is that, since a VPr cannot be preceded by an augment, the following
(Pre-)Initial must be a PPr. In the BRA scenario, this augment can have different
origins. It can be part of a relativizer that originates in a demonstrative of the form
v-PPr (where v is a vowel identical to that of the PPr), as appears to be the case in
Lamba non-subject relatives (16b).
(16) Lamba; Zambia
a. i-fi-ntu ifi
AUG-8-thing 8.DEM
‘these things’
b. i-fi-ntu [ifi tw-abwēne]
6
AUG-8-thing preIN
8
IN
1
PL
-saw
‘the things we saw’
(Doke 1938: 119, 434)
Alternatively, the augment can itself be (the reflex of) a relativizer.
7
It could have
been added to a relative verb form in Stage 2 (with an initial PPr) or, contra Nsuka-
Nkutsi, to a verb form in Stage 1, with an initial VPr.
Nsuka-Nkutsi’s second and fourth criteria ( 13b) and (13 d) look at the
segmental and tonal shape of the agreement markers. Segmentally, any dif-
ferences between VPrs and PPrs are r estricted to class 1, for which Meeussen
reconstructs á- and ʊ
́
- in the VPr paradigm and jʊ
̀
- in the PPr (see Nsuka-Nkutsi
1982: 118–121 for an overview table). Tonally, the differences are restricted to
classes 1 and 9 ( sometimes also to 4 and/or 6 ), where the VPr tends to have a
reflex of a *H and the PPr a reflex of a *L. The Mab ale ex amples i n (17) show tha t
theclass1markerininitialpositionoftherelativeverb(17b)differssegmentally
from that of the non-relative verb (17a), but is ide ntical to t he ag reement prefix
on the adnominal demonstrative (17c). The class 1 VPr a nd PPr both have a low
tone in Mabale, so ( 17) does not illustrate the tonal di fference that can be found
in some Bantu languages.
6 Doke separates the relativizer/agreement marker from the rest of the verb. It is not clear whether
this decision is based on independent evidence or whether it is simply an orthographic convention,
but this is of little importance.
7 Remember that relativizer is here used as a cover term for the element that appears between the
relativized NP and the relative clause in Stage 1. This element often starts out as a pronoun that
nominalizes the relative clause.
The Bantu relative agreement cycle
991
(17) Mabale; DRC
a. à-
mò-b
`
eng-í nà mbángù
IN
1
-IF
1
-pursue-PST with haste
‘She pursued him with haste.’
8
b. mò-tò [ò-w
´
e-í]
1-person
IN
1
-die-PST
‘the person who died ’
c. mò-tò ò-
bô ‘this man’
(Motingea 1996: 233, 243)
These two criteria need to be applied with care and knowledge of the individual
languages. On the one hand, reflexes of *jʊ and *ʊ have merged in many
languages and some languages tend to have a cl ass 1 verbal prefix *a- in most
verb forms, but *ʊ- in others. On the o ther hand, many Bantu languages have
tonal morphology, so that tonal differences between the Initial of a relative
verb and a non-relative verb may be due to other factors than their membership
of different paradigms. However, what counts for the tonal criterion is not so
much the presence of a low tone here or a high tone there, but the existence of a
tonal distinction between the m arkers of class 1 and 9 and those of the other
classes in the PPr paradigm, versus the absence of such a distinction in the VPr
paradigm.
Finally, Nsuka-Nkutsi used the absence of fir st and second person forms in
a paradigm of agreement markers to identify that paradigm as the one used to
mark agreement with nominal controllers on adnominal modifiers, i.e., as
Pronominal prefixes. When a first or s econd person pronoun is the head of a
relative clause construction in which the relativized NP is indexed by a PPr on
the relative verb, it triggers so-called enforced agreement (Corbett 1991: 204),
the details of which differ from language to language. In Orungu, for instance,
first and second person relativized NPs trigger class 9 agreement ( 18), the
default agreement pattern triggered by (pro)nominal controllers that lack a
gender specification, such as proper names. Most often, enforced agreement is
semantically motivated and results in a class 1 prefix for singular controllers
and a class 2 prefix for plurals, as in the Nkore-Ki ga example in (19), where the
class 2 marker a(-)ba- could be a succession of augment and V Pr or PPr, or a
reflex of a V-PPr relativizer of demonstrative origin. The absence of first and
8 In line with the glossing conventions adopted in this paper for indexes in verb forms, the third
person singular object marker in this example is glossed as
IF
1
, short for “Infix” of class 1, where
Infix is the name of a prefix position in Meeussen’s slot-filler model of the Bantu verb.
992 Van de Velde
second pe rson forms is the m ost straightforward of Nsuka-Nkutsi’s criteria, but
few grammatical descriptions provide examples of relative clauses with a
relativized NP of the first or second person.
(18) Orungu; Gabon
myɛ
́
/àwɛ
́
/ àzwɛ
́
/ ànwɛ
́
[y-áꜜmy
´
enì gó táŋgúnà]
1
SG 2SG 1PL 2PL IN
9
-can.REL LOC read
‘I/you(
SG)/we/you(PL) who can read’
(Van de Velde and Ambouroue 2011: 125)
(19) Nkore-Kiga; Uganda
imwe [a-ba-tuura aha], mu-raire buhooro
2
PL preIN
2
-IN
2
-live here IN
2
PL
-sleep well
‘You, who live here, how are you (lit. did you sleep well)?’
(Taylor 1985: 23)
Kagulu has a dedicated agreement marker go- for first and second person con-
trollers, which is of unknown origin (20). In subject relatives, it is added in Pre-
initial position to a VPr in Initial position, which marks subject agreement in
person and number. As in subject relatives with third person controllers, the Pre-
initial prefix marking agreement with the head NP is optional in Kagulu.
9
(20) Kagulu; Tanzania
aniye [(go-)ni-ku-kwenda]
1
SG preIN
1/2/
SG
/
PL
-IN
1
SG
-IF
2
SG
-love
‘I who love you’
(Petzell 2008: 189)
The possible existence of such subject relatives in which the relative verb agrees
twice with the same first or second person controller, once as the relativized NP
(with a third person form) and once as its subject, is perfectly predicted by the BRA
cycle. They are a clear illustration of the fact that the three types of agreement
summarized in (1) can be found as much in subject relatives as in non-subject
relatives. An example with a third person plural marker of class 2 in Pre-initial
position is provided in (21).
9 Petzell does not give examples of object relatives with first or second person heads. Note that
nothing identifies (20) as a relative clause construction when the Pre-initial is left out. There are
many examples in the Bantu languages of grammatical ambiguity between relative and non-
relative clauses. It is not clear whether and in which languages intonational means can or must be
used to disambiguate.
The Bantu relative agreement cycle
993
(21) Yao; Malawi, Mozambique, Tanzania
10
a. uwe [u-tu-li ŵa-yao]
1
PL preIN
2
-IN
1
PL
-be 2-yao
‘we who are Yao ’
b. mwe [u-m-jigele]
2
PL preIN
2
-IN
2
PL
-have.brought
‘you who have brought’
(Sanderson 1922: 73)
To summarize: the prefix paradigm used to index the subject on relative verb forms
in constructions of Type
SBJ differs from that used to index the head NP in lan-
guages of Type
NP
rel
. The latter is similar to the PPr paradigm of prefixes used to
mark agreement with the head noun on certain adnominal modifiers, but often not
fully identical to it. The BRA cycle explains why this is so: prefixes that index the
head NP on relative verb forms originate in agreeing relativizers, or are the result of
a merger between an original relativizer and a following subject prefix. The
following three sections discuss the three stages of the BRA cycle.
3 Stage 1
The Bantu Relative Agreement Cycle starts with the emergence of a nominalizer or
linker (here both called relativizer). The Bantu languages show impressive varia-
tion in the source of relativizers, but also in their initial syntactic status, position
and function. As pointed out in great detail by Nsuka-Nkutsi (1982: 3–93), the three
main sources of relativizers in the Bantu languages are demonstratives, personal
pronouns (aka substitutives) and genitive (aka connective) relators, in decreasing
order of frequency. I will illustrate Stage 1 with relativizers of demonstrative origin.
Note that Stage 1 is here used to refer to the emergence of a new relativizer. This
can happen in every logically possible type of source construction. In other words,
the BRA cycle does not have a stage zero. The left-hand side of the schematic
representation in (6), repeated below, is only one of very many alternative starting
points.
(6) possible starting point Stage 1
a.
HEAD
i
[AGR
j
-V SUBJECT
j
(…)] → HEAD
i
REL
i
[AGR
j
-V SUBJECT
j
(…)]
b.
HEAD
i
[SUBJECT
j
AGR
j
-V (…)] → HEAD
i
REL
i
[SUBJECT
j
AGR
j
-V (…)]
10 The shape of the class 2 Pre-initial in relative verb forms is unexpected, and arguably due to the
fact that it is a reflex of a relativizer, rather than of a PB Pronominal prefix. Important here is that
the same marker is used when the relativized NP is a noun of class 2.
994 Van de Velde
Alternatively, an extra relativizer can emerge in constructions that already have a
relativizer, or in constructions where the relative verb indexes both the relativized
NP and the subject, in constructions where the relative verb indexes the relativized
NP, or in any combination of these.
Examples (22)–(25) show relative clauses introduced by means of a relativizer
that is analyzed as a separate word in their sources, illustrating Stage 1. The
Nyakyusa example is a subject relative introduced by a near demonstrative. In
class 3, the paradigms of Verbal and Pronominal prefixes have identical forms, so
this example does not show whether the relative verb agrees with ʊn
̩
dʊngʊ as its
subject or as its head noun.
11
(22) Nyakyusa; Tanzania, Malawi
ʊn
̩
dʊngʊʊgʊ gʊkwisa
ʊ-mʊ-lʊngʊ [ʊ-gʊ gʊ-kʊ-is-a]
AUG
3
-3-week AUG
3
-NDEM
3
IN
3
-PRS-come-FV
‘next week’ (lit. ‘the week that is coming’)
(Persohn 2017: 39)
The Haya examples in (23) illustrate Stage 1 in non-subject relative clause con-
structions with a lexical (23a) and a pronominal (23b) subject. In the former, the
relativizer is separated from the relative verb by means of the lexical subject, which
is in preverbal position. In the latter, the relativizer is adjacent to the verb. A
construction such as that in (23a) cannot easily evolve into Stage 2, because the
integration of the relativizer into the relative verb is hampered by the intervening
subject. We will come back to this in Section 4.
(23) Haya; Tanzania
a.
`
e-bì-tòòk’ [
´
eby’
´
e-m-bwá y-á-lyà]
AUG
8
-8-banana DEM
8
AUG
9
-9-dog IN
9
-PST-eat
‘the bananas that the dog has eaten’
b. ò-mw-áán’ [ówó n-à-bónà]
AUG
1
-1-child DEM
1
IN
1
SG
-PST-eat
‘the child that I have seen’
(Duranti 1977: 121)
The examples in (23) show that the BRA cycle has the theoretical potential of
leading to languages in which the verb of a non-subject relative clause shows
double agreement when its subject is pronominal (Type
NP
rel
-SBJ, Stage 2) versus
11 Examples with a relativized noun of class 1, where the VPr and PPr paradigms are distinct, show
that the relative verb has a VPr in Nyakyusa subject relatives, which consequently show no
evidence of having gone through a BRA cycle before.
The Bantu relative agreement cycle
995
only subject agreement when its subject is lexical (Type SBJ, Stage 1). It is telling
that according to Nsuka-Nkutsi’s data this situation can be found in at least two
languages, viz. Bubi and Kimbundu (1982: 217, 244).
Finally, Examples (24) and (25) show a non-subject relative clause that has a
postverbal lexical subject and that is introduced by a morphologically indepen-
dent relativizer.
(24) Chokwe; DRC, Angola
ly-onda [líz
´
e a-a-mbách-ile pwo]
5-egg
REL
5
IN
1
-TNS-carry-RP 1.woman
‘the egg which the woman carried’
(Kawasha 2008: 50)
The Nzadi example (25) shows that relativizers can be non-agreeing, e.g., when
they consist of only the stem of a demonstrative. When such relativizers are inte-
grated into the relative verb, their merger with the following subject prefix can give
rise to a rare dedicated paradigm of Relative prefixes used to mark subject
agreement, but not to double agreement, or to agreement with the relativized NP.
(25) Nzadi; DRC
fùfú [nà ò dz
´
e múùr]
fufu
REL PST eat person
‘the fufu that the person ate’
(Crane et al. 2011: 160)
Relativizers of demonstrative origin are very common in both subject and non-
subject relatives. Among the languages that have a demonstrative relativizer in
Nsuka-Nkutsi’s sample of 107 Bantu languages, 61% have them in subject and non-
subject relatives, 27% only in non-subject relatives and 12% only in subject rela-
tives. They can be either optional or obligatory and they can be taken from one,
some or all types of demonstratives available in the language (e.g., near versus far,
neutral versus emphatic, and so on).
In many cases, the same demonstrative forms can be used pronominally or
adnominally. Heine and Kuteva (2002: 115) assume that relativizers of demon-
strative origin evolve naturally out of demonstrative pronouns, but in at least some
Bantu languages they must have begun their life as modifiers of the relativized
noun. Evidence comes from languages in which the relativizer can be placed
before or after the relativized noun (Nsuka-Nkutsi 1982: 23), an alternative word
order that also exists for adnominal demonstratives in many Bantu languages (Van
de Velde 2005). When relative clause constructions are marked by an obligatorily
prenominal demonstrative modifier of the head noun, this demonstrative can
become integrated into the head noun, resulting in a construct form of nouns,
996
Van de Velde
i.e., a form of the noun used when it takes certain modifi ers. This has happened in
Eton (Van de Velde 2017).
(26) Eton; Cameroon
a. kòpí í-nə
̀
ɲól î=kpə
̀
m
[9]coffee
IN
9
-be [9]color CON
9
=[9]cassava.leave
‘The coffee is green.’
b. í-kôpí í-nə
̌
ɲól î=kpə
̀
m
CSTR-[9]coffee IN
9
-be.REL [9]color CON
9
=[9]cassava.leave
‘green coffee’ (lit. ‘coffee that is green’)
(Van de Velde 2008: 347)
Stage 1 is often reinitiated by adding an augment or another demonstrative form in
front of a relative clause that is already marked by means of relativizer. Authors
tend to point out that this adds emphasis, most probably meaning contrastive
focus on the relative clause. These are examples of relative clauses being nomi-
nalized and placed in opposition to the head noun, mentioned in the character-
ization of Stage 1 in the introduction. The Nkore-Kiga examples in (27) differ in the
presence or absence of an augment a- in front of the relativizer ku. Taylor (1985)
describes the relative clause introduced by the augment in (27b) as a defining one,
as opposed to the non-defining one in (27a). In Rijkhoff’s (2002: 173–212) termi-
nology, the augmented relative clause is an anchoring modifier. It is used to enable
the hearer to identify the correct reference of the relativized noun and it carries a
presupposition of existence. The non-augmented relative clause, in contrast, is
used to qualify or classify the relativized noun.
(27) Nkore-Kiga; Uganda
a. a-ka-cumu ku w-aakozesa
AUG-12-pen preIN
12
IN
2
SG
-used
‘a pen that you used’
b. a-ka-cumu a-ku w-aakozesa
AUG
12
-12-pen AUG
12
-preIN
12
IN
2
SG
-used
‘the pen you used’
(Taylor 1985: 22)
In examples such as (27), Taylor writes the relativizer separately from the relative
verb, whereas Nsuka-Nkutsi treats it as a Pre-initial PPr written together with the
relative verb wherever he cites Nkore examples. Example (27) reflects Taylor’ s
analysis in the orthography and Nsuka-Nkutsi’s in the glosses. Criteria for word-
hood famously tend not to converge (see, for instance, Haspelmath 2011) and the
scenario of the BRA cycle predicts some degree of indeterminacy during
the gradual morphological integration of erstwhile independent relativizers into
The Bantu relative agreement cycle
997
the relative verb. If we assume like Nsuka-Nkutsi that the relativizer has become
prefixed to the verb, then the result is a verb-initial succession of three agreeing
prefixes. I have glossed the first one as an augment in (27b), but a more coherent
glossing in the slot-filler model used here would be to recognize a Pre-pre-initial
slot. Another noteworthy aspect of these Nkore-Kiga examples is that the vowel of
the relativizer ku is unexpected. In most classes, the relativizer is identical to the
Pronominal prefix, but in the classes that have a vowel /a/ in the PPr (classes 2, 6,
12 and 16), the vowel of the relativizer is /u/. A similar phenomenon is found in
Herero (Namibia, Botswana) and Mituku (DRC), which are geographically and
genealogically far from Nkore-Kiga and from each other. I currently have no pre-
cise explanation for this phenomenon. The most likely hypothesis in view of the
BRA cycle is that this unexpected vowel is a reflex of the demonstrative or pro-
nominal stem of a relativizer that was reduced in the process of its morphological
integration into the relative verb form.
4 Stage 2
Stage 2 constructions are those in which the relative verb has two agreeing pre-
fixes, one in Pre-initial position and one in Initial position. Typically, the Pre-initial
marker agrees with the relativized NP and the Initial marker with the subject,
corresponding to agreement of Type
NP
rel
-SBJ, as in (28).
(28) Songye; DRC
`
e-fùbá dì-bá-báá-dím-ì
5-field pre
IN
5
-IN
2
-RPT-cultivate-RPT
‘the field that they have just cultivated’
(Stappers 1964: 117)
However, a relativizer can also be integrated into a relative verb that had already
gone through a BRA cycle, so that both the Pre-initial and the Initial agree with the
relativized noun, as in the Dzamba example (29). This is the fourth agreement type
mentioned in the introduction: Type
NP
rel
-NP
rel
.
(29) Dzamba; DRC
i-zi-bata [i-zi-eza-áki oPoso ba-butu loome]
AUG
5
-5-duck preIN
5
-IN
5
-give-IMPF [1]Poso 2-guest today
‘the duck that Poso gave the guests today’
(Bokamba 1976) (cited in Demuth and Harford 1999: 52)
Evidence for the fact that i- and zi- are distinct morphemes, rather than a sim-
plex izi- prefix that develop ed from a VCV- shaped demonstrative comes from
998
Van de Velde
negative verb forms, in which the two are separate d from each ot her by the
negative prefix ta- .
12
The evolution from Stage 1 to Stage 2 involves the univerbation of a relativizer
and a following relative verb. The Bantu languages provide many examples of
constructions in which this gradual process is only partially completed. One type
of example involves relativizers of demonstrative origin that are optionally
reduced, for instance due to haplology. Nsuka-Nkutsi provides examples from
Swati, Tsonga and Ronga in which a demonstrative of the shape la-V-PPr (where V
is the vowel of the PPr) is optionally shortened to la-V when used as a relativizer in
front of an identical Initial in subject relatives (30). Crucially, when the relativizer
is so reduced, it has the CV-shape of typical verbal prefixes.
(30) Swati; Eswatini, South-Africa
si-lwane [lesi si-ɓona-ko] OR si-lwane [le-si-ɓona-ko]
7-animal
REL
7
IN
7
-see-REL 7-animal preIN
7
-IN
7
-see-REL
‘the (wild) animal that sees’
(Nsuka-Nkutsi 1982: 14)
Likewise, Nsuka-Nkutsi (1982: 24) lists four languages in which the tone patterns of
demonstratives change when they are used as relativizers, as compared to their
adnominal or pronominal use, viz. Gusii, Fipa, Konzo and Bemba. In Konzo, this
tone change is accompanied by vowel shortening. This too may be an indication for
their prosodic integration into the verb, although that would have to be verified
language by language.
A last example of ongoing reanalysis of the relativizer can be found in some of
the Great Lakes Bantu languages, where relativizers seem to jump across a pre-
verbal lexical subject in non-subject relative clauses. This can happen either
optionally, as in Haya (Duranti 1977: 128), or obligatorily, as in Luganda (31).
(31) Luganda; Uganda
o-mu-sajja [Petero gwe a-labye] musomesa
AUG
1
-1-man [1]Peter REL
1
IN
1
-has.seen teacher
‘The man that Peter has seen is a teacher.’
(Walusimbi 1976: 24)
12 A reviewer wondered why I analyze the relative verb form in (29) as having double agreement
with the relativized NP, instead of analyzing the first prefix as the augment of the pronominal
agreement in Initial position. I have no objection against calling this morpheme an augment, albeit
an augment of the relative verb form, rather than of its Initial. However, I see no contradiction
between using this term from comparative Bantu to name the prefix and analyzing it as a second
agreement marker. Calling it an augment by no means explains why it is there.
The Bantu relative agreement cycle
999
The gradual change of the relativizer from pre-subject to pre-verbal position can be
explained in terms of frequency: in languages that have preverbal nominal sub-
jects in non-subject relative clauses, the relativizer is still adjacent to the verb in the
great majority of cases, viz. whenever the subject is first or second person or a third
person pronoun. Those cases can lead to the gradual reanalysis of the relativizer as
forming one constituent with the relative verb. Interestingly, the relativizer in (31)
has clearly been reinterpreted as being syntactically linked to the relative verb, but
it has not (yet) been integrated into the verb phonologically. If it were, it would
surface with a long vowel (Hyman and Katamba 1990). An example of a con-
struction where the relativizer has been morphologically integrated in the relative
verb can be found in Ikalanga (32).
(32) Ikalanga; Botswana
13
isípá [Neo cha-á-ká-pá Nchídzi]
7.soap 1a.Neo preIN
7
-IN
1
-PST-give 1a.Nchidzi
‘the soap that Neo gave Nchidzi’
(Letsholo 2009: 139)
Finally, there is an interesting typological distinction between two types of con-
structions that are in Stage 2 of the BRA cycle. Meeussen (1971a) calls them the
Luba type and the Lega type. In relative clause constructions of the Luba type, the
prefix that indexes the relativized NP can only be followed by a prefix that indexes
the subject if the latter is of the first or second person (33a), (34a). Third person
subjects are expressed by a postverbal pronoun, which is sometimes analyzed as
free (33b), encliticized or suffixed (34b). Languages with constructions of the Luba
type are found in the Eastern half of the DRC and in Eastern Angola. Constructions
of the Lega type, in contrast, have double agreement regardless of the person of the
subject (35).
(33) Mituku; DRC
a. mʊ-ntʊʊ
́
-tʊ-tʊ
́
ma
1-person pre
IN
1
-IN
1
PL
-send
‘the person we send’
13 A reviewer objected to the glossing of this example as imprecise, because the gloss preIN
7
of the
cha- prefix fails to show that it is “in fact” a connective (=genitive) marker. The choice for reflecting
these morphemes’ positions in a slot-filler model of the Bantu verb in the glossing, versus their
function or paradigm was justified in the introduction. As pointed out by Nsuka-Nkutsi and
mentioned at the beginning of Section 3, genitive linkers are a common source of relativizers in
what I have called Stage 1. Whether or not the etymology of cha- is a genitive linker, its position in
the relative clause shows that it has evolved into an agreement marker, according to Letsholo
(2009)’s analysis. The scenario of the BRA cycle explains why it is there and why it has the form it
does.
1000 Van de Velde
b. mʊ-ntʊʊ
́
-ꜜtʊ
́
ma bô
1-person
IN
1
-send they
‘the person they send’
(Stappers 1973: 59)
(34) Luba; DRC
a. Kabundi, [u-tu-asɛki]
1.Kabundi pre
IN
1
-IN
1
PL
-mocked
‘Kabundi, whom we mocked’
b. ba-lumiana [ba-mbidi-yɛ]
2-man
IN
2
-address-3SG
‘the men he addressed ’
(Meeussen 1946: 174)
(35) Lega; DRC
a. mǎ-zɩ
̀
[mà-tʊ
́
-ꜜká-kúbʊ
̀
lá]
6-water pre
IN
6
-IN
1
PL
-FUT-pour
‘water that we will pour’
b. mǎ-zɩ
̀
[mà-bá-ꜜká-kúbʊ
̀
lá]
6-water pre
IN
6
-IN
2
-FUT-pour
‘water that they will pour’
(Meeussen 1971b: 28)
Constructions of the Luba type are therefore half way in between Stage 2 and Stage 3:
they are in Stage 2 when they have a first or second person subject and in Stage 3
with a third person subject. We will briefly come back to this with a tentative
explanation in the next section.
5 Stage 3
The passage to Stage 3 involves a reduction of the succession Pre-initial – Initial or
relativizer # Initial to just Initial. There is indirect and direct evidence for the
likeliness of such an evolution in the Bantu languages. Indirect evidence may be
found in other examples of reduced successions of prefi xes. Direct evidence comes
from languages where the shape of the Initial of relative verbs points to a merger of
two earlier prefixes.
Whether a Bantu language has prosodically prominent word or phrase
penultimate syllables (Hyman 2013), stem-initial prominence (Hyman 1989; Idia-
tov and Van de Velde 2016), or both, their prefixes are generally prosodically weak.
The Bantu relative agreement cycle
1001
The farther they are from the macro-stem, the more vulnerable they are to
erosion.
14
This is especially the case in the North-West of the Bantu domain, where
strong stem-initial prominence can be linked to maximality constraints on stems as
well as to the reduction and loss of prefixed material. In the verbal domain, the
prosodic weakness of prefixes contributes to the erosion of Tense-Aspect prefixes,
which survive as floating tones or disappear to be renewed out of auxiliary con-
structions (Nurse 2008: 48–49). In the nominal domain, the prosodic weakness of
prefixes drives another type of recurrent change: the creation of CV-shaped
augment prefixes out of weak prenominal demonstratives, the subsequent loss of
their initial consonant and finally full segmental loss, leaving only a H tone and/or
vowel alternations in preceding words that end in /a/ (Van de Velde 2019b).
Likewise, in languages where a segmental augment is preserved, the augment is
very often not realized when a locative prefix precedes the overt noun class marker
(Gr
´
egoire 1975: 159–163). This might be because the retention of the augment in the
presence of a locative prefix would lead to successions of three or four prefixes in a
prosodically weak position.
Direct evidence for the evolution responsible for the passage from Stage 2 to
Stage 3 can be found in languages that have a dedicated Relative prefix paradigm
that can be shown to be the result of a merger of two prefixes. One case is where
languages have V- prefixes in some classes (typically 1 and 9), versus CV- prefixes
elsewhere, and where rules of vowel hiatus resolution reduce a VV succession to V.
Haya subject relatives, for instance, have Initial á- for class 1 (36b) and
´
e- for class
9, but a Pre-initial – Initial succession for the other classes, e.g., á-bà- (cl2), ó-gù-
(cl3) and so on (Byarushengo 1977: 8) (36a). In other words, Haya subject relative
clauses are in Stage 2, except in class 1 and 9, where they are in Stage 3, because the
Pre-initial has merged with the Initial. Since it is the first vowel that goes in Haya
vowel hiatus resolution, the segmental shape of the VPr a- survives in class 1,
instead of the /o/ of the Augment.
15
(36) Haya; Tanzania
a. e-ki-ntw’
´
e-ki-tá-li ki-hâŋgo
AU
7
-7-thing preIN
7
-IN
7
-NEG-COP 7-big
‘the thing that is not big’
14 The macro-stem consists of the Infix (= object marker) and everything that follows it (= the
stem). Tense-aspect prefixes that recently grammaticalized from auxiliaries may retain some of
their prosodic strength.
15 Duranti notes the class 1 and 9 subject relative agreement prefixes with a short vowel, whereas
their vowel is long in Byarushengo’s overview table. When the TA marker in Formative position is
vowel-initial, the subject relative agreement prefixes of class 1 and 9 are followed by a glide y, e.g.,
omusháíj’ á-y-a-bon’ ómukâzi ‘the man who saw the woman’ (man pre
IN
1
-IN
1
-PST-see woman).
1002 Van de Velde
b. o-mu-ntw’ á-li mu-hâŋgo
AU
1
-1-person IN
1
-COP 1-big
‘the person who is big’
(Duranti 1977: 120)
There is a sizeable area in the Congo Basin where almost all languages have
agreement of Type
NP
rel
. Several of these languages, such as Mongo, Ntomba,
Mabale and Bolia have a separate paradigm of Relative prefixes used both in
subject and non-subject relative verb forms. This paradigm resembles the Pro-
nominal prefix paradigm in lacking fi rst and second person forms and in the
segmental shape of its class 1 prefix. However, it differs tonally from both the PPr
and the VPr paradigms in that its prefixes have a rising tone. The class 3 agreement
prefix of the relative verb form in (37b) has a rising tone, whereas the verbal prefix
of the corresponding non-relative verb form in (37a) is high. Example (37c) shows
that the prefi x paradigm of relative verb forms lacks a form for the first person
singular, and that it uses a class 1 prefi x instead, with the rising tone of relative verb
prefixes.
(37) Mongo; DRC
a. bò-támbá bó-fúk-à
3-tree
IN
3
-move-PRS
‘The tree moves.’
b. bò-támbá [bǒ-fúk-á]
3-tree
IN
3
-move-PRS.REL
‘the tree that moves’
c.
`
emí ǒ-k
`
el-ákí
I
IN
1
-do-PST
‘I who have done it’
(Hulstaert 1965: 477)
Another typological characteristic of the languages of this area is that successions
of a Pre-initial and Initial prefix always have a rising tone scheme, whatever the
lexical tone of the morphemes involved. This is especially clear in Mongo non-
relative verb forms, where the order between the negative prefix (n)tá- and the
verbal prefix that indexes the subject is pragmatically conditioned (38). Both
prefixes have a lexical high tone, but their succession is low-high.
(38) Mongo; DRC
a. tò-tá-k
`
el-ákí
pre
IN
1
PL
-NEG-do-PST
‘We didn’t do (it).’
The Bantu relative agreement cycle
1003
b. ntà-tó-k
`
el-ákí
NEG-preIN
1
PL
-do-PST
‘We didn’t do (it).’
(Hulstaert 1965: 310)
The omnipresence of Type
NP
rel
in the zone C languages of the Congo basin, which
are surrounded by areas where Type NP
rel
-SBJ is common, combined with the fact
that these languages have an obligatory low-high rising tone pattern on Pre-initial-
Initial successions makes it highly likely that their dedicated Relative prefix par-
adigms with a rising tone are a result of the merger of a succession of prefixes.
16
The reduction of the succession of a PPr that indexes the relativized NP and a
VPr that indexes the subject logically has the potential of leading to a single new
prefix that functions as the continuation of the second, verbal prefix (VPr), espe-
cially in its function of indexing the subject. An example of this can be found in
Mbagani (=Binji). Here, non-subject relative clauses are optionally introduced by a
close demonstrative of the shape à-PPr, giving rise to the succession à- PPr-VPr-
when this demonstrative relativizer is integrated into the relative verb form. In this
succession, the PPr can go, leading to a paradigm of àVPr- Relative prefixes (38a)
(Stage 3). Relative verbs formed with this prefix can again optionally be introduced
by a PPr or à-PPr relativizer (38b) (Stage 1). This is another rare pathway that leads
to a paradigm of Relative prefixes used to index the subject, next to the one
illustrated with a Nzadi example in (25).
17
16 In Mongo and some other closely related languages, the only other agreement targets that have
a prefix with a rising tone are possessive modifiers. There are reasons to assume that these
possessive forms are based on a relative form of the copula. This is clearest in Bolia, where the stem
of possessive pronouns consists of nkì, the relative form of the copula, and a personal pronoun
(Mamet 1960: 29). See also Nsuka-Nkutsi (1982: 57–67) for a discussion of the formal resemblances
between connective (=genitive) and relative constructions in the Bantu languages.
17 The Nguni languages have a relative clause construction for non-subject relatives with a
dedicated Relative prefix of the shape (l)v-VPr- (where v is a vowel determined by the quality of the
vowel of the Verbal prefix) that indexes the subject of the relative clause. It looks as if this were an
example of Type SBJ-SBJ- agreement, which is incompatible with the BRA-scenario. In reality, the
quality of the first vowel of the Relative prefix is determined by anticipatory assimilation, as is
clearly shown in relative verb forms with a first or second person subject. The Relative prefix here is
most probably a Mbagani-style prefix of which the first vowel comes from the invariable
demonstrative stem (l)a-. Alternatively, an originally complex prefix of the type found in Nguni
subject relatives, illustrated with a Swati example in (30), could have lost its function of indexing
the relativized NP due to anticipatory assimilation. Another peculiarity of this Nguni construction
is that the relativizer has “jumped across” the lexical subject before becoming integrated in the
relative verb form, as has happened in Luganda and Ikalanga. See Zeller (2004) and Nsuka-Nkutsi
(1982: 13–18) for analyses of Nguni relative clause constructions that resemble the BRA scenario.
1004 Van de Velde
(39) Mbagani/Binji; DRC
a. di-kamá [abá-bátúlɛ
́
ˑla]
5-foot
REL.IN
2
-cut.off
b. di-kamá [(a)di abá-bátúlɛ
́
ˑla]
5-foot
REL
5
REL.IN
2
-cut.off
‘the foot that they cut off’
(Van Coillie 1948: 272)
A third type of possible outcome of the reduction of a PPr-VPr- succession (on top
of a merger or the total loss of one of the two) is the situation in which either can be
dropped and speakers have a free or conditioned choice between using a PPr- or a
VPr-. Nsuka-Nkutsi (1982: 131–132) reports that such a free choice exists with a first
or second person pronoun head in languages across the Bantu domain, viz. Duala,
Sanga, Tabwa, Nyungwe, Sena and Zulu. Orungu and Punu have an uncondi-
tioned choice between an agreement prefix of the PPr- and the VPr- paradigm in
subject relatives with a nominal head noun, which is only visible for controllers of
class 1, the only class where the two paradigms are formally distinguished. A Punu
example will be provided and discussed in Section 6.
Finally, the question remains why constructions of Meeussen’sLubatype
have failed to reduce the PPr-VPr- succession in their relative verb forms to PPr-
when the VPr- is a first or second person form ( versus everywhere else). Lan-
guages with Luba type constructions normally u se a specific paradigm of short
postverbal or verb-final pronominal forms to express the subject in non-subject
relatives. According t o N suka-Nkutsi (1982: 42, 222) these paradigms lack first and
second person forms. If no alternative way of expressing first or second p erson
subjects is available, the original VPr has to remain in place. The inverse is not
true: the availability of alternative pronominal subject markers i s not a sufficient
condition for Pre-initial – Initialsuccessionstobereducedandwecanfind
examples of pronominal subjects being expressed twice in relative clause con-
structions, e.g., (40).
(40) Yao; Malawi, Mozambique, Tanzania
cáákúlyá [cí-tú
-kút
´
el
´
eká úw
´
e] cákúnóŋa
7.food pre
IN
7
-IN
1
PL
-are.cooking we tasty
‘The food which we are cooking is tasty.’
(Whiteley 1966: 139)
Independent evidence for this explanation of Luba type situations can be found in
the languages where the marker that indexes the subject on relative verb forms is
obligatorily or optionally omitted in the presence of a lexical subject. In Makhuwa
non-subject relatives, for instance, the subject is expressed by means of a suffix
The Bantu relative agreement cycle
1005
whose presence is obligatory in the absence of a lexical subject, but optional in the
presence of a lexical subject (van der Wal 2010). Non-subject relatives in the Beya
dialect of Lega have agreement of Type
NP
rel
-SBJ when no lexical subject is present,
also with third person subjects, but of Type
NP
rel
in the presence of a lexical subject.
(41) Lega (Beya dialect); DRC
a. nnyama [zɩ-bá-tá-gɩk-
´
e]
9.meat pre
IN
9
-IN
2
-NEG-cook-FV
‘the meat that they don’t cook’
b. i-d
´
egá [lɩ
́
-ꜜkʊ
́
-tágɩ
́
l-ág-á Syaba mǎ-nzɩ mu-ly-ó]
5-pot
IN
2
-CONT-draw-IPFV-FV Syaba 6-water 18-5-PRO
‘the pot that Syaba draws water in’
(Botne 2003: 446)
From here, it is only a small step to Luba-type constructions, where an obligatory
postverbal subject pronoun fulfills a role similar to that of the lexical subject in the
Lega examples in (41) and where the paradigm of these postverbal subject pro-
nouns lacks first and second person forms, so that a VPr indexing a subject of the
first or second person cannot be omitted.
6 Punu: an example of a language with traces of
multiple BRA cycles
As has been pointed out in the introduction, the BRA cycle applies to individual
constructions, rather than to languages. Constructions found in contemporary
Bantu languages can show no traces of having gone through it, or they can show
traces of having gone through it partially or fully, either once or several times. The
three previous sections have illustrated the different stages of the cycle – as well as
different types of outcomes at every stage – by means of examples from the five
hundred or so languages spoken throughout the Bantu domain. However, evi-
dence for the BRA cycle can also be found in individual languages that have
multiple alternative relative clause constructions. This section discusses the
relative clause constructions of one such language, viz. Punu, spoken in the South
of Gabon. Punu is one of the languages where the lexical subject remains in
preverbal position in non-subject relative clauses. Since preverbal lexical subjects
hamper the continuation of the BRA cycle beyond Stage 1, and since this obstacle is
not present in subject relatives, the BRA cycle has the potential of creating more
constructional variation in subject relatives than in non-subject relatives,
depending, of course, on whether and how often Stage 1 is triggered in subject
1006
Van de Velde
relatives. This potential is realized in Punu. The examples in (43) illustrate the
three alternative constructions that can be used to relativize the object of (42).
(42) Punu; Gabon, Congo
tú-tsi-gűkíga bâː-nà
IN
1
PL
-PST-gather 2-child
‘We gathered the children.’
(Blanchon 1980: 115)
The construction in (43a) shows no obvious traces of having gone through a BRA
cycle, either partially or fully. It shows Type
SBJ agreement and the relative clause is
not introduced by means of a relativizer. The morphological differences between
the relative verb form and its non-relative counterpart, realized tonally, are most
probably not related to the BRA cycle. Examples (42b) and (42c) show two con-
structions that entered Stage 1 through different paths.
(43) a. báː-nà [tú-tsi-gukíga]
2-child
IN
1
PL
-PST-gather
‘the children that we gathered’
b. báː-nà [bó tu-tsi-gukíga]
2-child
PRO
2
IN
1
PL
-PST-gather
‘the children that we gathered’
c. báː-nà [ába tu-tsi-gukíga]
2-child
DEM
2
IN
1
PL
-PST-gather
‘the children that we gathered’
The relativizer in (43b) originates in a pronoun of the shape PPr-o,wherevowelhiatus
resolution results in the elision of the vowel of the Pronominal prefix. The origin of the
alternative relativizer in (43c)isademonstrativeoftheformá-PPr. Its final /a/ is
realized as [ə], the normal realization of /a/ in word-final position. This is a mor-
phonological argument for assuming that the relativizer has not been morphologically
integrated into the relative verb. A syntactic argument that leads to the same
conclusion is that the relativizer precedes any lexical subject (but remember from the
discussion of Luganda that the two criteria do not necessarily coincide). The gloss in
(43) represents the etymology of the relativizers, rather than their current function.
Turning to subject relatives, we find only two possible constructions when the
relativized NP is a third person form that is not of class 1, illustrated in (44b)–(44c).
(44) Punu; Gabon, Congo
a. bá-tu bá-tsí-rariga bì-kǔtu
2-person
IN
2
-PST-sew 8-garment
‘The people sewed the garments.’
The Bantu relative agreement cycle
1007
b. ba-tu [bó ba-tsi-ráriga bi-kǔtu]
2-person
PRO
2
IN
2
-PST-sew 8-garment
‘the people who sewed the garments’
c. ba-tu [ába-tsi-ráriga bì-kǔtu]
2-person
IN
2
-PST-sew 8-garment
‘the people who sewed the garments’
(Blanchon 1980: 116)
The subject relative clause construction in (44b) is structurally identical to the non-
subject relative clause construction in (43b): the relative verb agrees with its
subject and the relative clause is introduced by a relativizer of pronominal origin. It
is clearly at Stage 1. Things become more interesting in (44c), showing a con-
struction that has the same origin as the non-subject relative clause construction
illustrated in (43c): an á-PPr relativizer of demonstrative origin followed by a
relative verb of agreement Type
SBJ. However, the original succession á-PPr VPr-
has merged and simplified into an agreement marker of a new Relative pre fi x
paradigm. Since Pronominal and Verbal prefixes are identical from class 2 up-
wards, it is impossible to decide which of the two survived the merger, the PPr part
of the á-PPr relativizer or the VPr. Consequently, the resulting agreement is
ambiguous between Type
SBJ and Type NP
rel
. From a phonological point of view, the
result of the merger is a prefix rather than a morphologically independent rela-
tivizer, because its final /a/ is realized [a], rather than [ə]. We are at one of the
possible outcomes of Stage 3 here, viz. the creation of a new paradigm of dedicated
Relative prefixes (RPr), which was illustrated with a strikingly similar Mbagani
example (39) in Section 5.
BRA-fueled constructional exuberance kicks in when the relativized NP of a
subject relative belongs to class 1. There are five alternative ways to relativize the
subject of (45).
(45) mú-tu á-tsí-rariga yikǔtu
1-person
IN
1
-PST-sew garment
‘Someone sewed the garment.’
The subject relative shown in (46a) is introduced by the same á-PPr relativizer as
the non-subject relative in (43c). The Initial of the relative verb form is the Pro-
nominal prefix of class 1, meaning that the verb has agreement of Type
NP
rel
.In
other words, the construction exemplified by (46a) has gone through a BRA cycle at
least once and has subsequently started another one. It is therefore either in Stage 1,
or in Stage 2, as there are no principled criteria to determine whether águ is a prefix
synchronically or an independent relativizer in (46a). Interestingly, the second
1008
Van de Velde
cycle has been completed in the alternative construction illustrated in (46b) in an
evolution parallel to the one that led to the construction shown in (44c).
(46) Punu; Gabon, Congo
a. mu-tu [águ(-)gu-tsi-ráriga yikǔtu]
1-person
DEM
1
(-)IN
1
-PST-sew garment
‘someone who sewed the garment’
b. mu-tu [águ-tsi-ráriga yikǔtu]
1-person
IN
1
-PST-sew garment
‘someone who sewed the garment’
(Blanchon 1980: 116)
The alternative constructions illustrated in (47) involve the PPr-o relativizer of
pronominal origin. When the class 1 PPr gu- is prefixed to the pronominal stem o,
the two merge to wo or o. This relativizer is followed by a verb form whose Initial
can be either of the PPr paradigm (47a), or of the VPr paradigm (47b). Conse-
quently, the relative verb in (47a) is of agreement Type
NP
rel
and that in (47b) of
Type
SBJ. Example (47a) illustrates another construction at Stage 3 that has entered
Stage 1 of a new cycle. Alternatively, it can be analyzed as a Stage 2 construction,
because here too there is no way to decide in a non-arbitrary way whether the
original relativizer (w)o is a prefix, a proclitic or an independent relativizer. And
again, an alternative construction exists in which this most recent cycle has
reached its Stage 3 end point, the original relativizer having replaced the original
Initial of the relative verb (47c).
(47) Punu; Gabon, Congo
a. mu-tu [(w)o(-)gu-tsi-ráriga yikǔtu]
1-person
PRO
1
(-)IN
1
-PST-sew garment
‘someone who sewed the garment’
b. mu-tu [(w)o(-)a-tsi-ráriga yikǔtu]
1-person
PRO
1
(-)IN
1
-PST-sewed garment
‘someone who sewed the garment’
(Blanchon 1980: 116)
c. mu-tu [(w)o-tsi-ráriga yikǔtu]
1-person
IN
1
-PST-sewed garment
‘someone who sewed the garment’
In the absence of written historical sources, it is impossible to know which evo-
lution has led to the coexistence of the alternatives shown in (47a) and (47b). The
most straightforward scenario in view of the BRA cycle is that they are alternative
Stage 3 outcomes of a previous Stage 2 with a gu-a- succession. Alternatively, (47b)
could be an older construction that remained in existence while a BRA cycle
The Bantu relative agreement cycle
1009
created (47a). We just saw that the survival of older stages once newer stages have
emerged is not unusual in Punu. Finally, the Pronominal prefix could have been
the original prefix and the verbal prefix an innovation due to analogy with non-
subject relative verb forms.
In conclusion, the Punu data show that the BRA cycle can be triggered several
times in the same language, with different relativizers in Stage 1, and that the
evolution can be fast.
7 Conclusions
A sizeable portion of the extensive typological variation found among relative
clause constructions in the Bantu languages is straightforwardly explained by a
scenario of recurrent morphosyntactic change that I call the Bantu Relative
Agreement (BRA) cycle. As I argue in Van de Velde (forthc.), contra Meeussen’s
reconstruction of a “direct” and an “indirect” relative construction in Proto-Bantu,
the first application of BRA in any given language must have started from a con-
struction in which the relative verb agreed with its subject using a prefix from the
VPr paradigm. There is ample evidence of constructions that have gone through
several BRA cycles, either fully or partly. However, once subject agreement has
been lost as a result of a BRA cycle, it cannot be brought back. From this point,
subsequent BRA cycles can only renew agreement of type
NP
rel
.
At its first stage, Stage 1, the BRA cycle involves the emergence of an element in
between the relativized NP (
NP
rel
) and the relative clause. This element tends to be
an agreement target, and when it is, it agrees with
NP
rel
. Its original function is
either that of a linker or of a nominalizer, i.e., a pronominal form that changes an
adnominal relative clause into an independent one. In both cases, I have called it
relativizer, for lack of a better cover term. This relativizer tends to originate in a
demonstrative, a personal pronoun or a genitive linker. Because relative clauses
are mostly verb-initial in the Bantu languages, the mutually adjacent relativizer
and relative verb tend to merge, so that the relativizer is reinterpreted as an
agreement marker indexing the relativized NP. This can lead to a succession of
agreement markers, one that indexes
NP
rel
and one that indexes the subject of the
relative verb (Stage 2). Alternatively, or subsequently, the succession of the orig-
inal relativizer and the original subject prefix can be simplified by means of a
merger or of the deletion of one of them, usually the subject marker (Stage 3). This
simplification can be conditioned by the shape of the agreement prefixes involved,
or by the person of the subject prefix.
Perhaps the most rewarding aspect of the BRA scenario is that it not only
makes sense of many of the major typological characteristics of Bantu relative
1010
Van de Velde
clause constructions, but also of many minute and otherwise puzzling details in
the grammars of individual languages. Among the former is van der Wal’s (2010)
observation that the verbs of non-subject relative clause constructions in
Makhuwa (Type
NP
rel
) are best analyzed as participles. This analysis has been taken
up for Cuwabo non-subject relative verb forms by Gu
´
erois and Creissels (2020),
who furthermore point out that these participles assign a specific grammatical role
to the agent of the relative verb, one that differs from that of the subject of non-
relative verbs and from that of an adnominal possessor. The BRA cycle shows that
the participial nature of such relative verbs is due to the integration of an adno-
minal modifier into the verb, where it becomes an agreement marker that tends to
wipe out the original subject agreement marker. The latter’s obligatory presence in
other verb forms is a defining criterion for the recognition of a subject role in the
Bantu languages. The participial nature of relative verb forms has been noted for
non-subject relative clause constructions, but nothing in the BRA excludes subject
relatives to become participial by the exact same mechanism, and examples pre-
dictably abound throughout the Bantu family. If this has not been noticed before, it
is probably due to the fact that the distinction between verbal subject-predicate
agreement and participial noun-modifier agreement is harder to notice if the
subject and the relativized NP are co-referential. The distinction is clearest with
relativized NPs of the first and second person, where adnominal/participial
agreement markers are necessarily third person forms.
Another result of the BRA mechanism that has gone largely unnoticed in the
comparative and descriptive literature on Bantu relative verb constructions is that
many languages have a dedicated paradigm of Relative prefixes. These are typi-
cally identified as Pronominal prefixes in Nsuka-Nkutsi (1982), who consequently
had many difficulties in determining the paradigm of agreement markers. The BRA
cycle explains the emergence of dedicated Relative prefix paradigms as being due
to the merger of a succession of prefixes, the first of which most recently developed
out of a relativizer. Examples include the fact that, in Nsuka-Nkutsi’s terms (1982:
102), relative verb forms in Nyoro, Ganda, Haya, Kerebe, Kete and Bemba can have
a verbal prefix when agreeing with a class 1 controller but a Pronominal prefix
elsewhere (see the discussion of Haya subject relatives in Section 5). In the same
vein, BRA provides a strong hypothesis for explaining the rising tone on agreement
prefixes of relative verb forms in Mongo and other languages spoken in the Congo
Basin.
Finally, the different logically possible outcomes at every stage of the cycle are
amply attested in constructions throughout the Bantu domain, sometimes in
alternative constructions in individual languages. The BRA scenario can theoret-
ically give rise to relative verbs agreeing twice with the relativized NP, and such
The Bantu relative agreement cycle
1011
constructions are indeed attested in the contemporary Bantu languages. In
contrast, it provides no path leading to relative verbs agreeing twice with their
subject, and to my knowledge such relative verb forms are not attested. The BRA
scenario also makes statistical predictions. Although agreement of Types
NP
rel
-SBJ
and NP
rel
in non-subject relative clause constructions with a preverbal lexical
subject are compatible with the BRA scenario, they are predicted to be rare,
because the integration of a relativizer in the relative verb form is hampered by the
intervening subject NP. As I showed, agreement Type
NP
rel
-SBJ is attested in such
constructions in a few languages. In contrast, I have so far found no examples with
Type
NP
rel
agreement in non-subject relatives with a preverbal lexical subject.
Acknowledgments: This work is partially supported by a public grant overseen by the
French National Research Agency (ANR) as part of the program “Investissements
d’Avenir” (reference: ANR-10-LABX-0083). It contributes to the IdEx Universit
´
ede
Paris – ANR-18-IDEX-0001. I wish to thank Denis Creissels, Larry Hyman, and Dmitry
Idiatov, as well as the audiences at CALL in Leiden (2017) and Reconstructing Proto-
Bantu Grammar in Ghent (2018) for their comments.
References
Atindogb
´
e, Gratien Gualbert & Rebecca Grollemund (eds.). 2017. Relative clauses in Cameroonian
languages, structure, function and semantics. Berlin & Boston: De Gruyter Mouton.
Blanchon, Jean A. 1980. La relativization. In François Nsuka Nkutsi (ed.), El
´
ements de description
du punu, 115–128. Lyon: Centre de Recherches Linguistiques et S
´
emiologiques (CRLS),
Universit
´
e Lumi
`
ere (Lyon 2).
de Blois, Kees F. 1970. The augment in the Bantu languages. Africana Linguistica IV.87–165.
Bokamba, Eyamba Georges. 1976. Question formation in some Bantu languages. Bloomington, IN:
Indiana University dissertation.
Botne, Robert. 2003. Lega (Beya dialect) (D25). In Derek Nurse & G
´
erard Philippson (eds.), The
Bantu languages (Routledge Language Family Series), 422–449. London & New York:
Routledge.
Byarushengo, Ernest Rugwa. 1977. Preliminaries. In Ernest Rugwa Byarushengo,
Alessandro Duranti & Larry M. Hyman (eds.), Haya grammatical structure: Phonology,
grammar, discourse (Southern California Occasional Papers in Linguistics), 1–15. Los
Angeles, CA: Department of Linguistics, University of Southern California.
Corbett, Greville G. 1991. Gender. Cambridge University Press.
Crane, Thera Marie, Larry M. Hyman & Simon Nsielanga Tukumu. 2011. A grammar of Nzadi [B865].
(University of California Publications in Linguistics 147). Berkeley, CA: University of California
Press.
Demuth, Katherine A. & Carolyn Harford. 1999. Verb raising and subject inversion in Bantu
relatives. Journal of African Languages and Linguistics 20. 41–61.
1012 Van de Velde
Diercks, Michael. 2009. Subject extraction and (so-called) anti-agreement effects in Lubukusu: A
criterial freezing approach. In Papers from the Annual Meeting of the Chicago Linguistic
Society 45(1). 55–69.
Doke, Clement M. 1938. Text book of Lamba grammar. Johannesburg: Witwatersrand University
Press.
Doke, Clement M. 1954. The southern Bantu languages. London: Oxford University Press.
Duranti, Alessandro. 1977. Relative clauses. In Ernest Rugwa Byarushengo, Alessandro Duranti &
Larry Hyman (eds.), Haya grammatical structure (Southern California Occasional Papers in
Linguistics 6). Los Angeles, CA: Department of Linguistics, University of Southern California.
Gr
´
egoire, Claire. 1975. Les locatifs en bantou. Tervuren: Royal Museum for Central Africa.
Gu
´
erois, Rozenn & Denis Creissels. 2020. The relative verb forms of Cuwabo (Bantu P34) as
contextually oriented participles. Linguistics 58(2). 463–491.
Haspelmath, Martin. 2011. The indeterminacy of word segmentation and the nature of morphology
and syntax. Folia Linguistica 45(1). 31–80.
Heine, Bernd & Tania Kuteva. 2002. World lexicon of grammaticalization. Cambridge: Cambridge
University Press.
Henderson, Brent. 2007. The syntax of agreement in Bantu relatives. In Frederick Hoyt,
Nikki Seifert, Alexandra Teodorescu & Jessica White (eds.), Morphosyntax of
underrepresented languages (Texas Linguistics Society 9), 167–184. Stanford, CA: CSLI
Publications.
Horton, Alonzo E. 1949. A grammar of Luvale, 2nd edn. Johannesburg: Witwatersrand University
Press.
Hulstaert, Gustaaf. 1965. Grammaire du lɔmɔ
́
ngɔ: Deuxi
`
eme partie: Morphologie (Annales Du
Mus
´
ee Royal de l’Afrique Centrale – S
´
erie in-8° – Sciences Humaines). Tervuren: Royal
Museum for Central Africa.
Hyman, Larry. 2013. Penultimate lengthening in Bantu. In Balthasar Bickel, Lenore Grenoble &
David Peterson & Alan Timberlake (eds.), Language typology and historical contingency: In
honor of Johanna Nichols (Typological Studies in Language 104), 309–330. Amsterdam &
Philadelphia: John Benjamins.
Hyman, Larry M. 1989. Accent in Bantu: An appraisal. Studies in the Linguistic Sciences 19.
115–134.
Hyman, Larry M. & Francis Katamba. 1990. Final vowel shortening in Luganda. Studies in African
Linguistics 21(1). 1–59.
Hyman, Larry Michael & Francis X. Katamba. 1993. The augment in Luganda: Syntax or pragmatics?
In
Theoretical aspects of Bantu grammar, 209–256. Stanford, CA: CSLI Publications.
Idiatov, Dmitry & Mark Van de Velde. 2016. Stem-initial accent and C-emphasis prosody in north-
western Bantu. In Paper presented at the 6th International Conference on Bantu Languages,
Helsinki, Finland. http://idiatov.mardi.myds.me/talks/2016_BANTU6_C-emphasis_Idiatov_
Van_de_Velde_SLIDES.pdf.
Kawasha, Boniface Kaumba. 2008. Relative clauses and subject inversion in Chokwe, Kaonde,
Lunda and Luvale. Africana Linguistica XIV. 37–61.
Letsholo, Rose. 2009. The “forgotten” structure of Ikalanga relatives. Studies in African Linguistics
38(2). 131–154.
Mamet, M. 1960. Le Langage des Bolia (Lac L
´
eopold II) (Annalen van Het Koninglijk Museum van
Belgisch-Congo: Reeks in 8 33). Tervuren: Royal Museum for Central Africa.
Meeussen, Achille Emile. 1946. Syntaxis van het Tshiluba (Kasayi) deel 5. Kongo-Overzee 12/13.
173–184.
The Bantu relative agreement cycle
1013
Meeussen, Achille Emile. 1967. Bantu grammatical reconstructions. Africana Linguistica III.
81–121.
Meeussen, Achille Emile. 1971a. Relative clauses in Bantu. Studies in African Linguistics
Supplement 2. 3–10.
Meeussen, Achille Emile. 1971b. El
´
ements de grammaire lega. Tervuren: Royal Museum for Central
Africa.
Meinhof, Carl. 1948. Grundzüge einer vergleichenden Grammatik der Bantusprachen. Hamburg &
Berlin: Eckardt & Messtorff.
Motingea, Mangulu. 1996. El
´
ements de grammaire mabale (bantou C.30): Sur la base des textes
de J. Tanghe. Afrika und Übersee 79(2). 203–258.
Nsuka-Nkutsi, François. 1982. Les structures fondamentales du relatif dans les langues bantoues
(Annalen – Reeks IN-8° – Menswetenschappen). Tervuren: Royal Museum for Central Africa.
Nurse, Derek. 2008. Tense and aspect in Bantu. Oxford: Oxford University Press.
Persohn, Bastian. 2017. The verb in Nyakyusa: A focus on tense, aspect, and modality
(Contemporary African Linguistics 2). Berlin: Language Science Press.
Petzell, Malin. 2008. The Kagulu language of Tanzania: Grammar, texts and vocabulary (East
African Languages and Dialects 19). Cologne: Rudiger Köppe Verlag.
Rijkhoff, Jan. 2002. The noun phrase. Oxford: Oxford University Press.
Sanderson, Meredith. 1922. A Yao grammar, 2nd edn. New York & Toronto: Macmillan.
Stappers, Leo. 1964. Morfologie van het Songye (Annalen Reeks In-8° Wetenschapen van de
Mens). Tervuren: Royal Museum for Central Africa.
Stappers, Leo. 1973. Esquisse de la langue mituku. Tervuren: Royal Museum for Central Africa.
Taylor, Charles. 1985. Nkore-Kiga (Croom Helm Descriptive Grammars). London: Croom Helm.
Valinande, Nzama Kambale. 1984. The structure of Kinande. Washington, DC: Georgetown
University dissertation.
Van Coillie, Gustaaf. 1948. Korte Mbagani-Spraakkunst: De taal van de “Babindji” in Kasayi.
Kongo-Overzee 14. 257–279.
Van de Velde, Mark. Forthcoming. Agreement on relative verb forms. In Koen Bostoen,
Gilles-Maurice de Schryver, Gu
´
erois Rozenn & Sara Pacchiarotti (eds.), Reconstructing Proto-
Bantu grammar. Berlin: Language Science Press.
Van de Velde, Mark. 2005. The order of noun and demonstrative in Bantu. In Koen Bostoen &
Jacky Maniacky (ed.), Studies in African comparative linguistics with special focus on Bantu
and Mande, 425–441. Tervuren: Royal Museum for Central Africa.
Van de Velde, Mark. 2008. A grammar of Eton (Mouton Grammar Library). Berlin & New York:
Mouton de Gruyter.
Van de Velde, Mark. 2013. The Bantu connective construction. In Anne Carlier &
Jean-Christophe Verstraete (eds.), The genitive (Case and Grammatical Relations across
Languages), 217–252. Amsterdam & Philadelphia: John Benjamins.
Van de Velde, Mark. 2017. The augment as a construct form marker in Eton relative clause
constructions. In Gratien Atindogbe & Rebecca Grollemund (eds.), Relative clauses in
Cameroonian languages: Structure, function and semantism (Empirical Approaches to
Language Typology [EALT] 58). Berlin & Boston: De Gruyter Mouton.
Van de Velde, Mark. 2019a. Les expressions nominales dans les langues bantu. Lyon: Universit
´
e
Lumi
`
ere Lyon 2 HDR Dissertation.
Van de Velde, Mark. 2019b. Nominal morphology and syntax. In Mark Van de Velde, Koen Bostoen,
Derek Nurse & G
´
erard Philippson (eds.), The Bantu languages, 2nd edn, 237–269. London &
New York: Taylor and Francis.
1014 Van de Velde
Van de Velde, Mark & Odette Ambouroue. 2011. The grammar of Orungu proper names. Journal of
African Languages and Linguistics 32. 113–141.
Van de Velde, Mark & Odette Ambouroue. 2017. The origin and use of a relative clause construction
that targets objects in Orungu (Bantu, Gabon). Studies in Language 41(3). 615–637.
van der Wal, Jenneke. 2010. Makhuwa non-subject relatives as participial modifiers. Journal of
African Languages and Linguistics 31. 205–231.
Walusimbi, Livingstone. 1976. Relativization and focusing in Luganda and Bantu. Los Angeles:
UCLA dissertation.
Whiteley, Wilfred Howell. 1966. A study of Yao sentences. Oxford: Clarendon Press.
Zeller, Jochen. 2004. Relative clause formation in the Bantu languages of South Africa. Southern
African Linguistics and Applied Language Studies 22(1/2). 75–93.
The Bantu relative agreement cycle
1015
