Locatives in Shona and Luganda
Pavel Caha
Marina Pantcheva
CASTL, Tromsø
September 12, 2014
Abstract
At a general level, this paper is concerned by the categorisation of expressions
in natural languages. We approach this question with a relatively new tool in hand:
phrasal spell-out. The basic idea is this: if phrasal spell-out exists, a single item
may correspond to several terminals, where each terminal has a distinct label. As
a consequence, the approach predicts the existence of expressions whose behaviour
corresponds to a mixture of prototypical categorial properties.
This paper applies this relatively new analytical option to locative markers in
Shona and Luganda. We contrast the behaviour of the locative markers in these
languages with more familiar Indo-European adpositions, in order to establish a new
class of items: locative class markers. The behaviour of the new class, however,
is not explained by positing a new category in the decomposed projection, but
by proposing that locative class markers correspond to a combination of several
existing categories.
1 Introduction
In Shona, Luganda as well as in other Bantu languages, location in space is expressed
by morphemes which are referred to as noun class markers. In their prototypical
instanciations, noun class markers express two functions: class/gender (e.g.,
animate vs. inanimate) and number (singular vs. plural). A consequence of juxtaposing
these two statements is the conclusion that location in Bantu is expressed
by gender/number portmanteau markers. This is an unexpected state of affairs.
We could compare this, for instance, to a situation where the grammar of Italian
includes a statement such that the relation of spatial inclusion (‘IN’) is expressed
by adding a neuter singular affix to the noun/noun phrase.
For Italian and other Indo-European languages, the descriptive grammars do not
include such statements. In Bantu, such statements are very common. According
to a large part of the literature (Myers 1987, Bresnan 1995, Bresnan and Mchombo
1995, Carstens 1997), the difference lies in the linguistic categorization of abstract
meaning types. Location in most of Indo-European has a dedicated category, P,
while location in most of Bantu has the category N. From this single difference, a
whole set of contrasts between Bantu and average Indo-European (which we review
below) is meant to follow.
We will argue that such a view is oversimplified. We present old and new
evidence which suggests that – while on the right track – the nominal analysis
1
accounts only for half of the facts: those which show that the locative class markers
share some properties with nouns. However, there is also evidence that they do not
behave like nouns with respect to some other tests (Carstens 1997), and this is the
part of the data which the nominal analysis captures only by stipulation.
The empirical evidence then suggests that we are looking at a category which
is in some sense intermediate between prototypical nouns and prototypical adpositions.
If that is so, the Bantu situation is in fact only one particular instance of a
wider pattern. In particular, Svenonius (2006) discusses a whole range of languages
with similar items that do not fit either a nominal or an adpositional category and
suggests that we need to postulate “a syntactic category which is distinct from both
N and P, which [he] call[s] AxPart for ‘Axial Part”.
Svenonius’ proposal is one out of two major responses to the existence of such
expressions. The other one, inspired by Kayne’s (2004) work, proposes that the
intermediate properties “stem from the fact that [locative markers] are modifiers [...]
of a non-phonologically realised noun [...] Place [...] whose presence in the syntactic
structure gives locatives a nominal flavor” (Terzi 2010, 197, see also Botwinik-Rotem
2008, Pantcheva 2008, Aboh 2010, Cinque 2010).
Trying to decide between the two approaches is difficult, as they seem to have
complementary strengths and weaknesses. The silent PLACE proposal has the
advantage that it includes both an adposition-like marker (the overt morpheme) and
an actual noun (the silent PLACE). This immediately yields the relevant mixture
of nominal and adpositional characteristics. However, it is not immediately clear
why the overt silent PLACE modifiers cannot modify a non-silent noun. Similarly,
it is not clear why the silent PLACE cannot be modified by other modifiers, e.g.,
adjectives.
Such facts are taken by Svenonius as a starting point for his alternative analysis
and lead him to conclude that there is in fact no (silent) noun. At the same time,
it seems that under the AxPart analysis, the mixture of nominal and adpositional
properties does not immediately follow from anything: AxPart is postulated as
an independent category whose relationship to nominal or other properties is not
apriori expected.
Against this background, our contribution is going to be twofold. First of all,
we will present a new type of empirical evidence from Shona and Luganda that
strengthens the idea of having a silent noun PLACE in the structural representation
of locative markers. Second of all, we will present a new version of the silent PLACE
proposal that relies on phrasal spell-out (Starke 2009 and related work). More
specifically, we are going to argue that some of the drawbacks associated to the
original silent PLACE analysis may be removed if the actual lexical items spell out
a whole phrasal projection that includes PLACE as well as the adpositional node.1
2 Shona
In this section, we highlight the relevant empirical facts concerning Shona locative
markers. These basic observations carry over to Luganda, which we discuss later
on against the shared background.
1
The idea can be traced back to a suggestion by Klaus Abels at one of the seminars ran by Peter
Svenonius in Tromsø in 2006. See http://www.hum.uit.no/mra/SemSumms/V06/Feb28.htm.
2
2.1 Noun class markers
Let us first introduce the notion of a class marker. Class markers are prefixes that
encode nominal class (or gender) and number. A couple of examples are below,
noun class markers are in boldface. Following the tradition, they are glossed by
numbers where odd numbers usually correspond to the singular meaning, and even
numbers to the plural meaning.
(1) a. mu-nhu
1-NHU
a person
b. va-nhu
2-NHU
people
c. chi-nhu
7-NHU
thing
d. zvi-nhu
8-NHU
things
As the contrast between (1-a,b) shows, the difference in the form of the class marker
encodes the difference between singular and plural. The pair (1-c,d) shows the same
thing. The contrast between (1-a,c) shows that the class marker also encodes the
distinction in animacy. These facts lead to the conclusion that noun class markers
are portmanteau morphemes for class/gender and number.
It should be also mentioned that the root and the marker form often a noncompositional
semantic unit (an idiom), as well as a phonological unit (tonal marking
is superimposed over the two markers). See, e.g., Taraldsen (2010) FerrariBridgers
(2008) or Carstens (1997) for a recent discussion of the issues in generatively
oriented frameworks.
Noun class markers enter into a rich set of concord/agreement relations. This
is illustrated in (2-a,b).
(2) a. mu-komana
1-boy
mu-kuru
1-big
uyu
1.that
u-ri
1-is
mu-mota
in-car
That big boy is in the car
b. va-komana
2-boy
va-kuru
2-big
ava
2.that
va-ri
2-is
mu-mota
in-car
Those big boys are in the car
The examples differ in the number of the initial head noun, marked by mu (sg)
and va (pl). The difference in the marking of the head noun is further reflected on
the adjective, the demonstrative and the verb.
2.2 Locative markers
The examples in (3) show one way of expressing location in Shona. What we can see
is that there are three distinct markers (pa, ku, mu) that precede the class-marked
noun, and yield the meaning of ON, BY and IN respectively. These morphemes
are called locative noun classes in the descriptive literature. Because of this, they
are also glossed by the relevant class number assigned to them in the traditional
classification.
(3) a. nhunzi
10-fly
dzi-ri
10-be
pa-ma-poto
16-6-pots
The flies are on the pots
b. nhunzi
10-fly
dzi-ri
10-be
ku-ma-poto
17-6-pots
The flies are there by the pots (flying around, pots further away)
3
c. nhunzi
10-fly
dzi-ri
10-be
mu-ma-poto
18-6-pots
The flies are in the pots
Despite their name, which suggests a relation to ordinary non-locative classes, locative
markers look like prepositions at the first blush. First of all, it can be shown
that they attach to the whole extended NP that includes the demonstrative and
other modifiers.
This is illustrated below in (4). The starting point is the observation that
demonstratives in Shona (as well as Luganda) can either precede (less common) or
follow the noun (more common). When the demonstrative precedes the noun, it
comes in between the locative marker and the noun, see (4-b). It cannot occur to the
left of the locative class marker, see (4-c). This shows that the syntactic position of
the locative markers is identical to the location of prototypical prepositions: they
take a whole DP as a complement, and turn it into a location. The analysis is
depicted at the first line of (4).
(4) [ LOC [ DemP ] ]
a. pa-mu-komana
16-1-boy
uyu
1.this
b. pa-uyu
16-1.this
mu-komana
1-boy
both: on this boy
c. *uyu
1.this
pa-mu-komana
16-1-boy
The phrasal analysis straightforwardly explains also the distribution of possessors in
such examples. Just like demonstratives, possessors in Shona (as well as Luganda)
may both follow (unmarked) or precede the noun (marked). When the precede,
they come in between the locative marker and the noun:
(5) a. pa-mu-komana
16-1-boy
w-angu
1-my
b. pa-w-angu
16-1-my
mu-komana
1-boy
both: on my boy
c. *w-angu
1-my
pa-mu-komana
16-1-boy
This contrasts with the behaviour of non-locative class markers, which are tightly
bound to their nominal host, and no phrase may intervene between the two. For
instance, diminutives in Shona (like locatives) are formed by stacking a diminutive
class marker on top of a regular class marker, see (6-a,b).2
(6) a. mu-ti
3-tree
‘a tree’
b. ka-mu-ti
13-3-tree
‘a little tree’
However, the diminutives may not be separated from the root by any means (a
fact stressed by Bresnan and Mchombo 1995). For instance, in (7-a), a possessor
2
This is where Shona and Luganda differ: Luganda attaches diminutives directly to the root, and the
regular class marker is absent. This is also an option in Shona, alongside the one discussed in the main
text.
4
is placed in between the diminutive prefix ka and the noun with its class prefix
mu-ti. But the combination is ungrammatical, which contrasts with the behaviour
of locative prefixes shown in (5-b).3 (7-b) completes the pattern and illustrates
the fact (which we already know) that possessors may both precede and follow the
noun; but crucially, the diminutive prefix is counted as part of the noun, and the
locative one is not.
(7) a. *ka-{w/k}-angu
13-3/13-my
mu-ti
3-tree
intended: my little tree
b. {k-angu}
13-my
ka-mu-ti
13-3-tree
{k-angu}
13-my
‘my little tree’
All facts summed up, the syntactic distribution of the locative markers seems unremarkable
in comparison to familiar adpositions, and contrasts with the behaviour
of all other class markers. For some Bantu languages (siSwati, see Marten 2010),
these properties have served as the basis for the claim that locative noun class
markers indeed are prepositions.
3 Properties of Shona locatives
However, in Shona, Luganda as well as other Bantu languages, the traditional
grammars establish additional properties which differentiate these morphemes from
familiar Indo-European prepositions (and Marten 2010 shows that these do not
carry over to siSwati). These additional properties are the cause of the fact that –
as Ashton et al. (1954, 56) put it – “in the opinion of the writers and most Bantuists,
these adverbial formatives should not be looked upon as prepositions, for this is to
view a Bantu language through the medium of European grammatical concepts.
Their behaviour is most un-preposition-like [...].” What is this behaviour?
3.1 Locatives can be used as ordinary class markers
The first property is that these markers may attach directly to the root, in which
case they function as regular class markers. A couple of examples is provided
below. Most striking is perhaps the fact that the root nhu, seen in (1), can be
embedded under the locative markers, and yield the meaning of ‘place’ (with vulgar
connotations).
(8) a. mu-nhu
1-NHU
person
b. mu-kati
18-KATI
inside
c. pa-nhu
16-NHU
d. ku-nhu
17-NHU
both: place
This can be analyzed in two ways. Either as an instance of homophony (the markers
have two distinct functions), or as a straightforward underlying identity. Most
approaches (including the traditional grammars) propose the latter (see, a.o., Myers
3
The two concord options in the example show that this fact holds regardless of what concord marker
(the diminutive or the regular one) there is on the possessor.
5
1987, Bresnan 1995 and Bresnan and Mchombo 1995). The claim is that in both
cases we are looking at a single function – that of nominal classification. From that,
it follows that locatives in Bantu are radically different from locatives in standard
average Indo-European, and that the two language groups have radically different
structures for locatives.
For instance, Bresnan and Mchombo (1995) would propose the structure (9) for
our string mu-ma-poto from (3-c) ‘in the pots.’ The crucial thing here is the label
N of the node that hosts the locative marker mu ‘in,’ which is the crucial ingredient
intended to explain the differences between Bantu and standard Indo-European.
(9) NP1
Ncl
mu
NP2
ma-poto
However, looking at the facts in (8) from the second conceivable perspective –
namely as an instance of two distinct functions that are expressed by the same
marker – the conclusion is less dramatic. We keep locatives (adpositions) and
nominal classifiers as separate functions (corresponding to different categories), and
the question really is how come they can be expressed the same.
The reason why majority of researchers believe that locatives in Bantu really
are instances of nominal classification structures has to do with the fact that the
apparent nominal characteristics of Bantu locatives do not stop here.
3.2 Locatives have the distribution of nouns
Another argument that has been put forth in the literature is that locatives in
Bantu have the distribution of NPs. For example, there is a large body of literature
(XXXX) that quite convincingly argues for the claim that the locatives in Bantu
may become subjects. An example from Shona is given in (10). The remarkable
thing is the absence of a formal subject (it), required in English, as well as other
subject-like characteristics, including verbal agreement.
(10) a. mu-ki-koro
18-7-school
m-aka-zara
18-past-full
It was full in the school
b. ku-mberi
17-front
kw-e-chi-koro
17-of-7-school
kw-aka-zara
17-past-full
It was full in front of the school
As the proponents of the nominal analysis of locative class markers point out,
analysing locatives as NPs also immediately explains the external distribution of
such phrases, specifically the fact that they may behave like subjects. On the other
hand – the argument goes – if they were PPs, their subject-like properties would
be surprising.
Note, however, that the argument starts from the unquestioned assumption that
only NPs may be subjects. In principle, we could also start from the assumption
that locatives are PPs, and conclude from the data that PPs may serve as subjects.
What is the right way to go?
We do not have much to say here, and only want to draw a parallel between this
analytical issue and the much more discussed issue of dative (‘quirky’) subjects in
Icelandic. The situation is almost identical: in Icelandic, dative marked NPs can
6
serve as subjects of various verbs (see Zaenen et al. 1985 for a seminal discussion).
If we start from the assumption that only nominatives may be subjects, then these
datives must be syntactically (abstractly) nominative (as some propose). On the
other hand, others conclude that datives may be subjects, and that the notion of
subject must be dissociated from case.4
Be that as it may – our point is that pending an ultimate solution to a tricky
theoretical issue, the facts themselves (as given in (14)) are not to be taken as a
strong argument against a PP status of the locative. It may simply be the case
that PPs can be subjects in some languages, just like datives behave as subjects in
Icelandic but not elsewhere.
In a similar spirit, the literature discusses data which show that locatives inside
complex NPs behave somewhat differently than in English. In order to see that,
let us start from an ordinary bi-nominal structure. It is shown in (11-a). The
dependent noun here is mu-komana ‘boy.’ Its dependent status (its genitive case)
is signalled by the marker e, which precedes the whole noun-phrase. This e is
analogous to the English of, and it is glossed as such. The final ingredient of the
bi-nominal structure is a concord marker w- which precedes the dependency marker
e and tracks the class of the head noun.
(11) nominal dependents
a. mu-fananidzo
3-picture
w-e-mu-komana
3-of-1-boy
a picture of the boy
b. mu-fananidzo
3-picture
w-e-mu-bhuku
3-of-18-5.book
a picture in the book (lit.: a picture of in the book)
When it comes to locatives, they are integrated in the complex NP in the same
way as NPs. Taking a phrase like mu-bhuku ‘in the book’ (with mu a locative class
marker), what we can observe is that this phrase must also be preceded by e, see
(11-b). Literally, one thus have to say ‘a picture of in the book.’ Starting from
there, Bresnan (1995) develops an argument for the nominal status of the locative.
The argument goes as follows.
The Shona e is the analogue of the English of. The presence of of has been
previously explained by the reference to the so-called Case Filter (Chomsky 1981):
since nouns cannot assign Case, of appears in the structure to assign case to the
nominal dependent. English of does not appear with PPs (a picture (*of) in the
book), because PPs do not need case. However, since e appears with locatives in
Shona, it looks as if PPs in Shona did need case, which is unexpected. However,
if locatives really are NPs, then there is no mystery: e appears exactly where we
would expect it.
The reasoning clearly relies on the correctness of the Case Filter, which is an
issue we cannot go into here. Note however that Romanian – similarly to Bantu
– must have an of -like marker before some NP-internal PPs as well. The example
in (12) illustrates this (see Giurgea 2014, where the example comes from). We are
not aware that such data would be taken as evidence for the nominal status of pe,
which in other respects acts as a run-of-the-mill preposition.
(12) cartea
book.def
*(de)
of
pe
on
masa
table.def
4
See Bobaljik and Wurmbrand 2009 for an overview of the literature on both sides of the spectrum.
7
‘the book on the table’
To sum up: it has been argued that PPs in Bantu have the distribution of NPs,
and that this independently supports the conclusion that they are indeed exactly
that. However, the argument relies on the assumption that we know independently
of Bantu what the distribution of NPs and PPs is. This assumption is, we think,
too simplistic. Across various languages (Icelandic, Romanian), we see that the
material in positions which we take to be hallmark of NPs on the basis of English
may in fact differ in its category from language to language. If that is so, the
arguments in this section simply evaporate.
However, there are additional arguments for the NP status of locatives. These
appear to us more robust.
3.3 Locatives enter into concord
No matter whether we look at locative class markers as ambiguous or not, it comes
as no surprise that when the morphemes mu, ku and pa are used as regular noun
class markers, they trigger concords. The reasoning is this: grammatical behaviour
is defined by function/the underlying category, and when these markers express the
class marker function, they are expected to behave exactly like other class markers.
The following example illustrates this:
(13) mu-kati
18-inside
m-e-mba
18-of-9.house
mu-chena
18-white
m-aka-sviba
18-past-dirty
The white inside of the house was dirty
What is more telling is that we observe concord also when the same markers are
used in their adpositional function. This is true first of all for verbal agreement. In
several sentence types (the so-called locative inversion construction among them),
the location comes before the verb, and in such cases, the verbal marking reflects
the class of the locative:
(14) a. mu-ki-koro
18-7-school
m-aka-zara
18-past-full
It was full in the school
b. ku-mberi
17-front
kw-e-chi-koro
17-of-7-school
kw-aka-zara
17-past-full
It was full in front of the school
To our mind, however, the strongest concord-based argument for the nominal status
of Bantu locatives is based on NP internal concord. As a starting point, recall the
fact that possessors such as ‘my’ in ‘my hand’ agree in Bantu with the possessee.
An example is shown in (15-a), where the possessor ‘my’ (angu) agrees with the
possessee ‘hand.’ What is remarkable is that when such structures are embedded
under a locative prefix, the possessor may optionally also exhibit concord with the
locative, see (15-b).
(15) Two distinct concords possible on (a subset of) modifiers
a. n-hunzi
9-fly
i-ri
9-is
pa-
16-
[ru-oko
11-hand
rw-angu]
11-my
b. n-hunzi
9-fly
i-ri
9-is
pa-ru-oko
16-11-hand
p-angu
16-my
The fly is on my hand
8
How strong an argument for the nominal status of locatives is this? So far, not very
strong. We know that there are languages, e.g., Finnish, where locative marking is
copied by concord; see the examples in (16).
(16) Locative concord in Finnish
a. iso
big
auto
car
‘a/the big car’
b. iso-ssa
big-in
auto-ssa
car-in
‘in a/the big car’
Whatever the status of the Finnish affixes is, they have never been analyzed as
heading a syntactic node of the category N. Hence, NP-internal locative concord
on its own is not a strong sign of nominal categorial status.5
What is more relevant is that in Shona (as well as Luganda), the locative and
non-locative marker are in a complementary distribution on the modifier. This is
shown in (17-a), where the combination of the locative (pa) and the class/number
marker (rw) on the possessor leads to ungrammaticality. In Finnish, this is not so;
the number marker (i) and the case marker (ssa) are both copied by concord, and
happily co-occur, see (17-b).
(17) a. *n-hunzi
9-fly
i-ri
9-is
pa-ru-oko
16-5-hand
pa-rw-angu
16-5-my
The fly is on my hand
b. iso-i-ssa
big-pl-in
auto-i-ssa
car-pl-in
‘in (the) big cars’
What does this contrast mean? Let us start with Finnish. We will call the category
of ssa ‘Case,’ and the category of i ‘Number’. Then we can state the rule of concord
by saying that the modifier agrees for ‘Number’ and ‘Case.’ Perhaps the best way
to analyze this theoretically is to say that concord consists in reproducing a part
of the nominal functional sequence on top of the adjective, an approach pioneered
for Bantu in Taraldsen (2010).
The situation is different in Shona. If we assume (on analogy to Finnish) that
pa and ru are different categories (let us say that pa is a preposition (P) and ru a
Class marker), we must say that concord targets either P or Class, but never both
at the same time. This is a suspicious rule of concord, and the question really is
whether such a rule is a possible rule of grammar. Adopting Taraldsen’s approach,
the issue is even more pressing. It is simply impossible to state a rule for concord
which uniquely determines a part of the functional sequence that is reproduced on
the modifier.
The important conclusion is that these problems arise only because of the initial
assumption, namely that pa and ru are of a distinct category. If, on the contrary,
we assume that locative and non-locative classes are both instances of the same
category (one which the tradition calls Class), then we can easily state the facts:
5
In several studies (van Riemsdijk and Huybregts 2002, Asbury 2006, Svenonius 2010, Pantcheva
2011), arguments have been given for analysing such locative morphemes as the affixal spell-out of a
syntactically independent adpositional syntactic head. Vainikka (1993) claims that the appositional
head is silent in Finnish, and it is responsible for assigning the relevant case to the whole DP, which is
reflected by the overt morphology.
9
modifiers in Shona agree in Class – a specific part of the nominal sequence – with
the modifi-ee.
This approach gains support from additional facts. In order to show that, let
us turn to the question of what determines which concord is chosen. The answer
is that this depends on the syntactic position of the modifier. Modifiers with locative
concord are higher than the locative class marker; modifiers with non-locative
concord are lower than the locative class marker. The facts which lead to the conclusion
are given below. We have already seen that when a pre-nominal possessor
bears the non-locative concord, it comes in between the locative class prefix and
the noun, see (18-a). It cannot precede the locative prefix, see (18-b).
(18) a. pa-rw-angu
16-11-my
ru-oko
11-hand
on my hand
b. *rw-angu
11-my
pa-ru-oko
16-11-hand
on my hand
Structurally, this means that the possessor with non-locative concord is contained
inside the DP which is embedded under the locative marker, see (19). It cannot be
higher than pa, since that would incorrectly allow it to be to the left of pa.
(19) [ pa [ rw-angu [ ru oko ] ] ]
On the other hand, when the possessor shows locative concord, the facts are the
exact opposite: a pre-nominal possessor has to precede the locative marker (as
shown in (20-a)), and cannot follow (as shown in (20-b)).
(20) a. p-angu
16-my
pa-ru-oko
16-11-hand
on my hand
b. *pa-p-angu
16-16-my
ru-oko
11-hand
on my hand
This means that the possessor with the locative concord must be higher than pa,
and cannot be lower than pa:
(21) [ p-angu [ pa [ ru oko ] ] ]
With the syntactic difference between the two concords clarified, and under the
assumption that both locative and non-locative classes belong to a single category,
we end up with a clear and completely general rule for concord: the possessor agrees
in Class with the first class marker lower down.
3.4 Summary
In this section, we have seen several remarkable properties of Shona locatives, all
of which have been cited as evidence for their nominal status. However, under
scrutiny, the evidence in favour of the label N is not completely unequivocal; the
strongest piece of evidence so far appears to be the way DP internal concord works.
In the following section, we turn to two properties of the locative markers that sets
them apart from nouns, and then turn to the analysis.
10
4 How locatives differ from nouns
The properties reviewed in the preceding section motivate the traditional analysis
of locative class markers as items which are essentially identical to ordinary
class markers. Under this approach, it becomes tempting to assimilate the constructions
completely, and propose that just like ordinary class markers attach to
nouns, the locative class markers do too, just that the noun is silent. This leads
to the consequence that locatives really are bi-nominal structures (perhaps with a
silent PLACE, Carstens 1997). However, there are at least two differences between
locatives and regular bi-nominal structures.
4.1 The absence of a ‘linker’
The first distinction is that in ordinary bi-nominal structures, a linker/genitive
marker e (analogous to of ) is present between the two nouns, see (11-a), repeated
in (22-a).
(22) a. mu-fananidzo
3-picture
w-e-mu-komana
3-of-1-boy
a picture of the boy
b. *mu-PLACE
18-place
m-e-ma-poto
18-of-6-pot
intended: in the pots
c. mu-ma-poto
18-6-pots
in the pots
If we try to construct an example along the lines of the bi-nominal example, just
substituting the head noun by the hypothesised silent noun PLACE, we end up
with an ungrammatical string (22-b). The comparison with (22-c) reveals that the
problem is the presence of the linker e with the accompanying concord marker. Put
simply, one cannot say inside of the pots in Shona, one has to say inside the pots.
4.2 Only a subset of modifiers is allowed
In (15), we have seen that it is possible for certain modifiers to bear locative concord
even when the locative noun class markers are used as spatial markers. We repeat
the relevant example below:
(23) n-hunzi
9-fly
i-ri
9-is
pa-ru-oko
16-11-hand
p-angu
16-my
The fly is on my hand
We have also concluded on the basis of their order that they sit higher in the
functional spine than pa. Myers (1987) and Bresnan and Mchombo 1995 have
argued that in order to accommodate such modifiers in the structure, we need
to say that the locative marker heads a regular nominal projection, where such
modifiers are added.
However, this argument has a double edge, because the set of such modifiers
is restricted. For example, while part-whole relations such as (23) are fine with a
locative concord, ordinary possessors cannot bear it, see (24). Specifically, (24-b)
shows that only non-locative concord is possible here.
11
(24) Possessors
a. n-hunzi
9-fly
i-ri
9-be
pa-mu-komana
on-1-boy
w-angu
1-my
The fly is on my boy
b. *n-hunzi
9-fly
i-ri
9-be
pa-mu-komana
on-1-boy
pa-ngu
16-my
Similarly adjectives may not have locative concord. The following example illustrates
that.
(25) a. *pa-mu-komana
16-1-boy
pa-chena
1-white
b. pa-mu-komana
16-1-boy
mu-chena
1-white
‘on the white boy’
The example (25-a) does not have any of the two conceivable interpretations: it
cannot mean either ‘on the white surface of the boy’ or ‘on the white boy.’ It is
simply unacceptable. (25-b) shows that the non-locative concord is fine.
This should be compared against the fact that there is no general ban on adjectival
concord with the locative class markers. When the class markers are used in
a non-spatial meaning (to refer to a part of an object, and not to a space defined
with respect to a part of an object), adjectival modification becomes possible:
(26) mu-kati
18-inside
m-e-mba
18-of-9.house
mu-chena
18-white
m-aka-sviba
18-past-dirty
The white inside of the house was dirty
These facts taken together are problematic for the proposal that the locatives
project a second nominal projection where nominal modifiers may freely be added.
Clearly, something more is going on in here.
The facts are highly reminiscent of Svenonius’ (2006) discussion of English. He
points out that “[a] distinction between the [non-spatial] use and the [spatial] use
of front is adjectival modification, which is only acceptable with [non-spatial use].”
(27) a. There was a kangaroo in the smashed-up front of the car.
b. *There was a kangaroo in smashed-up front of the car.
For Bantu, we may adopt the same conclusion, only replacing ‘front’ in Svenonius’
quote by ‘class markers’: a distinction between the non-spatial use and the spatial
use of class markers is adjectival modification, which is only acceptable with nonspatial
use.
Hence, a potential conclusion to be drawn here is that the Bantu locatives fall
into the large set of expressions found across various languages which Svenonius
calls AxParts; expressions that have some – but not all – nominal characteristics.
5 The analysis
With the basic facts laid out, let us turn to the analysis. What we found in general
terms is that locative class prefixes have some nominal properties, but at the same
time differ from nouns. As highlighted at the outset, the current literature contains
two classes of approaches. Svenonius (2006) proposes that since such expressions
cannot be identified as fully identical to any of the existing categories, they belong
12
to a special category which he calls AxPart. AxPart is understood as a special
category label and occupies a slot in the functional sequence between the DP and a
prototypical adpositional region (what we will call here PlaceP, following Svenonius’
terminology). Nothing else said, it seems to us that this approach deals easily with
the fact that such expressions differ from nouns (as well as adpositions), but leaves
the overlap in properties with nouns accidental. There is no guarantee from the
label AxPart itself that expressions instantiating it will trigger concord, etc.
The second approach is to say that there is a second noun in the structure. For
some, the marker itself is the second noun (for Bantu: Myers 1987, Bresnan and
Mchombo 1995), or there is a silent noun PLACE, and the marker is related to it
in one way or another (for Bantu: Carstens 1997). The second noun is responsible
for the nominal properties. However, as far as we can see, the fact that the extra
noun cannot be modified by adjectives or possessors is left unexplained, as well as
the lack of the overt linker.
All summed up, we stand in a situation where each analysis has complementary
merits and drawbacks. What we would like to propose here is a third type of analysis,
which keeps the noun in the structure, but adds to this that AxPart is indeed a
special category (as argued by Svenonius). However, it is not a special category in
the sense that it corresponds to a specific label in the functional sequence. Instead,
we suggest that it is a special type of a lexical item, namely one that combines
several labels (both N and P). In particular, we propose that ‘AxParts’ (of which
the locative classes are an instance) are stored in the lexicon as expressions that
may spell out a whole phrase that contains minimally a nominal terminal, a class
node, and an adpositional terminal:6
(28) AxPart ⇔ PlaceP
Place ClassP
Class NP
PLACE
The constituent has an NP at the bottom which corresponds to the silent PLACE
of various approaches. However, there is no unpronounced terminal; the NP node
corresponding to PLACE is pronounced as a part of a larger structure.
We believe that this analysis has the potential to capture both the nominal
and non-nominal properties, as we argue below. In the following sub-sections, we
review the relevant properties one by one and show how these can be explained if
one adopts the hypothesis (28).
5.1 The lack of the linker
The lack of the linker follows from the way expressions such as (28) are integrated
into the structure. Our assumptions here follow traditional Cartography (Cinque
and Rizzi 2010) in that we assume a cross-linguistically fixed sequence of labels
(the functional sequence), which is responsible for the underlying order of heads
and their specifiers. For our purposes, only a small subset of all the categories will
be relevant. They are given in (29).
6
For the theory of phrasal spell out assumed here, see Starke (2009), Caha (2009), Caha (2013).
13
(29) PlaceP
Place NumP
Num ClP
Cl NP
...
First of all, there is the ‘lexical head’ N at the bottom, with nominal class and
number higher up in the structure in that order. Another relevant projection will
be called Place, after Koopman (2000) and Svenonius (2010). Following further
Svenonius (2008b), we assume that the role of this head is to introduce a region
relative to the noun that it takes as a complement.7
As in traditional cartography, we assume that these heads may have specifiers
which have to ‘match’ the particular head. For example, various types of adjectives
have been proposed to be generated in the Spec of their corresponding projection
(an approach pioneered in Cinque 1999 for adverbs, and adopted for adjectives by
Scott 2002 and Svenonius 2008a, a.o.).
Putting these assumptions together, the proposal for Bantu locative markers is
that they are similar to adjectives in being introduced as the Spec of their corresponding
category, i.e., in the Spec of PlaceP. Their role is to specify the precise
content of the relation (interior, surface, surroundings) between the reference object
and the region constructed on its basis. The following tree depicts the proposal.
(30) PlaceP
PlaceP ⇔ mu
Place ClassP
Class NP
PLACE
PlaceP
Place NumP
Num ClP
Cl NP
...
Boldface of the label Place indicates the assumed Spec-Head matching, which allows
the Place head to remain silent, as in Koopman (1996) or Starke (2004). There is
no matching beyond the top node; the two NPs are independent of each other.
The Spec-Head nature of the relationship between PLACE and the second noun
on the right branch is an alternative to an ordinary bi-nominal complement structure
as proposed for Bantu by Myers (1987), Bresnan and Mchombo (1995) and
7
Even higher up is a Path region, which encodes distinctions related to how the region denoted by
PlaceP is to be understood, i.e., whether it is a static location, or whether it is a source/goal of a motion
event. In essence, the Place region corresponds to what Kracht (2002) calls the localizer function, and
the Path region to what he calls the modalizer function, although we assume that both Place and Path
may be more complex (see, e.g., Svenonius 2010 for an elaborated Place region, and Pantcheva 2011 for
an elaborated Path region of the functional sequence).
14
Carstens (1997), see (9).8 We propose that this alternative structure is the reason
for the absence of a linker, which we understand as associated to complement structures.
Importantly, the absence of the linker does not mean the absence of a second
noun (PLACE); it’s just that the second noun is generated inside the complex Spec.
Note as well that ‘ordinary’ nouns cannot be used in this manner, i.e., as a Spec
of a functional projection Place. This option is dependent on Spec-Head matching,
which is only an option if the noun spells out Place in addition to the nominal
projections. For instance, the following structure is illicit, because the boldfaced
labels do not match:
(31) illicit structure: PlaceP
NP
...
PlaceP
Place NumP
Num ClP
Cl NP
...
Since ordinary nouns cannot be generated as Specs of functional heads, they have
to enter into regular complement structures, which (by assumption) lead to the
occurrence of the linker.
In other words, we have now explained both why locative prefixes do not have
a linker, and why ordinary nouns do. The source of this difference, however, is not
that the former is not a noun and the latter is. The difference is in how the second
noun is introduced in the structure. If the noun’s the lexical entry has a functional
head in addition to the nominal part, it may be merged as a Spec, and avoid the
need for a linker.
5.2 Locatives can be used as noun class markers
Recall from the discussion above that locative class markers can also attach directly
to roots, and serve as ordinary class markers:
(32) a. mu-nhu
1-NHU
person
b. mu-kati
18-KATI
inside
c. pa-nhu
16-NHU
d. ku-nhu
17-NHU
both: place
How can we account for that? The analysis has two ingredients. The first one is
the fact that the phrasal spell-out model we use here (see Starke 2009, Caha 2013)
contains a matching principle that allows a single lexical item to spell-out (pronounce)
multiple structures (i.e., more than just the one specified in the lexicon).
The principle is given below in (33); see the literature quoted for an independent
motivation for this particular spell-out rule.
8
All other silent PLACE approaches that we cite above seem to us to assume the same, namely that
the Ground is introduced as the possessor of the silent PLACE. We do not adopt this proposal here.
15
(33) The Superset Principle (Starke 2009): A lexically stored tree matches a
syntactic node iff the lexically stored tree contains the syntactic node.
The principle allows that a locative class prefix with the entry (34-a), repeated
from above, can also spell-out syntactic structures such as (34-b), because (34-b)
is contained in the entry (34-a).
(34) a. PlaceP
Place ClP
Cl NP
PLACE
b. ClP
Cl NP
PLACE
The second ingredient of the analysis consists in wider application of the principles
of Spec-Head matching that we have sketched above. Recall that our working
hypothesis here is that a particular functional head may have a complex Spec, and
remain silent. We have applied this idea to the Place node, but the very same
principles may also apply at different levels of structure. Among others, we can
generate structures such as the following:
(35) Class
ClassP
Class NP
PLACE
Class
Class NP
...
In this structure, the complex Spec indicates the particular content of Class, and
the whole structure corresponds to an ordinary noun-class construction. Recall that
beyond the Class node, no matching is required. So the right branch NP in the
structure (35) may be spelled out by the root nhu, and classified as a particular
concept (ANIMATE, THING, PLACE) by the class marker, which spells out the
left branch. The boldfaced head Class is left unpronounced, a by-product of the
Spec-Head configuration.
Putting now the classifier structure (35) together with the proposal that complex
locative markers may ‘shrink’, as in (34-b), we have an account of the fact that
locative class markers may also function as non-locative class markers.
In general terms, shrinking gives us an account of the fact that ‘AxParts’ (like
front) have a locative use and a nominal use, where the nominal use is always a
by-product of ‘shrinking,’ predicted by the Superset Principle.
Let us also mention here that for regular class markers, we assume that they
have lexical entries like the ones in (36).
(36) a. singular inanimate ⇔ ClassP
Class NP
THING
16
b. plural inanimate ⇔ NumP
Num ClassP
Class NP
THING
This allows them to function as class markers in structures such as (35), but not
in locative structures (they lack the Place node, crucial for the possibility to be
used as Spec,PlaceP). We hasten to add that this is a tentative proposal created to
have something concrete to start from; for instance, the object described here by
simple THING may correspond to a complex structure which includes syntactically
relevant heads such as COUNT, INDIVIDUATED, etc. Similarly, the animate class
markers (classes 1 and 2) may include a decomposed head ANIM, in addition to
those mentioned above, etc.
The reason why we mention such entries here is to show explicitly what kind of
relationship our analysis proposes for the relation between locative and non-locative
class. It is not identity, but overlap (they have some nodes in common). The nodes
which are not shared help us explain the differences (ordinary class markers cannot
occur in Spec,PlaceP). The shared nodes are important in order to understand the
parts of the grammar where locative and non-locative classes pattern alike. One
such domain is concord.
5.3 Concord
As highlighted in section 3.3, locative class markers may be borne by modifiers
which are located higher than the locative, as in (37-a). Modifiers which are lower
than the locative bear the class of the noun, as in (37-b). The two markers cannot
be copied simultaneously, (37-c).
(37) a. [ p-angu [ pa [ ru oko ] ] ]
b. [ pa [ rw-angu [ ru oko ] ] ]
c. *pa-ru-oko pa-rw-angu
As discussed in section 3.3, the complementary distribution seen in (37-c) suggests
that whatever we call the category that is targeted by concord, both locative and
non-locative classes have that category. The present account allows us to formalise
this insight. First of all, we assume that concord replicates a ClassP projection
(potentially decomposed, as in Taraldsen 2010) on top of the modifier, and the
locative marker may ‘shrink’ to spell out whatever is copied.
(38) Class
ClassP
Class NP
PLACE
Class
Class modifier
...
The particular Class which is present on top of the modifier is determined by two
principles. In a given structure, the modifiers carry the class of the noun which they
17
(i) c-command and (ii) which is closest to them (i.e., whose extended projection –
as revealed by precedence – a symmetrically c-commands the other noun).
For example, in the following structure, Modifier 1 agrees with N1 (the modifier
c-commands both N1 and N2, but N1 is chosen because it is closer than N2).
Modifier 2 agrees with N2 (the modifier does not c-command N1).
(39)
Modifier 1
N1
...
Modifier 2
... N2
...
This mechanics then ensures that the result we get is the one shown in (37). But
most importantly, the proposal explains both the fact that we get a complementary
distribution between locative and non-locative concord markers on the modifier, but
not on the noun. The crucial ingredient of this analysis is the notion of overlap.
Locative and non-locative class share a part of their structure, and this shared part
is targeted by concord. As a consequence, they show complementary distribution,
even though they are not identical. On the noun, the locative marker may stack,
because of its extra Place specification: it uses the part which is not shared.
In the remainder of this section, we would like to point out that the behaviour of
concord in Shona (as well as Luganda) locatives is completely parallel to what can
be observed elsewhere in similar type of non-complement bi-nominal constructions.
Consider for example the behaviour of measure constructions (a cup of coffee) in
Swedish. The starting point is the observation that there are two ways of conveying
this meaning. One, by a regular bi-nominal construction with a preposition, where
N1 takes N2 as a complement (see (40-a)). Two, by a construction without a
preposition (as in (40-b)). The construction (40-a) appears to be an ordinary
bi-nominal construction, and each noun may have its own determiner, as (41-a)
shows. The construction without the preposition behaves differently, and only a
single determiner is allowed, see (41-b).
(40) a. en
a
kopp
cup
med
with
kaffe
coffee
b. en
a
kopp
cup
kaffe
coffee
‘a cup of coffee’
(41) a. en
a
kop
cup
med
with
det
that
kaffet
coffee
a cup of that coffee
b. *en
a
kopp
cup
det
that
kaffet
coffee
The construction without the preposition is, we believe, analogous in abstract structural
aspects to the Bantu locative construction. What we have is a special type
of a quantity noun, which can be analysed as a Spec of a dedicated functional
head. This explains both the lack of the preposition and the fact that only a single
determiner is allowed (because we are looking at a single extended functional
sequence).
The relevant point is that the behaviour of concord in the relevant construction
is strongly reminiscent of Bantu. Whatever modifier c-commands both kopp ‘cup’
and kaffe ‘coffee’ agrees with ‘cup’ (because it is the closest noun). Whatever
modifier c-commands only kaffe ‘coffee,’ tracks the class of ‘coffee.’
This is easy to show since kopp and kaffe differ in their gender. In (42-a,b), the
modifiers c-command kopp, and reflect its nominal class called common gender in
18
Swedish grammar. In (42-c), the modifier c-commnads only kaffe and agrees with
it, showing the neuter gender instead.
(42) controls concord like ordinary nouns
a. en/*ett
a-cg/a-neut
kopp
cup-cg
kaffe
coffee-neut
‘a cup of coffee’
b. en
a
{varm/*
warm-cg
varmt}
warm-neut
kopp
cup-cg
kaffee
coffee-neut
c. en
a
kopp
cup-cg
{*varm/
warm-cg
varmt}
warm-neut
kaffee
coffee-neut
both b and c: ‘a cup of warm coffee’
To conclude: the Spec-Head structure proposed here in (30) contains two nouns,
and two distinct class projections. Modifiers that appear in this structure agree in
Class with one of the two nouns, subject to c-command and locality. Because of
the fact that locative and non-locative class markers overlap for Class, they are in
complementary distribution on the modifier. This can be so even though they are
not in complementary distribution on the noun, where they stack one on top of the
other. Stacking here is allowed because unlike concord, stacking on nouns involves
crucially the Place head, for which there is no overlap.
We have further shown that understood this way, the concord phenomenon seen
in Bantu is perfectly analogous to the behaviour of structurally similar examples
in unrelated languages.
5.4 No adjectival modification
The final observation to be tackled here is the fact that when locative class markers
are used as ordinary class markers (when they attach to the root and together with
the root denote a thing), such an NP may be modified by adjectives. When they
act as locative markers, adjectives are excluded. Why is this so?
In order to answer this question, we must place adjectives in the functional
sequence. Before we get more specific about that, we would like to say that the
only thing which is really crucial for us is that they are introduced somewhere
between the NP and the Place head; if that is so, the effect described in the first
paragraph can be derived. That said, we will now suggest one particular solution
along these lines.
What we will be assuming is that since the class marker and the root form a
constituent for idiomatic interpretation and other processes, adjectives are generated
outside of that constituent, in the specifier of a dedicated head F, see (43).
We are not explicitly representing concord on the adjective, but we assume it is
base-generated as a part of the extended AP in Spec,FP, and subject to a matching
requirement respecting the requirements of c-command and locality formulated
earlier.
19
(43) PlaceP
Place FP
AP
...
F’
F ClassP
Class NP
...
We assume further that ClassP moves across FP, which yields the surface position
of the adjective after the noun. This movement is not represented in the tree.
With this issue clarified, let us now come back to our original structure for
locatives, given first in (30). In this structure, it is fine to modify the right branch
NP1 by adjectives. This requires that an additional head F is added in between
PlaceP and ClassP, hosting an adjective in its Spec, see (44). Just for illustration,
we have filled in the structure by lexical material.
(44) PlaceP
PlaceP ⇔ mu
Place ClassP
Class NP-2
PLACE
PlaceP
Place FP
AP
ma-chena
white
F’
F ClassP
Class NP-1
ma-poto
pots
However, it is impossible to modify NP2 (the silent PLACE) by an adjective. Such
an adjective would have to be introduced inside the constituent on the left branch,
in the specifier of a head that would come just above the class node associated to
the NP2 PLACE:
20
(45) PlaceP
PlaceP ⇔ mu
Place FP
AP
...
F’
F ClassP
Class NP-2
PLACE
PlaceP
Place ClassP
Class NP1
...
However, if such an adjective were present, it would no longer be possible to
use mu as the spell out the whole PlaceP structure. That is because the lexical
entry of mu, given in (29), does not contain the F head and its Spec. The Superset
Principle, however, requires that for mu to spell out the PlaceP in (45), it would
have to contain these elements. Hence, the conclusion is that it is impossibility to
modify the silent PLACE by adjectives. The reason for that is that the PLACE is
silent only because it is spelled out by a phrasal item, and it cannot remain silent
(be spelled out by a phrasal item) when it is modified by an adjective.
5.5 Summary
This section has introduced a new proposal for the structure of Bantu locatives. We
combined the proposal that there is a second noun in the structure, PLACE, with
the observations that the Bantu locative class markers are special items that stand
somewhere in between the prototypical categories N and P, strongly resembling a
category labelled AxPart by Svenonius 2006.
The noun we have in the structure is the source of the nominal properties of
the locative, it triggers concord on c-commanding modifiers and it allows for an
account of non-locative uses. We remain agnostic as to how this noun influences
the external distribution of the locatives, see section 3.2, as we are not convinced
that the facts show what they are supposed to show.
Despite the presence of the second noun, we are not looking at an ordinary binominal
structure where one noun is the complement of the other noun. The role
of PLACE in the structure may perhaps be best described as an ‘auxiliary noun’
function, on analogy with ‘auxiliary verbs.’ Its nominal part – detectable by various
tests – only appears in the structure in order to support functional material.
The novel aspect of this proposal is that the noun does not correspond to a
silent terminal; it is silent because it is spelled out as a part of a larger constituent.
The argument in favour of this hypothesis is that the phrasal nature of the lexical
item helps us shed light on the properties in which locative class markers differ from
ordinary nouns: they do not take the linker, and they do not allow certain types
of modification. These issues have remained unexplained under the original silent
PLACE hypothesis, as far as we are aware.
There is one question that remains. So far, we have been looking at the function
of the noun class prefixes as essentially modifiers of an NP. In this function, they
21
sit on the left branch in a Spec of a particular noun. However, the proposal is that
these morphemes include a noun at the bottom, and could in principle be also used
beyond the modification function as self standing nouns. Just like most auxiliary
verbs can also be used as main verbs, we would expect these auxiliary nouns to
have some use as regular nouns.
This expectation is based on the Superset principle. Since lexical items may
shrink, these morphemes may spell out just the NP (or ClassP) and act as regular
nouns. For instance, it should be fine to use mu in order to express something like
the notion of ‘place’ or ‘interior’ or whatever its meaning is. On the first blush,
this does not seem to be the case. However, there is interesting evidence that this
prediction may in fact be correct.
6 Luganda
In order to shed some light on this issue, we are going to switch to a different
Bantu langauge, Luganda. In all abstract aspects discussed up to now, Luganda
behaves just like Shona. However, there is one thing that distinguishes the noun
class morphology in the two languages, and this is what will become relevant here.
6.1 The initial vowel
The difference is that noun class markers in Luganda are generally bi-morphemic.
The initial part of the class marker corresponds to a single vowel, variously called
the pre-prefix, the augment, or the initial vowel. This marker is absent in Shona.
The second part of a class marker is called the prefix, and it is a CV marker very
similar to the Shona prefixes:
(46) Luganda
a. o-mu-ntu
1-1-NTU
a person
b. a-ba-ntu
2-2-NTU
people
c. e-ki-ntu
7-7-NTU
thing
d. e-bi-ntu
8-8-NTU
things
The shape of the pre-prefix is determined by the prefix; the pre-prefix corresponds
to the coalescence of A with the vowel of the prefix (u+a=o, i+a=e,
a+a=a).
There is evidence that the pre-prefix is a separate morpheme, because it is not
always present on the noun. Its presence/absence is controlled by factors which are
notoriously complex (see, e.g., Taraldsen 2010). However, the intuition many follow
is that the initial vowel is similar to a determiner. For example, it must be present
when the noun is the subject of the sentence, as in (47-a). Here, the adjective is
interpreted as a predicate of the subject. When the pre-prefix is missing on the
noun, as in (47-b), the noun is interpreted as a predicate. The adjective is then
modifying the predicate, and the subject is pro-dropped.
(47) a. e-ki-kopo
7-7-cup
ki-nene
7-big
The cup is big.
b. ki-kopo
7-cup
ki-nene
7-big
22
It’s a big cup.
Hence, we will proceed under the assumption that the initial vowel corresponds to a
separate head in the syntactic tree, and resides somewhere around the place where
definiteness and specificity are determined. This idea is further supported by the
behaviour of objects under negation. Specifically, the initial vowel is missing on
non-specific objects in the scope of negation, as in (48-a), but it is present when
the object is interpreted as specific (48-b).
(48) a. ss-a-yas-izza
neg.1-past-break-perf
ki-kopo
7-cup
I didn’t break any cup
b. ss-a-ky-as-izza
neg.1sg-past-7oc-break-perf
e-ki-kopo
7-7-cup
I didn’t break the cup
What is interesting is that the locative classes in Luganda lack the initial vowel.
In (49-a), a locative phrase acts as a subject. In this position, the initial vowel is
expected to appear. However, (49-a) has only mu with no apparent initial vowel,
and its inclusion, as in (49-b), yields ungrammaticality.
(49) a. mu-ki-no
18-7-this
e-ki-kopo
7-7-cup
mu-ddugala
18-dirty
It is dirty inside the cup (literally in the cup is dirty)
b. *o-mu-ki-no
18-18-7-cup
e-ki-kopo
7-7-cup
mu-ddugala
18-dirty
The absence of the initial vowel is characteristic for the locative classes. In informant
sessions, we have tested locatives in all contexts where a regular noun would
have to have an initial vowel, and – with one exception – the locative phrase is
not allowed to have it. In fact, sequences where the locative class marker appears
with an initial vowel (such as the one given in (50)) are judged unacceptable in any
context:
(50) *o-mu-ki-kopo
18-18-7-cup
intended: in the cup
Note that this contributes to the list of differences between locative and non-locative
classes, i.e., to the list of problems for a theory according to which locative constructions
in Bantu are an instance of an ordinary bi-nominal complement structue
(as in Myers 1987 or Bresnan and Mchombo 1995).
Let us now turn to the theoretical interpretation of this fact. Given the plausible
analysis of initial vowels as members of the determiner system, the fact that locative
class prefixes lack the initial vowel can be reformulated as: locatives do not accept
determiners. Seen this way, this only adds up to the expected profile of an AxPart
type of category: AxParts cannot be modified by determiners. As Svenonius (2006)
convincingly argues for (51), only the example without the determiner (i.e., (51-a))
is an AxPart construction, (51-b) features an ordinary nominal use of front.
(51) a. There was a kangaroo in front of the car.
b. There was a kangaroo in the front of the car.
All summed up, locative classes in Luganda do not have the initial vowel because
23
they are AxParts, and AxParts cannot be modified by determiners, which the initial
vowels are an instance of.
As for the theoretical reasons why initial vowels are excluded, recall the discussion
in section 5.4. Here we have proposed that AxParts (in their locative use)
spell out a whole PlaceP, and nothing that would have to be introduced inside
that PlaceP may appear. Since determiners (like adjectives) are generated inside
PlaceP, they are incompatible with the AxPart use of the prefix.
6.2 Class markers in a nominal use
We are now in a position to return to the question from the end of section 5:
can mu and its kin be used as regular nouns as the proposal predicts, or are these
morphemes only ever attested as modifiers (‘auxiliary nouns’) that occupy the Spec
of a particular projection?
We will answer this question by introducing a puzzling construction where paradoxically
we do find an initial vowel with a locative class marker. We will then argue
that the best way to analyse this construction is by assuming the existence of a
noun, meaning PLACE, and pronounced as mu.
Let us start by mentioning an example where an initial vowels appear with a
locative class marker; it is shown in (52). We found this construction by trying to
elicit examples where mu ‘in’ appears with the initial vowel o, yielding o-mu.
(52) n-jagala
1-love
o-mw-a-Ssembatya
18-18-of-Ssembatya
I like it at Ssembatya’s place
The example itself has o-mw, rather than o-mu, but this is a predictable phonological
change in the context of a following vowel (a). As the interpretation suggests,
this is a somewhat irregular locative construction. Ssembatya here is not the
Ground (the reference object with respect to which the location is determined), it
is the possessor of the Ground. This is in line with the morphosyntax: Ssembatya
is preceded by the genitive like marker a ‘of.’
Hence, it looks like we are looking at an ellipsis type of construction, where the
Ground NP (‘place’) is elided, and only its possessor remains. Let me then look
at ellipsis with possessor-like remnants in Luganda in more detail in order to see
whether this is the correct way of looking at the construction.
The first sentence we will consider in this context is given in (53-a); it is a noneliptical
sentence on the basis of which the ellipsis will be constructed. The head
noun ‘water’ belongs to class 6, and has a bi-morphemic class marker a-ma. The
head noun has a locative modifier (‘in the well’), which (recall) is embedded in the
DP just like possessors. The possessive marker in Luganda is a, but we do not see it
in (53-a), because of a regular phonological process which elides this a in contexts
where another vowel follows. It is replaced by the sign ’ in the orthography.
(53) a. a-ma-zzi
6-6-water
g-’-oku-lu-zzi
6-of-17-11-well
lw-affe
11-our
g’a-kalidde
6past-dry.perf
‘The water in our well dried.’
b. a-ma-zzi
6-6-water
g-’-oku-lu-zzi
6-of-17-11-well
lw-affe
11-our
nago
also
g’a-kalidde
6past-dry.perf
‘The one in our well dried as well.
c. *a-ma-zzi
6-6-water
g-’-oku-lu-zzi
6-of-17-11-well
lw-affe
11-our
nago
also
g’a-kalidde
6-past-dry.perf
24
The important thing is how ellipsis is constructed on the basis of this sentence.
The example in (53-b) shows that NP ellipsis consists in deletion of the root plus
the prefix, but leaving the initial vowel present. The initial vowel cliticises on the
genitive remnant, so the orthography of (53-b) would be ag’okuluzzi. As (53-c)
shows, it is impossible to eliminate the initial vowel, leaving only the possessor.
Returning to the sentence in (52), and parsing it on analogy with (53-b), the
analysis must be that o is the initial vowel of an elided noun, which belongs to class
18 and means ‘place,’ and mw is a concord on the possessor which reflects the class
of the elided head:
(54) n-jagala
1-love
o-mu-N
18-18-place
mw-a-Ssembatya
18-of-Ssembatya
I like it at Ssembatya’s place
The problem is that there is no such noun in Luganda. The nouns that would
semantically fit in the non-elliptical construction (such as a-wa-ntu ‘place’ or a-waka
‘home’) belong to class 16, rather than 18. Hence, we have a problem in finding
the non-elliptical counterpart of this structure.
The point of all this is that when we put this problem together with the problem
identified in section 5, both problems are solved. The problem at the end of section
5 was that we predicted that locative class markers like the class 18 mu should have
a use as a regular noun meaning PLACE. The problem we have in (54) is that we
lack a class 18 noun meaning PLACE. The natural conclusion is that mu is the
noun we are looking for.
The idea as to why the non-elliptical counterpart is unattested is this: the noun
would be pronounced just as mu. But this is superfluous, since this material can
be without any problems recovered from the concord marker mu. Hence, economy
dictates that ellipsis obligatorily applies.
6.3 The demonstrative ‘here’
With this conclusion in hand, let us turn to demonstratives meaning ‘here.’ Both
in Luganda as well as Shona, these look like the regular demonstrative ‘this,’ just
bearing the locative concord, and with no accompanying noun.
To illustrate that, let us first look at (55-a). Here we see an ordinary demonstrative,
agreeing in class with the head noun. (55-b) shows that the head noun can be
elided, leaving only the demonstrative. (55-c,d) shows that when the demonstrative
root is augmented by a locative concord marker, the meaning is ‘here.’
(55) a. ki-no
7-dem
eki-kopo
7-cup
ki-nene
7-big
This cup is big
b. ki-no
7-dem
ki-nene
7-big
This one (thing of class 7) is big
c. m-beera
I-live
wa-no
16-dem
I live here.
d. eki-kopo
7-cup
ki-ri
7-cop
mu-no
18-dem
The cup is in here
The point is that (55-c,d) very much look like ellipsis structures with the head noun
25
omitted, as in (55-b). Kayne (2004) in fact argues that also in English, here is a
modifier in a complex NP, where the head noun is a silent PLACE. If that is so,
then we once again have the need for a particular type of noun to fit in the structure
(55-d). But as before, there is no such noun safe the one we have postulated to
reside inside the locative class marker.
Hence, it seems that our prediction that locative class markers will have a purely
nominal use is in fact correct, but it can only be seen indirectly through ellipsis
structures.
7 Conclusion
We have argued that Bantu locative class markers are an instantiation of the category
AxPart, and we have proposed a novel analysis of this recently established
category. Specifically, we have suggested that AxParts do not correspond to a specific
label in the functional sequence, but that they are a specific type of a lexical
item that combines several labels. We have shown that under this assumption,
their complex behaviour can be explained.
If correct, this line of research suggests that perhaps some of the labels that
have been postulated in the description of various phenomena may in fact be reinterpreted
as a combination of several independently needed categories. At the
same time, the gist of the Cartographic enterprise remains intact; we are still working
under the assumption that there is a rich and cross-linguistically rigid sequence
of heads and specifiers.
The general lesson for future research is that phrasal spell-out increases our
descriptive power beyond the traditional view, which works under the assumption
that a morpheme (or a class of morphemes) that behaves unlike all other existing
morphemes necessarily means that a new category label (corresponding to those
morphemes) must be postulated in the functional spine.
References
Aboh, Enoch O. 2010. The P route. In The Cartography of Syntactic Structure,
vol.6, edited by Guglielmo Cinque and Luigi Rizzi, pp. 225 – 260. Oxford University
Press, Oxford.
Adger, David. 2012. A syntax of substance. Ms., Queen Mary, London.
Asbury, Anna. 2006. Towards a typology of morphological case. In Nordlyd 33.1:
Special issue on Adpositions, edited by Peter Svenonius, pp. 120 – 148. University
of Tromsø, Tromsø. Available at www.ub.uit.no/munin/nordlyd/.
Ashton, E.O., E. M. K. Mulira, E. G. M. Ndawula, and A.N. Tucker. 1954. A
Luganda Grammar. Longmans, Green and Co, London, New York, Toronto.
Bobaljik, Jonathan David and Susi Wurmbrand. 2009. Case in GB/Minimalism. In
The Oxford Handbook of Case, edited by Andrej Malchukov and Andrew Spencer,
pp. 44–58. OUP, Oxford.
Botwinik-Rotem, Irena. 2008. Why are they different? An exploration of Hebrew
locative PPs. In Syntax and semantics of spatial P, edited by Anna Asbury,
Jakub Dotlaˇcil, Berit Gehrke, and Rick Nouwen, p. 331364. John Benjamins,
Amsterdam.
26
Bresnan, Joan. 1995. Category mismatches. In Theoretical Approaches to African
Linguistics, edited by Akiniyi Akinlabi, pp. 19–45. Africa World Press, Trenton,
NJ.
Bresnan, Joan and Sam A. Mchombo. 1995. The Lexical Integrity Principle: evidence
from Bantu. Natural Language and Linguistic Theory 13: 181–254.
Caha, Pavel. 2009. The Nanosyntax of Case. Ph.D. thesis, CASTL, University of
Tromsø.
Caha, Pavel. 2013. Explaining the structure of case paradigms through the mechanisms
of Nanosyntax. Natural Language and Linguistic Theory 31: 1015–1066.
Carstens, Vicki. 1997. Empty nouns in Bantu locatives. The Linguistic Review 14:
361–410.
Chomsky, Noam. 1981. Lectures on Government and Binding. Foris, Dordrecht.
Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective.
Oxford University Press, New York.
Cinque, Guglielmo. 2010. Mapping spatial PPs: an introduction. In The Cartography
of Syntactic Structure, vol.6, edited by Guglielmo Cinque and Luigi Rizzi,
pp. 3–25. Oxford University Press, New York.
Cinque, Guglielmo and Luigi Rizzi. 2010. The cartography of syntactic structures.
In The Oxford handbook of linguistic analysis, edited by Bernd Heine and Heiko
Narrog, pp. 51–65. Oxford University Press, Oxford.
Ferrari-Bridgers, Franca. 2008. A unified syntactic analysis of Italian and Luganda
nouns. In The Bantu-Romance connection: A comparative investigation of verbal
agreement, DPs, and information structure, edited by Ccile de Cat and Katherine
Demuth, pp. 239–258. John Benjamins, Amsterdam.
Giurgea, Ion Tudor. 2014. Romanian adnominal locative PPs and argument structure.
Talk at Olinco 2014, Olomouc.
Kayne, Richard S. 2004. Here and there. In Lexique, Syntaxe et LexiqueGrammaire/Syntax,
Lexis & Lexicon-Grammar: Papers in Honor of Maurice
Gross, edited by C. Lecl`ere, E. Laporte, M. Piot, and M. Silberztein, pp. 253–
275. John Benjamins, Amsterdam.
Koopman, Hilda. 1996. The spec head configuration. In Syntax at Sunset, edited by
Edward Garrett and Felicia Lee, pp. 37–64. Department of Linguistics, UCLA,
Los Angeles.
Koopman, Hilda. 2000. Prepositions, postpositions, circumpositions, and particles.
In The Syntax of Specifiers and Heads, edited by Hilda Koopman, pp. 204–260.
Routledge, London.
Kracht, Marcus. 2002. On the semantics of locatives. Linguistics and Philosophy
25: 157–232.
Marten, Lutz. 2010. The great siSwati locative shift. In Continuity and change
in grammar, edited by Anne Breitbarth, Christopher Lucas, and Sheila Watts.
John Benjamins, Amsterdam.
27
Myers, Scott. 1987. Tone and the Structure of Words in Shona. Ph.D. thesis,
University of Massachusetts at Amherst.
Pantcheva, Marina. 2008. The place of PLACE in Persian. In Syntax and semantics
of spatial P, edited by Anna Asbury, Jakub Dotlaˇcil, Berit Gehrke, and Rick
Nouwen, pp. 305–330. John Benjamins, Amsterdam.
Pantcheva, Marina. 2011. Decomposing Path. The nanosyntax of directional expressions.
Ph.D. thesis, CASTL, Tromsø.
van Riemsdijk, Henk and Riny Huybregts. 2002. Location and locality. In
Progress in Grammar: Articles at the 20th Anniversary of the Comparison
of Grammatical Models Group in Tilburg, edited by Marc van Oostendorp
and Elena Anagnostopoulou, pp. 1–23. Meertens Instituut, Amsterdam.
www.meertens.knaw.nl/books/progressingrammar/.
Scott, Garry John. 2002. Stacked adjectival modification and the structure of
nominal phrases. In Functional Structure in DP and IP: The Cartography of
Syntactic Structures, edited by Guglielmo Cinque, pp. 91–120. Oxford University
Press, New York.
Starke, Michal. 2004. On the inexistence of specifiers and the nature of heads. In
Structures and Beyond: The Cartography of Syntactic Structures, vol. 3, edited
by Adriana Belletti, pp. 252–268. Oxford University Press, New York.
Starke, Michal. 2009. Nanosyntax. A short primer to a new approach to language.
In Nordlyd 36: Special issue on Nanosyntax, edited by Peter Svenonius, Gillian
Ramchand, Michal Starke, and Tarald Taraldsen, pp. 1–6. University of Tromsø,
Tromsø. Available at www.ub.uit.no/munin/nordlyd/.
Svenonius, Peter. 2006. The emergence of axial parts. In Nordlyd, special issue
on Adpositions, edited by Peter Svenonius and Marina Pantcheva, no. 33 in
Tromsø Working Papers in Linguistics, pp. 1 – 22. CASTL, University of Tromsø,
Tromsø.
Svenonius, Peter. 2008a. The position of adjectives and other phrasal modifiers
in the decomposition of DP. In Adjectives and Adverbs: Syntax, Semantics,
and Discourse, edited by Louise McNally and Chris Kennedy, pp. 16–42. Oxford
University Press, Oxford.
Svenonius, Peter. 2008b. Projections of P. In Syntax and Semantics of Spatial P,
edited by Anna Asbury, Jakub Dotlaˇcil, Berit Gehrke, and Rick Nouwen, pp.
63–84. John Benjamins, Amsterdam.
Svenonius, Peter. 2010. Spatial P in English. In The Cartography of Syntactic Structure,
vol.6, edited by Guglielmo Cinque and Luigi Rizzi, pp. 127 – 160. Oxford
University Press, Oxford. Available at http://ling.auf.net/lingBuzz/000001.
Taraldsen, Tarald. 2010. The nanosyntax of Nguni noun class prefixes and concords.
Lingua 120 6: 1522 – 1548.
Terzi, Arhonto. 2010. Locative prepositions and ‘Place’. In The Cartography of
Syntactic Structure, vol.6, edited by Guglielmo Cinque and Luigi Rizzi, pp. 196–
224. Oxford University Press, Oxford.
28
Vainikka, Anne. 1993. The three structural cases in Finnish. In Case and Other
Functional Categories in Finnish Syntax, edited by Anders Holmberg and Urpo
Nikanne, Studies in Generative Grammar, pp. 129–159. Mouton de Gruyter,
Berlin.
Zaenen, Annie, Joan Maling, and H¨oskuldur Thr´ainsson. 1985. Case and grammatical
functions: The Icelandic passive. Natural Language and Linguistic Theory
3: 441–483.
29