A Distributed Morphology Approach to Nominal Coor

Ai Taniguchi
LIN892: Distributed Morphology (Squib II)
A Distributed Morphology Approach to Nominal Coordination and Verbal Agreement in Swahili
3 May 2013
1 Introduction
Although much has been said in the literature regarding morphosyntactic agreement in Swahili, very
few researchers have addressed its relevance to coordination. There are a number of facts about
Swahili nominal coordination, observed by Marten (2000) and others; namely, that conjoined animate subjects show resolved verbal agreement (the plural of the corresponding class), inanimates
show default Class 7 plural agreement, and objects show closest conjunct agreement. The description is striking, yet little attempt has been made to theoretically account for these generalizations.
The goal of this paper is in two parts: 1. To provide cleaner data to describe the coordination agreement facts in Swahili, and 2. To provide a Distributed Morphology-driven analysis to theoretically
unify resolved, default, and closest conjunct agreement in Swahili. I will ultimately argue that the
interplay of interpretable/uninterpretable features, Probe/Goal relations, and EPP/pied-piping facts
are responsible for the agreement patterns of nominal coordination in Swahili.
This paper will be organized as follows: In Section 2, I will give a brief overview of Swahili morphosyntax, with focus placed on agreement relations. In Section 3, I will provide original nominal
coordination data, elicited from a native Swahili speaker. In Section 4 I will outline some previous
works on coordinated agreement in Swahili, and in Section 5 I will describe Distributed Morphology notions to be used in my analysis, followed by the application of it to my data. I will briefly
discuss the theoretical implications of this study in the end.
2 Background
Swahili is a Bantu language spoken primarily in the east coast of Africa (including Kenya and Tanzania). A polysynthetic language rich in agreement, it is a language of interest in morphosyntactic
theory.
2.1 Swahili Noun Classes
Like many Bantu languages, Swahili has the characteristic noun class system as a part of its grammar. All nouns in Swahili are put into one of the 7 (or more, depending on what is considered a
class) noun classes, all of which have a plural counterpart. The classes are roughly categorized
semantically (e.g., humans, natural elements, manufactured objects, fruits, etc.) but the categorization is in reality rather arbitrary. The nouns themselves are typically (but not always) marked with
a class-specific prefix, and show strict agreement with verbs, adjectives, prepositions, etc. Subject agreement with the verb is obligatory, and object agreement is mandatory for animate nouns
but optional for inanimate nouns. Table 1 below summarizes the 7 noun classes and their verbal
agreement morphemes:
Table 1: Swahili Noun Classes
Class1
“Semantic Category”
Nominal Prefix
Verbal Agreement
Example
1SG
1PL
human
human
mwa-
a-, mwa-
mtoto alicheka ‘the child laughed’
watoto walicheka ‘the children laughed’
3SG
3PL
tree
tree
mmi-
ui-
mti ulianguka ‘the tree fell’
miti ilianguka ‘the trees fell’
5SG
5PL
manufactured objects
manufactured objects
ji-/∅ma-
liya-
blanketi lilianguka ‘the blanket fell’
mablanketi yalianguka ‘the blankets fell’
7SG
7PL
“things”
“things”
kivi-
kivi-
kiti kilianguka ‘the chair fell’
viti vilianguka ‘the chairs fell’
9SG
9PL
fruits, etc. “catch-all”
fruits, etc. “catch-all”
n-/∅n-/∅
izi-
kalamu ilianguka ‘the pen fell’
kalamu zilianguka ‘the pens fell’
11SG
(11PL)
abstract/mass nouns
abstract/mass nouns
uN/A
uN/A
ugali umepikwa ‘the cornmeal has been cooked’
**Lost over time
(13SG)
13PL
abstraction
abstraction
N/A
u-
N/A
u-
**Merged with Class7
usiku ulikuja ‘night came’
Note that a couple of the noun classes have lost their singular or plural counterpart over time. Class
9 is the biggest noun class that houses the most diverse set of nouns (including a lot of loanwords),
and is sometimes referred to as the “catch-all” class.
2.2 Basic Verbal Morphology
Based on the noun class of the argument(s), verbs in Swahili take agreement prefixes. Other than
this agreement, Swahili verbs take on tense-marking prefixes. A final vowel is added post-verbally
1
It should be noted that traditionally, the plural of Class 1 is labeled as “Class 2,” the plural of Class 3 as “Class
4,” etc. I have relabeled the classes for the purposes of this paper, not only because it is more intuitive, but because it
works out better this way for feature Valuing, etc., in DM.
to conform to the CVCV syllable structure, and although not exemplified here, certain voice markers (e.g. passive marker) are optionally added as verbal suffixes as well. Below is the basic morphology of Swahili verbs2 :
(1)
Yohana a-li-chek-a
John.1SG 1SG-PST-laugh-FV
‘John laughed’
(2)
Yohana a-li-ki-vunj-a
kiti
John.1SG 1SG-PST-7SG-break-FV chair.7SG
‘John broke the chair’
Swahili is primarily SVO, but has relatively free word order. The subject can optionally be moved
to a post-verbal position in Swahili:
(3)
A-li-chek-a
Yohana
1SG-PST-laugh-FV John.1SG
‘John laughed’
(4)
A-li-ki-vunj-a
kiti
Yohana
1SG-PST-SG7-break-FV chair.7SG John.1SG
‘John broke the chair’
The semantic difference between (1)/(2) and (3)/(4) is that a sense of completion is often associated
with the latter. For the discussion of the syntax of this post-position form in Swahili, see my other
unpublished squib, Taniguchi (2013).
The purpose of this paper is to examine verbal agreement when there are conjoined DPs in either
the subject of the object position, the data for which will be presented in the following section.
3
Nominal Coordination: The Data
Various nominal coordination data were elicited from native Swahili speakers, with focus placed
on subject and object agreement. Overall, there seems to be three agreement patterns possible with
conjoined DPs: Resolved agreement, closest conjunct agreement (CCA, including First Conjunct
Agreement (FCA) and Last Conjunct Agreement (LCA)), and default agreement. The agreement
patterns show sensitivity to the subject/object and animate/inanimate distinction.
2
SG = Singular noun class, PL = Plural noun class, # = Noun class, FV = Final vowel
3.1 Subject
Let us examine the data on subject coordination first. For animate (Class 1, “human”) nouns, there
is resolved agreement– that is, with two Class 1 nouns, it takes agreement for their corresponding
plural class agreement, as in (5a):
(5) Animate subjects (CL1)
a.
Mwanamume na mtoto
wa-li-chek-a
man.1SG
and child.1SG 1PL-PST-laugh-FV
‘the man and the child laughed’
b.
* Mwanamume na mtoto
a-li-chek-a
man.1SG
and child.1SG 1SG-PST-laugh-FV
‘the man and the child laughed’
c.
Wa-li-chek-a
mwanaume na mtoto
1PL-PST-laugh-FV man.1SG and child.1SG
‘the man and the child laughed’
(5c) shows that this agreement pattern is not affected by the inversion of the subject. As seen in
(5b), CCA (in the case of pre-verbal subjects, FCA) is not possible with (uninverted) subjects.
However, Marten (2000) suggests that CCA is possible with post-verbal subjects:
(6) Animate subjects (CL1)3
a.
Haroub
na Naila
wa-li-kuj-a
Haroub.1SG and Naila.1SG 1PL-PST-come-FV
‘Haroub and Naila came’
b.
* Haroub
na Naila
a-li-kuj-a
Haroub.1SG and Naila.1SG 1SG-PST-come-FV
Int: ‘Haroub and Naila came’
c.
Wa-li-kuj-a
Haroub
na Naila
1PL-PST-come-FV Haroub.1SG and Naila.1SG
‘Haroub and Naila came’
d.
A-li-kuj-a
Haroub
na Naila
1SG-PST-come-FV Haroub.1SG and Naila.1SG
‘Haroub and Naila came’
(6a) through (6c) are consistent with my data in (5a) through (5c), respectively. However, (6d)
suggests that CCA(FCA) is possible as long as the subject is moved post-verbally, which was not
elicited from my consultant (who insists that the agreement must be resolved). I take instances like
3
Marten (2000)
this to be an example of commitative coordination. That is to say, na indicates accompaniment,
rather than “true” coordination– conveniently, na in Swahili can also be glossed as the preposition
with. Therefore, I will not treat this data as legitimate FCA, and exclude it from my analysis henceforth.
The curious case is when two inanimate nouns are conjoined as subjects:
(7) Inanimate subjects (Same class)
a.
Mti
na mpira vi-li-anguk-a
tree.3SG and ball.3SG 7PL-PST-fall-FV
‘the tree and the ball fell’
b.
Vi-li-anguk-a
mti
na mpira
7PL-PST-fall-FV tree.3SG and ball.3SG
‘the tree and the ball fell’
c.
* Mti
na mpira mi-li-anguk-a
tree.3SG and ball.3SG 3PL-PST-fall-FV
‘the tree and the ball fell’
Although resolved 3PL agreement should be possible in (7), (7c) shows otherwise. Rather, conjoined inanimates take default agreement, which is the plural of Class 74 (the “things” class, appropriately), as seen in (7a). Post-positioning the subject does not seem to have an effect on this
agreement in (7b).
As expected, inanimate subjects of different noun classes take default agreement as well:
(8) Inanimate subjects (Different classes)
a.
Sanduku na mpira vi-li-anguk-a
box.5SG and ball.3SG 7PL-PST-fall-FV
‘the box and the ball fell’
b.
Vi-li-anguk-a
sanduku na mpira
7PL-PST-fall-FV box.5SG and ball.3SG
‘the box and the ball fell’
Above in (8), the mismatching classes of the conjuncts trigger the default Class 7 agreement, again,
regardless of subject position.
Again, Marten (2000) claims that CCA is possible, in this case with pre-verbal subjects:
(9) Inanimate subjects (Different classes)5
4
5
For some speakers/dialects, this default class is Class 9 plural, which is the general “catch-all” class.
Marten (2000)
a.
ki-me-vunj-ik-a
Mguu wa meza na kiti
leg.3SG of table and chair.7SG 7SG-PERF-break-PSV-FV
‘The leg of the table and the chair are broken’
However, the preposition and the passive voice pollutes the data too much for it to be a good generalization. No reliable minimal pairs were listed in Marten (2000), so I will leave this data for future
research. My consultant did not list LCA as a possible solution; for the purposes of this paper, I
will only look at default agreement for inanimate subjects.
Lastly, the data in (10) is to show that this pattern is truly about the spec TP position associated
with subjects, not necessarily their original thematic position:
(10) Inanimate subjects (Transitive verb)
a.
Sanduku na mpira vi-li-(ya)-vunj-a
madirisha
box.5SG and ball.3SG 7PL-PST-5PL-break-FV windows.5PL
‘the box and the ball broke the windows’
b.
Vi-li-(ya)-vunj-a
madirisha
sanduku na mpira
7PL-PST-5PL-break-FV windows.5PL box.5SG and ball.3SG
‘the box and the ball broke the windows’
In contrast to to (7) and (8), which had unaccusative verbs (subject originally in complement of
V), break in (10) is a transitive verb (subject originally in spec vP). The same pattern (default
agreement) still holds.
3.2 Object
In the presence of conjoined objects, the verb agrees with the first conjunct:
(11)
Animate objects (CL1)
a.
(Mimi) ni-na-m-pend-a
mwanamume na mtoto
(I)
1SG-PRES-1SG-like-FV man.1SG
and child.1SG
‘I like the man and the child’
b.
(Mimi) ni-na-wa-pend-a
wanaume na mtoto
(I)
1SG-PRES-1PL-like-FV men.1PL and child.1PL
‘I like the men and the child’
c.
(Mimi) ni-na-m-pend-a
mwanamume na watoto
(I)
1SG-PRES-1SG-like-FV man.1SG
and children.1PL
‘I like the man and the children’
d.
wanaume na watoto
(Mimi) ni-na-wa-pend-a
(I)
1SG-PRES-1PL-like-FV men.1PL and children.1PL
‘I like the men and the children’
e.
* (Mimi) ni-na-wa-pend-a
mwanamume na mtoto
(I)
1SG-PRES-1PL-like-FV man.1SG
and child.1SG
‘I like the man and the child’
Regardless of the number feature on the DP, the closest conjunct always agrees with the verb; there
is no resolved agreement, as attempted in (11e). This pattern holds regardless of the animacy of the
DPs:
(12) Inanimate objects (Same class)
a.
(Mimi) ni-na-u-pend-a
mti
na mpira
(I)
1SG-PRES-3SG-like-FV tree.3SG and ball.3SG
‘I like the tree and the ball’
b.
(Mimi) ni-na-i-pend-a
miti
na mpira
(I)
SA1-PRES-3PL-like-FV trees.3PL and ball.3SG
‘I like the trees and the ball’
c.
(Mimi) ni-na-u-pend-a
mti
na mipira
(I)
1SG-PRES-3SG-like-FV tree.3SG and balls.3PL
‘I like the tree and the balls’
d.
(Mimi) ni-na-i-pend-a
miti
na mipira
(I)
1SG-PRES-3PL-like-FV tree.3PL and balls.3PL
‘I like the tree and the balls’
e.
* (Mimi) ni-na-i-pend-a
mti
na mpira
(I)
1SG-PRES-3PL-like-FV tree.3SG and ball.3SG
‘I like the tree and the ball’
f.
* (Mimi) ni-na-vi-pend-a
mti
na mpira
(I)
1SG-PRES-7PL-like-FV tree.3SG and ball.3SG
‘I like the tree and the ball’
And as predicted, if the noun classes of the objects differ, the verb still agrees with the closest
conjunct:
(13) Inanimate objects (Different classes)
a.
(Mimi) ni-na-li-pend-a
sanduku na mpira
(I)
1SG-PRES-5SG-like-FV box.5SG and ball.3SG
‘I like the box and the ball’
b.
(Mimi) ni-na-ya-pend-a
masanduku na mpira
(I)
1SG-PRES-5PL-like-FV boxes.5PL and ball.3SG
‘I like the boxes and the ball’
c.
(Mimi) ni-na-li-pend-a
sanduku na mipira
(I)
1SG-PRES-5SG-like-FV box.5SG and balls.3PL
‘I like the box and the balls’
d.
(Mimi) ni-na-ya-pend-a
masanduku na mipira
(I)
1SG-PRES-5PL-like-FV boxes.5PL and ball.3SG
‘I like the boxes and the ball’
e.
* (Mimi) ni-na-vi-pend-a
sanduku na mpira
(I)
1SG-PRES-7PL-like-FV box.5SG and ball.3SG
‘I like the box and the ball’
f.
* (Mimi) ni-na-ya-pend-a
sanduku na mpira
(I)
1SG-PRES-5PL-like-FV box.5SG and ball.3SG
‘I like the box and the ball’
However, there is the problem again as to if (11) through (13) is really FCA, since the object
agreement on the verb is singular. Marten (2000) has suggestive data regarding the number status
of a conjunction phrase (although, again, without a reliable minimal pair):
(14)
Matukio... ya ku-m-fung-a
Jeejee
na Sichana
pamoja
result... of INF-1SG-tie-FV Jeejee.1SG and Sichana.1SG together
‘the result of tying Jeejee and Sichana together’
Pamoja, or together, requires a plural antecedent, suggesting that Jeejee na Sichana truly is a plural
constituent. Similar data can be found in English:
(15)
a.
John, with Mary, is going to the store
b.
John and Mary are going to the store
c.
I saw John and Mary together
d.
* I saw John with Mary together
e.
? I saw John together with Mary
f.
John, together with Mary, is going to the store
The verbal inflection contrast between (16a) and (16b) reiterates that a comitative construction is
not syntactically plural, while true coordination is. The grammaticality of (16c) and (16d), then,
can be attributed to this– together must modify something plural, which, the comitative example
is not. The sentence (to me) is slightly better if together precedes with as in (16e), but the verbal
inflection in (16f) shows that the constituent John together with Mary is still singular– together
with is perhaps a preposition of itself (Swahili has a prallel construction to this: pamoja na ‘together with’). Unless the position of pamoja in Swahili is relatively free with respect to comitative
constructions, this seems to suggest that the object coordination in Swahili is actually not comitative. Although tests and data for comitatives was not a part of this study, careful separation of
coordination and comitatives is worthwhile for future research.
In either case, the important fact is that resolved and default agreement, which is possible with
conjoined subjects, is NOT possible with objects. This asymmetry will be the focus of this paper.
In summary, coordinated agreement patterns in Swahili (for the purposes of this paper) are as follows:
• Subject (animate): Resolved agreement
• Subject (inanimate): Default agreement
• Object (animate/inanimate): First conjunct agreement (or agreement impossible, unless made
comitative)
4 Previous Analyses: Marten (2000)
The above generalizations about Swahili (and other Bantu) coordination agreement has been made
(Marten (2000), Marten and Ramadhani (2001), Marten et al (2007)), but little has been said about
the actual workings of its syntax. I will briefly outline what Marten (2001) has claimed in regards
to the syntax of Swahili coordination agreement.
Marten sees three distinct types of agreement that explains the resolved, default, and FCA patterns. He calls resolved agreement “morphological agreement,” the structure of which is illustrated
below:
(16) “Morphologial
. Agreement”
.
S
.
. .
Verb
Conj
.
.
Cdβ
NP1 NP2
. Clα Clα
β is the plural class corresponding to α. Although intuitive, this is a mere restatement of the facts.
He cites several examples (including how LCA is not possible with animates but is with inanimates,
as in (6b) vs. (9a)) on the “special status” of animates in Swahili and implies that this is why they
favor resolved agreement, but he leaves the mechanics of it to “further research.”
Next, he analyzes default agreement as “anaphoric agreement,” in which conjoined DP/NPs are
an anaphor of “empty” nominal subject. The structure is illustrated below:
(17) Anaphoric
Agreement
.
.
S
.
.
.
V
NP
Conj
.
7PL/9PL
NP
and
NP
.
The idea is that that there is a null syntactic subject, which is of the default noun class (class 8 or
10). What looks likes the surface subject is actually is an adjunct, to which the null subject refers
to. He lists several examples in support of this analysis, including the fact that adjectives cannot
modify conjoined DPs in Swahili, the idea being that you cannot modify a null constituent. He also
gives an “overt” version of this null subject, in support of this analysis:
(18)
...uso, mabega, kifua, tumbo– sehemu zote
hizi
zi-li-kuw-a
ndizo
...face, shoulders, chest, stomach– parts.9PL all.9PL these.9PL 9PL-PST-be-FV ??
za karibu
of near6
‘face, shoulders, chest, stomach– all of these parts were indeed closer’
Sehemu zote ‘all parts’ is what is supposed to be the subject that is referring to the “adjunct” list of
the actual DPs. This is actually quite intuitive, and similar data can be found in other languages as
well, e.g. French:
(19)
Moi, j’aime beaucoup le fromage
me, I.like a lot
the cheese
‘I like cheese a lot’
The introduction of an utterance with an accusative pronoun is not a marked form in French, highlighting the salience of this “anaphoric” structure.
However, there are cases when the “anaphor” is not necessarily a coordination of DPs:
(20)
6
Mpira u-li-anguk-a
juu ya mti.
Vi-li-vunj-ik-a.
ball.3SG 3SG-PST-fall-FV over of tree.3SG 7PL-PST-break-STAT-FV
‘The ball fell on the tree. They (both) broke.’
No gloss given in his paper
The empty subject in ‘they broke’ would be referring back to the two arguments of the previous
utterance in the discourse– ball and tree. Given this data, it would be hard for Marten to argue that
this is a coordination-specific mechanism.
Lastly, he refers to CCA as “syntactic agreement.” He offers the following structures for FCA versus resolved agreement with animate subjects (recall that he considers certain cases of post-verbal
subjects FCA, not comitative coordination):
(21) “Syntactic Agreement”
a.
.
.
S
.
. .
Conj V
.
.
NP
NP
.
b.
.
.
V
.
.
S
. .
Conj
.
.
NP
NP
.
The idea seems to be that since the two NPs are introduced first linearly in (21a), the verb can
refer to the “information available” (namely, that the conjunction is plural) to make the agreement.
However, in (21b), he claims that the verb does not know that the subject is plural until it gets to
the second NP, essentially– hence, at the introduction of the first NP, it hastily makes the singular
agreement with it.
Other than the obvious problem that the V head seems to be changing its anchoring from right
to left depending on the word order, one could argue that the Conj node is already encoded as
inherently plural, and that the verb would “know” to agree with a plural in either of the cases. Additionally, Marten’s step-by-step introduction of the nodes from left to right does not very seem
very compatible with Merge– it is hard to argue that the verb is “introduced” first in the syntax,
then the first NP, then the second NP in (21b), for example. Rather, the two NPs are likely merged
“first”, which is then merged with the verb, and so on; that is, Merge is bottom-up. Again, Marten’s
analysis seems to be a restatement of the facts– the verb agrees with the closest conjunct if subject
is post-verbal.
5 Distributed Morphology
In order to unify the three “types” of agreement in Swahili, I will use a Distributed Morphology
(DM) (Halle and Marantz (1993) et seq.) approach analysis. In this “late insertion” approach, terminal nodes only have bundles of abstract features, and phonological information is only inserted
later in the derivation. The idea is that the “Vocabulary Item” (VI) that matches all or most of the
features on the terminal nodes gets to be inserted. It is only after this has happened that the syntax
is sent off to PF/LF for pronunciation and interpretation. Optionally, features can systematically
undergo deletion (or “Impoverishment”) before VI insertion takes place, so the appropriate candidate can “win.”
Agreement in DM works in terms of Probe, Goal, Match, and Value– a Probe (v or T usually
for verbal agreement) has uninterpretable features that need to be valued; it finds a Goal down the
tree that has these features, such as a DP with interpretable phi-features, and Matches with it. If
this Goal is still “active,” this can trigger the Valuing of the uninterpretable features. After being
valued, the feature can delete.
5.1 Boskovic (2009)
Boskovic (2009) uses DM tools to account for FCA and LCA facts in Serbo-Croatian (SC), which
we briefly touched on earlier. The facts seem to be that agreement involving EPP features seems
to differ from agreement simply “down the tree,” which in terms of SC manifests as the FCA and
LCA asymmetry.
In addition to the above notion, Boskovic assumes that in addition to inherent interpretable features
on a Goal, there could also be pre-valued uninterpretable features. The idea is that interpretable
features have an effect on the semantics (e.g., number feature), and uninterpretable features do not
(e.g., grammatical gender features). Importantly, only uninterpretable features can delete.
In terms of movement, if a Probe has an EPP feature, what exactly moves (and pied-pipes the
rest) has to be determined. The key to this is that the Probe’s feature valuator determines movement. When there are more than one potential valuator, which is the case with coordination, some
uninterpretable features must undergo deletion in order to be excluded as a valuator candidate.
The goal of this paper will be to account for the subject/object agreement asymmetry, and to an
extent the animate/inanimate asymmetry in Swahili with these tools.
5.2 Application to Swahili: Preliminary Analysis
Following Carstens (2000), I will assume that noun class in Swahili is a combination of number and
gender features. I will also follow Ngonyani (1998) and others in that object agreement in Swahili
is actually agreement, not pronoun incorporation. I will also assume the Spec/head &P(BP) structure for coordination, as in Munn (1987) et seq (this will change in my alternative analysis of the
data later).
Below are the preliminary structures for animate subjects, inanimate subjects, and animate/inanimate
objects:
(22) Animate
Subjects. (Resolved)
.
TP
.
.
.
.
.
T


uN um : +PL


 uGen : CL1 
EP P
.
.
vP
.
.
. .
&PiN um:+P L,uGen:CL1
v′
.
.
.
.
.
′
v
VP
DP
&
.
1
[
]
iN um : −P L
.
.
.
V
DP
&
2
[
]
uGen : CL1 .
√
√
iN um : −P L
laugh
man
.
uGen : CL1
√
child
.
(23) Inanimate
Subjects (Default)
.
.
TP
.
.
.
.
.
T′
.
T


uN um : +PL


uGen : CL7DEF AU LT 
EP P
.
T′
.
.
vP
.
.
.
.
&PiN um:+P L,!uGen:CL1
v′
.
.
.
.
.
.
′
v
DP
VP
&
.
1
[
]
iN um : −P L
.
.
.
.
DP
DP2
V
]
!uGen : CL3 . & [
√
√
iN um : −P L
broke window
tree
.
.
!uGen : CL3
√
ball
.
(24) Objects
(Comitative)
.
.
TP
. .
.
.
T′
.
.
T
.
.
DP
.
. I
[
.
.
.
vP
.
v
]
uN um : -PL
uGen : CL1
.
.
vP
.
.
v′
.
PP
.
.
DP
P [
]
iN
um
:
−P
L
.
.
. with
V
DP
[
]
uGen : CL1
√
√
iN
um
:
−P
L
child
like
.
uGen : CL1
√
man
.
(25) Objects
.
. (FCA)
TP
.
. .
.
T′
.
.
.
vP
T
.
.
.
.
DP
v′
.
.
. I
v
[
]
uN um : -PL
.
uGen : CL1
V
.
√
like
.
.
VP
.
VP
.
.
&PiN um:+P L,uGen:CL1
.
.
.
′
DP
&
1
[
]
iN um : −P L
.
.
DP
&
2
[
]
uGen : CL1 .
√
iN
um
:
−P
L
man
.
uGen : CL1
√
child
One proposal that I must make is that &P not only has number, but gender as well; and, this gender
is the animate Class 1. This is how I will account for default agreement with inanimate subjects,
which I will elaborate on below. That &P has Class 1 gender may be related to the fact that object
agreement with Class 1 nouns is obligatory in Swahili, but not with other noun classes7 . This is
similar to Boskovic’s remarks that &P may have feminine gender in SC because of its historical
semantic importance.
Here is the operation of (22): T probes down the tree, and Matches &P, which is plural and Class
1. It checks inside &P to see that the gender of the contained DPs are the same as that of &P’s. It
passes this test, and &P is subsequently able to value the phi features on T, and satisfies EPP. Note
that we must assume that the number feature on the DPs do not have to match the number on &P;
somehow, the interpretable plural number feature on &P wins over the DPs’ singular features– this
may not be so implausible, particularly if number and gender were separated to begin with because
of the former’s stronger semantic importance.
With inanimates in (23), T probes down the tree in a similar way. It finds &P again, but only to find
out that the DPs that the &P contains does not match it in terms of gender (i.e., Class 1 and Class
3 clash). Seeing this, it defaults to Class 7. It still gets valued for Plural by &P, which satisfies EPP.
Object agreement in (24) is quite simple if this is a case of comitative coordination– the conjunction
is not a plural period, since the second DP is simply adjoined to the first as a part of a PP. T probes
down and Matches the closest DP, and is successfully valued by it. There is no need to “check
inside,” since there is no true coordination.
However, if this is not comitative and actually is an &P, this is problematic with the structure
in (25). At this point, there should be no reason that resolved agreement should not take place. I
offer two potential solutions to this dilemma:
1. A non-EPP Probe cares if the number on &P and the DPs match or not. When it finds out
that the numbers do not match, it starts over, and agrees with the next closest thing that has both
Number and Gender, and is self-contained (unlike an &P). This ends up being the closest DP conjunct.
2. The uninterpretable features for object agreement is on a terminal node higher than v (Agr or T,
presumably). If this is true, little v would have assigned case to &P already, making the features
on it inactive for valuing (in the sense of Bhatt and Walkow, forthcoming). Therefore it Matches,
but cannot Value. So a higher object agreement probe must fully agree with the next closest thing,
7
Which also could be further related to the fact that Class 1 nouns are always animate, thus having semantic importance
which is the closest DP conjunct inside the &P.
Solution 2 seems the more promising, and consistent. An updated structure based on this solution is illustrated below:
(26) Objects
. . (FCA)- Take 2
TP
. .
.
.
T′
. .
.
T
AgrP
.
.
.
.
.
Agr′
.
.
.
vP
Agr
[
]
uN um :
.
. .
DP
v′
uGen :
.
.
.
. I v.
VP
.
.
.
.
V
&PiN um:+P L,uGen:CL1<<IN ACT IV E
√
.
.
.
like
.
′
DP
&
1
[
]
iN um : −P L
.
.
DP
&
2
[
]
uGen : CL1 .
√
iN
um
:
−P
L
man
.
uGen : CL1
√
child
If movement to Spec,AgrP is another EPP feature, then Probe likely deletes the Gender feature on
DP2 once it encounters it, making DP1 the only possible valuator. It therefore moves, and piedpipes the rest of the &P with it. To account for the SVO order that we still must maintain, there is
V to T movement of main verbs in Swahili, as proposed by Ngonyani (1998).
5.3 Problems, and Alternative Analysis
There are obvious problems with this analysis. First, the “default” agreement in (23) is not wellmotivated. It does capture the generalization of “if the conjuncts are not Class 1, take Class 7
agreement,” but the mechanics of it is a tad ad hoc. Secondly, the AgrP in (26) is slightly redun-
dant if little v can do the same job. However, again, there is no reason that conjoined objects should
not get resolved or default agreement (depending on the noun class) if the features are on v.
Also, a prediction that the structure in (23) makes is that when the subject is a conjoined animate
DP and an inanimate DP, we should expect default Class 7 agreement because of the class clash
with the &P . My follow-up data suggests in such cases, the derivation fails all together:
(27) Animate and Inanimate
a.
* Mwanamume na mti
vi-li-anguk-a
man.1SG
and tree.3SG 7PL-PST-fall-FV
Int: ‘the man and the tree fell’
b.
* Mwanamume na mti
u-li-anguk-a
man.1SG
and tree.3SG 3SG-PST-fall-FV
Int: ‘the man and the tree fell’
c.
* Mwanamume na mti
a-li-anguk-a
man.1SG
and tree.3SG 1SG-PST-fall-FV
Int: ‘the man and the tree fell’
d.
Mwanamume a-li-anguk-a
na mti u-li-anguk-a
man.1SG
1SG-PST-fall-FV and tree 3SG-PST-fall-FV
‘the man fell and the tree fell’
You cannot express (27) with a single verb; you must show independent agreement on independent verbs for each DP, as in (27d). Although the intuition that something is crashing when the
conjuncts are not animate seems to be right, the present analysis cannot explain why this cannot
be saved by Class 7 agreement when the conjuncts differ in animacy. This could potentially be
explained if Class 1 gender is interpretable, not uninterpretable– this is justifiable in that Class 1
nouns are consistently animate, thus having semantic importance8 . If this is true, the fact that the
interpretable Class 1 gender cannot be deleted may be the reason for the unsavable crash.
Another way to approach this problem is to utilize Munn (1993)’s “adjoined BP” structure for
coordination, which I have replaced the structures below with. The following are the alternate
structures for (23) and (26):
8
Thanks Hannah for pointing this out
(28) Animate
Subjects. (Resolved): Adjoined BP
.
TP
.
.
.
.
T′
.
.
.
vP
T


uN um : +PL
.
.
. .


DP
v′
 uGen : CL1 
.
.
.
.
.
EP P
v
.
VP
BP
DP
.
iN um:+P L
1
[
]
iN um : −P L
.
.
.
B
V
DP
2
[
]
uGen : CL1 .
√
√
iN um : −P L
laugh
man
.
uGen : CL1
√
child
.
(29) Objects
.
. (FCA): Adjoined BP
TP
. .
.
.
T′
.
.
.
vP
T
.
.
.
.
DP
v′
.
.
. I
v
[
]
uN um : -PL
.
uGen : CL1
V
.
√
like
.
.
VP
.
.
DP
.
.
.
BP
DP
iN
um:+P
L
1
[
]
iN um : −P L
.
.
DP
B
2
[
]
uGen : CL1 .
√
iN
um
:
−P
L
man
.
uGen : CL1
√
child
Now the FCA case is easier to account for– in (29), the little v simply agrees with the closest
DP available, which happens to be the first conjunct in this adjoined BP construction. Resolved
agreement in (28), however, is not as straight forward anymore; we must appeal to some property
of EPP and pied-piping in order to say that the first conjunct fails to be a proper valuator. Then, T
would re-probe to find the next available pied-piper/valuator, which will be the adjoined BP. There
are advantages to using the Spec/head &P , but the adjoined BP has its pros as well.
5.4 Theoretical Implications
The Swahili data appears very promising in terms of Boskovic’s notion of interpretable and uninterpretable features, and the DM framework more generally. Although much refining is needed,
we seem to be on track for theoretically unifying resolved, default, and first conjunct agreement.
More broadly, my analysis reiterates the Spec/head &P vs. Adjoined BP debate– the former seems
to be able to account for resolved agreement well, but requires extra mechanics for FCA; the latter
is good for explaining FCA, but relies on other facts for resolved agreement. Both have inverse
pros and cons that cannot be immediately compromised for, but perhaps the most important contribution of this Swahili coordination data is that it shows a clear asymmetry between agreement
“down the tree” without EPP and agreement involving movement via EPP, all of which is expected
in a way. This study absolutely merits further research.
6 Conclusion
In this squib I have argued for the compatibility of the Distributed Morphology framework (with
particular interest in Boskovic’s notion of features) in accounting for coordination agreement facts
in Swahili. Although much more work is needed, the data that I have provided shows clear evidence of the subject/object asymmetry in agreement mechanisms, and my preliminary analyses
sufficiently show that with careful selection of tools, the verbal agreement patterns of coordinated
nouns are unifiable as a theory. Further examination of Swahili will certainly yield enlightening
directions in the research of the syntax of coordination and agreement more generally.
7 Acknowledgment
A big thank you to graduate students Jonathan Choti (Linguistics, MSU) and Alem Otieno (Horticulture, UGA) for their help with the coordination data! Asante sana!! I would also like to thank all
of my LIN892 classmates and Alan for their comments and advice during my in-class presentation.
8 References
Bhatt, R., & Walkow, M. (2011). Locating agreement in grammar: An argument from agreement
in conjunctions. Natural Language and Linguistic Theory, to appear.
Bošković, Ž. (2009). Unifying first and last conjunct agreement. Natural Language & Linguistic
Theory, 27(3), 455-496.
Carstens, V. (2000). Concord in minimalist theory. Linguistic Inquiry, 31(2), 319-355.
Halle, M., & Marantz, A. (1993). Distributed morphology and the pieces of inflection. The view
from building, 20, 111-176.
Marten, L. (2000). Agreement with conjoined noun phrases in Swahili. In Afrikanistische Arbeitspapiere: Swahili Forum VII, 64, 75-96.
Marten, L., Kula, N. C., & Thwala, N. (2007). Parameters of morphosyntactic variation in Bantu.
Transactions of the Philological Society, 105(3), 253-338.
Marten, L., & Ramadhani, D. (2001). An overview of object marking in Kiluguru. SOAS Working
Papers in Linguistics and Phonetics, (11), 259-275.
Munn, A. (1987). Coordinate Structure and X-bar Theory. McGill Working Papers in Linguistics,
4(1), 121-140.
Munn, A. (1993). Topics in the syntax and semantics of coordination. University of Maryland,
College Park dissertation.
Ngonyani, D. (1998). V-to-I movement in Kiswahili. Afrikanistische Arbeitspapiere, 55, 129-144.