Stress and tone in Dagaare
Arto Anttila Adams Bo domo
Stanford University Stanford University/
[email protected] Norwegian University
of Science and Technology
adams.b o [email protected]
Decemb er 29, 1996
Abstract
Dagaare Gur is a two-tone language of northwestern Ghana with approximately
1,000,000 native sp eakers. Most of the canonical disyllabic nominals fall into three
tonal classes: L-H, H-L and H-H. In the rst two, the second tone is a p olarity
tone: a sux assumes the opp osite tone to that of the ro ot. L-L is systematically
absent. Based on our primary descriptivework, supplemented by the eldnotes of
Kennedy 1966, we show that several tonal pro cesses distinguish b etween lexical
and derived tones and, in addition, there is evidence for p enultimate stress. The
surface tonal patterns result from an attempt to optimally satisfy i one-to-one
corresp ondence b etween input tones, output tones and TBUs and ii the preference
of stressed syllables for H and lexical tones. We show that the lexical/derived
distinction and several familiar tonal phenomena, e.g. spreading, contour formation
and oating tones can b e captured by the notion of corresp ondence McCarthy and
Prince 1995. Tonal p olarity is analyzed as a stress phenomenon: p enultimate stress
attracts lexical tones and leaves the unstressed nal syllable with whatever is the
contextually optimal derived tone, i.e. the p olarity tone. Similarly, underlyingly
toneless words which are at least two syllables long b ecome H-H due to stress which
explains the systematic absence of L-L.
1 Intro duction
Dagaare Gur, Niger-Congo is a two-tone language sp oken in northwestern Ghana and
1
adjoining areas of Burkina Faso. Minimal pairs like b a `to move fast' vs. b a `to put up a
p ole' and b a a `a dog' vs. b a a `to grow up' are prima facie evidence for a lexical contrast
between H and L tones. However, in the nominal system, tonal contrasts are most of the
time restricted to the p enultimate syllable, the remaining tones b eing predictable. The
following nouns are typical memb ers of the disyllabic core vo cabulary: 1
1 a H-L b L-H c H-H d L-L
yi-ri wi-ri p o g- o {
. .
house-sg horse-sg woman-sg {
The ro ot morphemes yi and wi- in 1a and 1b are H and L resp ectively. The singular
sux -ri shows tonal polarity: its tone is opp osite to that of the adjacentrootand
thus predictable. 1c shows that tonal p olarity falls short of predicting all suxal tones
as here one single H tone is spread over b oth the ro ot and the sux. L-L nouns are
systematically absent.
In the rst part of the pap er, we argue that Dagaare is a mixed system in two or-
thogonal dimensions. First, several tonal pro cesses distinguish b etween lexical and
derived tones. In particular, the ro ot tones in 1ab are lexical, the p olarity tones and
the spreading H tone in 1c are derived. Second, we present evidence for p enultimate
stress indep endent of tone, showing that Dagaare falls somewhere b etween free tone
languages and metrical stress languages McCawley 1970 , McCawley 1978 , Hyman 1978 ,
van der Hulst and Smith 1988 . Thus, for each syllable, wehave three binary choices:
H vs. L, lexical vs. derived, stressed vs. unstressed. This amounts to eight p ossible
combinations. Not all combinations are equally well represented: a lexical H is typically
assigned to a stressed syllable, a derived L is always assigned to an unstressed syllable
and derived H and lexical L may o ccur on either.
2 The distribution of Dagaare tones. T = lexical tone, [T] = derived tone.
stressed unstressed
H + +
[H] + +
L + +
[L] { +
In the second part, we prop ose an explanation for the surface tonal patterns of
Dagaare nominals in terms of Optimality Theory OT Prince and Smolensky 1993 ,
McCarthy and Prince 1995 . The analysis is driven by a uniform system of constraints
on the asso ciations b etween tones and tone-b earing units. Classical tone rules suchas
asso ciation, spreading, insertion and deletion are not primitives of the grammar, but arise
from an attempt to get as close as p ossible to the optimal con guration. Abstracting
away from the lexical/derived distinction, this optimal con guration is shown in 3:
3 Configuration Constraints
TBU a Every tone is linked to exactly one TBU.
b Every TBU is linked to exactly one tone.
T 2
As is usual in Optimality Theory, what is optimal for a subsystem is not necessarily
optimal for the language as a whole and the optimal tone con guration do es not always get
realized. For example, the numb er of tones and TBUs may not match, yet faithfulness
keeps one from deleting and inserting elements at will. In the present approach, the
grammar will simply state the optimal goal, leaving it up to constraintinteraction to
gure out the cheap est repair. Consider a TBU without a tone. Constraint 3b will
rep ort a problem, but will not knowhowto xit. There are several p ossibilities, for
example \Insert tone" or \Spread tone", b oth of which give rise to further choices \Insert
H/L", \Spread from Left/Right". One might also try deleting the o ending TBU. Which
of these options is actually chosen is decided by the language-sp eci c constraint ranking.
This is the mo de of explanation native to Optimality Theory: constraints are maximally
simple and context-free; the sp eci c repairs are emergent prop erties of the system.
2 Lexical and derived tones
2.1 Intro duction
In Dagaare, the vast ma jority of simple nominals fall into three tonal classes: LH, HL
2
and HH. 3
4 b a a river-sg b ari river-pl
.
b a al a sick p erson-sg b a alb a sick p erson-pl
b ogi hole-sg b ogri hole-pl
b ong o donkey-sg b onni donkey-pl
sacri ce-sg b o g o sacri ce-pl b o ori
. . . . .
di e ro om-sg d eri ro om-pl
d u o pig-sg d ori pig-pl
fili sore-sg fil e sore-pl
. . . .
gyili xylophone-sg gyil e xylophone-pl
gbi e forehead-sg gb eri forehead-pl
kp olu u termite-sg kp ol o termite-sg
k u uri ho e-sg k u e ho e-pl
squirrel-pl l onn o squirrel-sg l onni
. . . .
m ol uu kind of animal-sg m olli kind of animal-sg
. . . . .
o ng o rat-sg o nni rat-pl
. . . .
pi e basket-sg p eri basket-pl
. . . .
piiri ro ck-sg pi e ro ck-pl Central dialect
pi e ro ck-sg piiri ro ck-pl Southern dialect
p o li path-sg for rats p o l o path-pl for rats
. . . .
s eng e b ed-sg s enni b ed-pl
. . . .
ti e tree-sg tiiri tree-pl
. . .. .
tiri sp o on-sg ti e sp o on-pl
. . . .
t o ori ear-sg t ob o ear-pl
t o ori far-away-sg t o e far-away-pl
. . . . .
wi e farm-sg w eri farm-pl
. . . .
wiri horse-sg wi e horse-pl
y u oni year-sg y u om o year-pl
y u o ri p enis-sg y o e p enis-pl
. . . . .
zi e red-sg ziiri red-pl
. . .. .
z ag a animal house-sg z agri animal house-pl
.
z ammu onion-sg z am a onion-pl
. 4
5 biri seed-sg bi e seed-pl
d oli dry sp ot-sg d ol o dry sp ot-pl
g ani b o ok-sg g am a b o ok-sg
.
g u o thorn-sg g u u ri thorn-pl
. . . . .
gy eli egg-sg gy el e egg-pl
. . . .
gb e ri leg-sg gb e e leg-pl
. . . .
k og o mahogany-sg k ogri mahogany-pl
miri rop e-sg mi e rop e-pl
n eni meat-sg n em e meat-pl
. . . .
ni e p erson-sg n u b o p erson-pl
. . . .
n u o hen-sg n u u ri hen-pl
. . . . .
n u ori mouth-sg n o e mouth-pl
. . . . .
o r a a typ e of b erry-sg o rri typ e of b erry-pl
. . .
p a ari vagina-sg p e e vagina-pl
. . .
p ani do or-sg p am a do or-pl
. .
s ani debt-sg s am a debt-pl
.
w eg e log-sg w egri log-pl
. . . .
w ogi tall-sg w ogri tall-pl
yiri house-sg yi e house-pl
y u ori name-sg y o e name-pl 5
6 b a a dog-sg b a ari dog-pl
.
bi e child-sg biiri child-pl
b o g o shoulder-sg b ogri shoulder-pl
. . . .
b u o goat-sg b uu ri goat-pl
. . . . .
d angn a owner-sg d angm a owner-sg
d u ng o animal-sg d u nni animal-pl
. . . .
d o o man-sg d ob o man-pl
. . . .
f a a bad-sg f a ari bad-pl
.
op en p ot-sg gb a al a op en p ot-pl gb a ali
.
g o ll a a wild mouse-sg g o lli wild mouse-pl
. . .
k u l a a river-sg k u lli river-pl
. . .
k u o rat-sg k u uri rat-pl
em e kind of fruit-pl kind of fruit-sg li eni li
......
l ugri prop-sg l ug o prop-pl
l u or a a lion-sg l u ori lion-pl
. . . . .
ny agri ro ot-sg ny ag a ro ot-pl
.
ngm ani calabash-sg ngm am a calabash-pl
.
p egi shell-sg p egri shell-pl
. . . .
pir a a button-sg pirri button-pl
iri ro of-pl pi e ro of-sg pi
. . . . .
p o g o woman-sg p ogb o woman-pl
. . . .
p u o stomach-sg p uu ri stomach-pl
. . . . .
t ang a mountain-sg t anni mountain-pl
.
ting e town-sg tinni town-pl
. . . .
t u u forest-sg t u uri forest-pl
v a ali rubbish-sg v a al a rubbish-pl
.
w al a a antelop e-sg w alli antelop e-pl
.
zi e place-sg ziiri place-pl
z u mmu sh-sg z u m o sh-sg
. . . .
Before addressing the sp eci c tonal prop erties of Dagaare nominals, a few general
observations are in order. Following Kennedy 1966 , we p osit the syllable as the TBU.
On his analysis, the Dagaare word consists of a sequence of one or more primary syllables,
each with contrastive length, followed byaword- nal optional secondary syllable which
is always CV. The canonical simple word is disyllabic. Some representative examples
are given in 7. 6
7 a CV.CV ti.ri `sp o on'
. .
b CVV.CV tii.ri `trees'
.. .
!
CVV.V n a a. u `cow'
.
c CV.V yi. e `houses'
CVV d a a `push'
d CVC.CV b og.ri `holes'
Examples b and c suggest that word-medially CVV sequences are monosyllabic,
but word- nally either CVV or CV.V is p ossible. This asymmetry is noted by Kennedy
1966 , 8-11 who shows that word-medial VV-sequences are either long vowels /i:, i:, a:,
.
u:, u:/ or diphthongs /ie, ie, uo, u o/, whereas word- nally genuine vowel sequences are
. . . . .
also found. In supp ort of this, we note that VV-sequences like /ee, o e, ue/ are absent
3
word-medially, but o ccur word- nally. The CV.V syllabi cation obviates the need for a
new syllable typ e found only in word- nal p osition. As for tonal evidence, we note that
the tonal patterns CVVCV and CVVCV whichwould b e exp ected under the CV.V.CV
syllabi cation and also if the TBU were the mora are systematically absent, yet word-
nally we nd yi e `houses' and vi l a a `go o d' where a two-tone melo dy is split over a nal
.
CVV, suggesting CV.V. Word- nal CVV-sequences also show contextual variation which
can b e attributed to comp eting syllabi cations. For example, s a a `father', d a a `b eer' and
vil a a `good' have alternative pronunciations s a a , d a a and vil a a . The second tone emerges
. .
esp ecially clearly under emphatic lengthening as in [s a:: :] `father!'. In the closely related
Dagbane, the emergent second tone has b een explained in terms of an additional TBU
Wilson 1970 , Hyman 1993 . We attribute this variation to the variable resyllabi cation
of CV.V as CVV in p ostlexical phonology. It follows, correctly, that similar variation is
excluded in CVCV-words which are unambiguously CV.CV.
The more marginal word typ es include b a `friend', z u `head', n u `hand' CV, b a al.b a
`sick p eople' CVVC.CV, vi.l a. a `go o d' CV.CV.V and kpo.lu u `termite' CV.CVV.
.
There are also simple words with more than two tones, for example kp aru u LHL `shirt-
. .
sg'. These cases will b e discussed later. Trisyllabic words, for example n o otiri `sho e' and
. . . .
simiri `eye', generally seem either recent loans or comp ounds and involve additional tonal
4
complexities we will not attempt to analyze in this pap er.
2.2 Tonal p olarity
The surface inventory LH, HL and HH is not the exp ected one. In a maximal two-tone
system, each syllable would have its own tone with no regard for the tone of other syllables,
yielding two kinds of monosyllables H, L four kinds of disyllables HH, HL, LH, LL and
5
so on. In words like wi-ri `horse' and yi-ri `house' the rst syllable b ears a contrastive tone
while the second syllable automatically assumes the opp osite tone. This syndrome, known
as tonal p olarity, is attested in several West African languages, for example Hausa, Margi,
Kanakuru Chadic Ho mann 1963 , Pulleyblank 1986 , Newman 1995 , Mo ore and Lama
Gur Kenstowicz et al. 1988 . On this view, the H-H typ e constitutes an exceptional
class. Put in another way, while an initial L tone is always p olarizing, initial H tones 7
come in two avors: one is p olarizing, the other spreads. Nothing in the word segmental
phonology, meaning helps in guring out the tonal typ e.
8 w eg-ri log-pl p eg-ri shell-pl
. . . .
g u u -ri thorn-pl b uu -ri goat-pl
......
n u u -ri hen-pl p uu -ri stomach-pl
......
n u -b o p erson-pl d o-b o man-pl
. . . .
We prop ose the following representations for wiri `horse', yiri `house' and p og o
. .
`woman' and call them classes A, B and C, resp ectively. The underlying forms are given
in 9a, the surface forms in 9b.
9 a wi -ri yi -ri po g-o b wi -ri yi -ri po g-o
. . . .
L H L [H] H [L] [H]
In typ es A and B, the ro ot tone is lexical and do es not spread, consequently forcing
tone insertion on the toneless sux. This newly inserted derived tone annotated by
square brackets will assume a value opp osite to the ro ot- nal tone by the Obligatory
Contour Principle Leb en 1973 which prohibits two adjacent identical tones *HH, *LL,
hence p olarity. The analysis implies that there is no synchronic reason to presume that
the sux tone started underlyingly as some sp eci ed tone or other. In class C, b oth the
ro ot and the sux are lexically toneless which induces the insertion of a derived H tone
which spreads on b oth syllables. This may seem surprising given that L is the typical
default tone. We will later prop ose that this H tone is induced by the presence of word
stress whichoverrides the general preference for L tones.
10 The main features of the analysis:
a Lexical tones induce p olarity but do not spread.
b Derived tones spread and do not induce p olarity.
c Polarity tones are derived tones.
There are of course many alternatives available. One initially plausible solution would
b e that A and B stems come with the complex tones LH and HL, C stems are H and
suxes are toneless. In the case of complex-toned stems, the second tone would always
b e realized on the numb er sux due to a constraint against contour tones. We call this
the Complex Tone Analysis. 8
11 The Complex Tone Analysis
UR Associate Spread Output
wi -ri wi -ri { wi -ri
LH L H L H
yi -ri yi -ri { yi -ri
HL H L H L
po g -o po g-o po g-o po g-o
......
H H H H
Kenstowicz et al. 1988 , 78-79 consider this analysis for Mo ore, but reject it for two
reasons. First, it fails to account for the systematic absence of L-L words, a glaring gap
that any analysis should explain. The analysis raises, but do es not answer, the question
why there are no L stems analogous to H stems. Second, phonological inventories are
typically hierarchical in nature and the presence of more complex memb ers presupp oses
the existence of simpler ones. In this case, the p osited tonal inventory violates this basic
principle of phonological organization: it p osits b oth HL and LH contours in absence of
a basic L.
Alternatively, one might stick with the assumption that the ro ots are H, L and toneless,
but prop ose that all suxes are underlyingly H and that there is a dissimilation rule which
lowers H suxes after H stems, thus yielding the app earance of tonal p olarity.We call
this the Dissimilation Analysis.
12 Lowering rule = Meeussen's rule
HH! HL
13 The Dissimilation Analysis
UR Associate Spread Meeussen Output
wi -ri wi -ri { { wi -ri
L H L H L H
yi -ri yi -ri { yi -ri yi -ri
H H H H H L H L
po g -o po g-o po g-o { po g-o
......
H H H H
This analysis has b een defended for at least Mo ore Kenstowicz et al. 1988 , Dagbane
6
Hyman 1993 and Margi Pulleyblank 1986 . The absence of p olarity in class C now
follows: the only tone present is the suxal H and the lowering rule needs two H tones to 9
apply. The absence of L{L words follows as well: suxes are underlyingly H and lowering
is not p ossible after L tones.
However, the analysis do es not re ect the fundamental insight that tonal p olarity
is essentially neutralization, i.e. susp ension of tonal contrast. It is exactly for this
reason that it typically o ccurs in proso dically weak p ositions: word- nally rather than
word-medially, on clitics and axes rather than ro ots, very much like extrametricality
Pulleyblank 1986 , 214. In the dissimilation analysis it is precisely the sux { sometimes
even only the sux { that carries the underlying H tone and b ears the descriptive burden
in the analysis.
The reason why p olarity is so natural and its o ccurrence so exp ected is b ecause tone
functions proso dically and works naturally with melo dic patterns [...] Whereas
contentwords esp ecially nouns and verbs have the b o dy to carry distinctive tone,
with short, unstressed grammatical morphemes, a sp eci c tone, whether it b e high
or low, has very little saliency Newman 1995, 776.
Neither do es the dissimilation analysis generalize in the rightway. In Dagaare, free-
standing nonemphatic CV-pronouns are L toned. However, if the pronoun is cliticized on
a L-toned verb, it p olarizes to H as 16 shows.
14 N/fu / u Nmi e-r e m a/fu / u l a
. . .
. . . .
1p/2p/3p b eat-imperf 1p/2p/3p fact
I/you/he.she.it was b eating me/you/him.her.it
15 u ni e m a/fu / u l a
. .
. . .
3p step 1p/2p/3p fact
He stepp ed on me/you/him.her.it
16u n a l a Nm e m a/fu / u
. .
. .
3p.emph fact b eat 1p/2p/3p
It was he who b eat me/you/him.her.it
These facts show that tonal p olarity cannot b e accounted for merely by the lowering
rule in 12. We will later suggest that nonemphatic pronouns are toneless and get a
default L in non-p olarity contexts. This agrees with our claim that p olarity tones are
derived, not lexical. Finally, the absence of L{L words only follows given the stipulation
that all suxes are H, a fact hardly less puzzling than the absence of L{L words itself.
To summarize our prop osal, tonal p olarity is tone insertion on a toneless TBU. If there
is an adjacent lexical tone, the OCP induces the opp osite tone. If no tone is present, a
single H tone is inserted and spread across the whole word. In the sections to come, we
will present more evidence for the distinction b etween lexical and derived tones and for
the prop osed analysis. 10
2.3 High tone suppression
Our analysis claims that the H tones in yiri and p og o are of di erent origin whichis
. .
re ected in the presence vs. absence of tonal p olarity. Another di erence emerges in the
noun+adjective N+A construction. In Dagaare, as in Dagbane and Mo ore, adjectives
come after the noun. The numb er sux app ears only once, on the last adjective, which
also determines the shap e of the sux allomorph.
17 A wir-i+t o o r-i ! wir-t o or-i horse-far.away-sg
......
wir-i+zi- e +t o or-i ! wir-zi-t o o r-i horse-red-far.away-sg
......
wir-i+f a- a ! wir-f a- a horse-bad-sg
B yi-ri+ t o or-i ! yi-t o o r-i house-far.away-sg
......
yi-ri+ zi- e +t o o r-i ! yi-zi-t o or-i house-red-far.away-sg
......
!
yi-ri+ f a- a ! yi- f a- a house-bad-sg
C p og- o +t o or-i ! p o g-t o or-i woman-far.away-sg
......
p og- o +zi - e +t o or-i ! p o g-zi -t o o r-i woman-red-far.away-sg
......
p og- o +f a- a ! p o g-f a- a woman-bad-sg
. . .
The main observation is that class C stems lose their H tones b efore adjectives. In
Class B, the H tone stays put and we observea downstep if a H toned adjective follows
!
yi- f a- a . Again, we prop ose that derived H tones are suppressed, lexical H tones are
not. As 18 shows, p olarity [H] is suppressed even in cases like s a a `father' whichhave
7
no numb er sux:
18
A s a a+s a a ! s a as a a father's father N+N, no suppression
s a a+s a a+vil- a a ! s a as a a-vi l- a a father's go o d father N+N+A
. .
s a a+vil- a a+ s a a ! s a a-vi l- a as a a go o d father's father N+A+N
. .
Wenowhavetwo indep endent tests for checking H tones: derived tones should b oth
spread and b e suppressed; lexical tones should do neither. Note that our classi cation is
no longer inno cent nomenclature: it would b e entirely p ossible for a H tone to spread, but
not b e suppressed, or not to spread i.e. trigger tonal p olarity and yet b e suppressed.
However, such mixed b ehavior do es not exist. As the following examples show, even non-
canonical longer words, some of which are obvious loans, b ehave as predicted: whenever
there is tonal p olarity, the H tone survives B; whenever there is spreading, the H tone
is suppressed C: 11
!
19 B fintil- e +f a- a ! fintil- f a- a lamp-bad-sg
. . . . .
!
dind o -ri +f a- a ! dind o - f a- a do or-bad-sg
. . . . .
!
k ar ansu n-i +f a- a ! k ar ansu n- f a- a kerosine-bad-sg
. . .
!
n o oti-ri f a- a ! n o o ti- f a- a sho e-bad-sg
......
!
s aku u-ri+f a a ! s aku- f a a scho ol-bad-sg
C b ad e r-i +f a- a ! b ad er-f a- a spider-bad-sg
. . .
b angyi r- a a+f a- a ! b angyir-f a- a latrine-bad-sg
. .
d ankyin-i+f a- a ! d ankyin-f a- a wall-bad-sg
gb ati r-u u +f a- a ! gb atir-f a- a kind of plant-bad-sg
. . . .
gyirimin-ii +f a- a ! gyirimin-f a- a slimness-bad-sg
......
k ommi-ru u+f a- a ! k ommi-f a- a tomato-bad-sg
kp ant ol- o+f a- a ! kp ant ol-f a- a topknot-bad-sg
k unkun- i+f a- a ! k unkun-f a- a tortoise-bad-sg
simi- e+ f a- a ! simi-f a- a Bambara b eans-bad-sg
s o w o l- o +f a- a ! s ow o l-f a- a kind of fo o d-bad-sg
. . . . .
t ak or- o+f a- a ! t ak or-f a- a window-bad-sg
At this p oint, we return to the Dissimilation Analysis discussed in section 2.2. The
analysis claims that class C stems are toneless, the surface H tone b eing of suxal origin.
This generalizes to H tone suppression in p og- o + f a- a ! p og-f a- a . Since the sux - o is not
. . . .
present, the toneless stem pog- surfaces with the default L tone Kenstowicz et al. 1988 ,
.
85. However, the examples in 19 remain problematic: the H tone do es surface except
in the last syllable of the ro ot. If the H tone in b ad e r-i and other HHH words indeed were
. .
of suxal origin, then wewould exp ect the following unattested forms:
20 C b ad e r-i +f a- a 6! *b ad er-f a- a spider-bad-sg
. . .
b angyi r- a a+f a- a 6! *b angyi r-f a- a latrine-bad-sg
. .
gb ati r-u u +f a- a 6! *gb atir-f a- a kind of plant-bad-sg
. . . .
To salvage the analysis, one could suggest that the stable rst tones are lexically
presp eci ed as H and thus not susceptible to suppression. However, one would still wonder
what keeps them from spreading onto the toneless syllable. This problem of partial H-
tone suppression in long words and the descriptive generalization \no [H] tones on the
TBU on the immediate left of a N+A juncture" will remain a challenge for our analysis
as well. Our main result is in showing that, at the N+A juncture, derived [H] tones are
suppressed, but lexical H tones survive, with downstep if a high tone lexical or derived
follows.
2.4 Downstep and downdrift
In this section, we will take a closer lo ok at downstep and downdrift in various contexts.
Once again, lexical and derived tones turn out di erent. 12
In many tone languages, a L tone lowers the overall pitch range or key.Thus, in the
sequence H LH ,H hasalower pitch than H . If the L tone is explicitly presenton
1 2 2 1
the surface we call the phenomenon downdrift, if \hidden" wehave downstep. Since
downdrift is predictable it is usually left untranscrib ed. Downstep on the other hand can
b e surface-phonemic. Here we follow a standard interpretation by whichdownstep equals
a oating L tone Pulleyblank 1986 .
21
Representation Transcription Phenomenon
!
a ta t a.t a. t a downdrift, pitch levels = 1-3-2 ta ta
H L H
!
b ta ta t a. t a downstep, pitch levels = 1-2
HL H
c ta ta t a.t a neither, pitch levels = 1-1
H H
Dagaare has b oth downstep and downdrift Kennedy 1966 . We will now examine
downstep/downdrift in the following environments:
22 a across ro ot+sux juncture
b across N+A juncture
c across N+N juncture asso ciative constructions
!
At this p oint, weintro duce nouns of the form H H and call them Class D. Here the
situation is a bit more complicated as the same stem may b e class B in the singular, but
class D in the plural or vice versa.
!
thing-sg b u m m a thing-pl 23 b u ni
. . .
!
d a a b eer-sg d a a n e e b eer-pl
. .
!
ii li horn-sg iil e horn-pl
......
!
k a a oil-sg k a a n e e oil-pl
. .
!
n e e water-pl o water-sg k u o k u
......
! !
k u o r a a farmer-sg k ori b o farmer-pl
. . . . .
!
kp a a u guineafowl-sg kpinni guineafowl-pl
. .
!
n a a u cow-sg nii cow-pl
. ..
!
pi r uu sheep-sg piiri sheep-pl
......
!
s a an a stranger-sg s a a m a stranger-pl
!
s o g l a a black-sg s ogl o black-sg
. . .
! !
s u a rabbit-sg s u o n i rabbit-pl
. . . .
! !
s u o N a a rabbit-sg s u o ni rabbit-pl
. . . . .
! !
y u o r a a tourist-sg y ori b o tourist-pl
. . . . .
!
zi i blood-sg zi i n e e blood-pl
......
!
z u head-sg z u ri head-pl 13
Assuming that downstep is due to a oating L tone, we can p osit the tonal melo dy
HLH for class D. Next, we observe that class D patterns together with class B with resp ect
to the N+A construction: the ro ot H is not suppressed and if a H-toned adjective follows
we hear an intervening L tone.
!
24 B yi- f a- a house-bad-sg
!
D d a a- f a- a b eer-bad-sg
!
kp a a- f a a guineafowl-bad-sg
!
pi- f a- a sheep-bad-sg
.
!
s u o- f a- a rabbit-bad-sg
. .
!
z u- f a- a head-bad-sg
!
iil- f a- a horn-bad-sg
..
!
k u o r- f a- a farmer-bad-sg
. .
!
s u on- f a- a rabbit-bad-sg
. .
!
y u o r- f a- a tourist-bad-sg
. .
!
Phonetically, the intervening L is realized either as a downstep H- H or an actual con-
tour HL-H. While we do not have a formal explanation for this di erence, we note that
all the contour cases wehave observed involve a syllable closed by a sonorant consonant.
One p ossible hyp othesis is that B and D stems are b oth underlyingly H, but that D
stems select inherently H-toned numb er suxes singular, plural, or b oth. This selection
must b e morphologically conditioned as the same stem may take a regular p olarity sux
in the singular, but a H sux in the plural, or vice versa. The intervening L tone can
now b e explained as an automatic consequence of the OCP: two adjacent H tones trigger
L-interp olation.
The closely related Dagbane shows that the internal downstep may also originate from
a HL stem. As in Dagaare, canonical nouns fall into four tonal classes and in the N+A
construction the noun app ears without a sux.
25 Dagbane Hyman 1993 , 237
A /w ah-u/ + /tit a-li/ w ar-tit a-li horse-big-sg
B /s an- a/ + /tit a-li/ s a an-tit a-li stranger-big-sg
C /p ag- a/ + /tit a-li/ p ag-tit a-li woman-big-sg
! !
D /kp aN- a/ + /tit a-li/ kp aN- tit a-li guineafowl-big-sg
In Dagbane, B and D b ehave di erently: only D triggers downstep. This can b e
explained by assuming that in Dagbane the OCP do es not trigger downstep insertion and
that the stems are H and HL resp ectively.Thus, it would not b e to o surprising if Dagaare
to o had genuine HL toned stems. For now, we simply note that b oth 26a and 26b are
p ossible representations for class D.
! !
26 a kp a a - u b kp a a - u
. .
H [L] H H L H 14
With these preliminaries b ehind us, wenow show that lexical tones trigger downstep,
derived tones do not. For the data, we turn to the asso ciative construction N+N which
typically has a p ossessive meaning.
! !
27 a kp a a - u yi-ri guineafowl-sg house-sg
.
! !
kp a a - u bi- e guineafowl-sg child-sg
.
! ! !
kp a a - u kp a a - u guineafowl-sg guineafowl-sg
. .
! !
pi- r uu yi-ri sheep-sg house-sg
. . .
! !
pi- r uu bi- e sheep-sg child-sg
. . .
! ! !
pi- r uu kp a a- u sheep-sg guineafowl-sg
. . . .
! !
s u o- N a a yi-ri rabbit-sg house-sg
. .
! !
s u o- N a a bi- e rabbit-sg child-sg
. .
! ! !
s u o- N a a kp a a- u rabbit-sg guineafowl-sg
. . .
b bi- eyi-ri child-sg house-sg
bi- ebi- e child-sg child-sg
!
bi- e kp a a - u child-sg guineafowl-sg
.
p o -g o yi-ri woman-sg house-sg
. .
p o -g o bi- e woman-sg child-sg
. .
!
p o -g o kp a a - u woman-sg guineafowl-sg
. . .
wi-riyi-ri horse-sg house-sg
wi-ribi- e horse-sg child-sg
!
wi-ri kp a a - u horse-sg guineafowl-sg
.
b a al- ayi-ri sick.p erson-sg house-sg
b a al- abi- e sick.p erson-sg child-sg
!
b a al- a kp a a- u sick.p erson-sg guineafowl-sg
.
kp ol-u uyi-ri termite-sg house-sg
kp ol-u ubi- e termite-sg child-sg
!
kp o-lu u kp a a- u termite-sg guineafowl-sg
.
The examples in 27 show abutting H tones across a word b oundary. In a, the rst
tone is lexical class D sux with a following downstep, in b the rst tone is derived
8
p olarity tone or class C tone with no downstep; b oth H tones remain equally high.
This reveals a parallel b etween tonal p olarity and downstep.
28 The Downstep Generalization:
!
HH ! H [L] H = H H
!
H [H] ! H [L] H = H H
[H] H ! no change = HH
[H] [H] ! no change = HH 15
Similar facts are found in Dagara and Dagbane although descriptive statements vary.
According to Delplanque 1983 , in Dagara a ro ot H causes downstep across words,
whereas a sux H do es not:
29 Dagara Delplanque 1983 , 124
!
a z el ez u u n u *zu un u
'C'est la t^ete du mendiant.'
!
b s o-r on a *s o-r o n a
'Ce sont les couteaux.'
Delplanque's generalization which is stated in terms of morphology ro ot vs. sux,
not phonology lexical vs. derived, is only approximately true in Dagaare: there are H
suxes which trigger downstep class D and H ro ots which do not class C. Yet it seems
clear why the generalization should work in most cases: statistically, ro ot tones tend to
b e lexical underlyingly sp eci ed whereas suxal tones tend to b e derived.
In 30 we see the parallel facts in Dagbane Wilson 1970 , Hyman 1993 . The data
lo oks di erent due to the p ostlexical pro cesses of vowel elision and rightward tone spread-
ing which destroy surface p olarity.At the lexical level, however, we encounter the familiar
Dagaare pattern:
30 Dagbane Hyman 1993 , 238
Lexical Spreading Elision Gloss
A /w ahu/ w ahu [w ah yili] horse's house
!
B /s an a/ s an a [s an yili] stranger's house
C /p ag a/ p ag a [p ag yili] woman's house
! ! !
D /kp aN a/ kp aN a [kp aN yili] guineafowl's house
In the three cases where downstep is audible in principle B, C, D, we get the exp ected
pattern: the downstep in B is due to the suxal [L]-tone; in C there is no downstep as
!
predicted; in D the internal downstep of kp aN a is heard on the following word due to
elision.
Finally,we consider downdrift across words. Here two H tones are separated bya word-
nal L toned syllable. The absolute pitch levels are indicated numerically: 1 denotes the
highest, 5 the lowest tone.
!
31 a e-b a a bi- e cro co dile-sg child-sg 1-3 3-3
.
!
b e-b a a p o g- o cro co dile-sg woman-sg 1-3 3-3
. . .
!
c e-b a a yi-ri cro co dile-sg house-sg 1-3 3-5
.
Compared to downstep, the pitchlowering is remarkably dramatic. In 31a, the
downdrift-triggering second tone of e -b a a and the second word are phonetically equally
.
low = 3. However, the phonological H/L contrast is preserved. This is evident from 31c 16
where -ri remains lower than yi-.Now, consider ay o o `female name', at rst hearing similar
. .
to e -b a a . In this case, downdrift is noticeably less dramatic: high remains higher than
.
low.
32 ay o o bi- e Ayoo -sg child-sg 1-3 2-2
. . . .
ay o o p o g- o Ayoo -sg woman-sg 1-3 2-2
......
ay o o yi-ri Ayoo -sg house-sg 1-3 2-4
. . . .
The crucial di erence b etween the low-toned syllables in a-y o o and e -b a a is that in
. . .
the former the low tone is lexical, in the latter derived. This di erence emerges clearly
under emphatic lengthening: e -b a a::: has a steadily falling contour whereas ay o o:: : has
. . .
a nal rise, i.e. a p olarity tone. While not audible on the surface of a-y o o, the p olarity
. .
tone [H] blo cks downdrift on the following H.
Wenow summarize our results so far. Three indep endent diagnostics for identifying
lexical vs. derived tones have b een prop osed: i tonal p olarity; ii H tone suppression;
iii downstep. If applied to the same tone, the results should converge. Indeed, words
which do not show tonal p olarity undergo H suppression p og- o ,p og-f a a and do not
. . .
trigger downstep p og- o bi e . Polarity tones, b eing derived, are suppressed s a a, s a a-
. .
vil a a and do not trigger downstep wir-iyi-ri . Words which trigger downstep on the
.
!
following adjectiveyi- f a a p olarize their suxes yi-ri . It is dicult to see how all
these facts could b e connected under an analysis which do es not di erentiate b etween
two kinds of tones along the lines of our lexical vs. derived distinction.
2.5 Word stress
Finally,we turn to evidence for stress in Dagaare. Our conclusion will b e that stress
interacts with b oth lexical and derived tones, yet it can b e identi ed with neither.
In many tone languages, there is evidence for accent indep endent of tone. While
in some cases accent is hardly more than a diacritic, in others it has genuine stress-like
prop erties. Typically, only one contrastive tone p er word is found; this tone is assigned
to a metrically strong p osition and it do es not spread or assimilate. All these are well-
known prop erties of stress McCawley 1978 , Hyman 1978 , Hayes 1995 . In Bantu, H tone
attraction to strong p ositions is common. To take a few examples from recent litera-
ture, in Lamba underlying H tones are phonetically manifested at the heads of tro chaic
feet, creating an alternating tonal pattern Bickmore 1995 . In Luganda, the second
mora b ears an accent which induces H tone attraction and lengthening and blo cks con-
sonant deletion/mutation Hyman and Katamba 1993 . In Kizigua, lexically supplied H
tones are attracted to prominent p ositions of the metrical grid: p enult and the initial
syllable, a metrical system like that of Polish Kenstowicz 1989 . In Chichewa, ^ lexi-
cal words end in a disyllabic fo ot with accompanying p enultimate H tone assignment
Kanerva 1989 . In Zulu and Xhosa, H tones shift to the metrically strong p osition in the
nal fo ot Goldsmith 1988 . In Kimatuumbi, nouns have a single accented syllable which
receives a H tone, all others receive L. Moreover, accented H tones distinguish themselves
by refusing to ob ey tone shift Pulleyblank 1983 , Odden 1995 . Finally,inTonga, a L 17
tone is asso ciated with the accented syllable, much as in the intonation asso ciated with
yes-no questions in English Goldsmith 1983 . As for Dagaare, we prop ose that word
stress falls on the p enultimate syllable.
The rst hint ab out stress in Dagaare comes from segmental asymmetries. According
to Kennedy 1966 , 3-4, the simple word consists of a primary syllable followed byan
optional secondary syllable. Kennedy's diagram slightly mo di ed for clarity is given
9
b elow.
33 The structure of the simple word Kennedy 1966:
word
1 2
onset nucleus coda onset nucleus
1 1 1 2 2
The secondary syllable is weaker than the primary syllable in several ways. Not only
is its structure simpler optional onset, no co da, but it also supp orts fewer contrasts.
The crucial nuclear asymmetry is that only nucleus may carry contrastive length k u.ri
1
`to hammer' vs. k u u.ri `a ho e'. As for margins, onset p ermits all the 24 consonant
1
phonemes listed in 34, whereas b oth coda and onset maximally p ermit six, namely
1 2
those segments not enclosed in parentheses Kennedy 1966 , 3. If b oth coda and onset
1 2
are present, the restrictions are even more stringent. According to Kennedy, the only
10
p ossible combinations in this case are /gr/ and /gl/.
34 p kp t ky k ?
b gb dr gy g
m Nm n ny N
f s h
v z
l y w
Secondary i.e. noninitial nal CV syllables also undergo p ostlexical reduction, a
phenomenon quite widespread in Gur languages Rialland 1985 , Hyman 1993 .
35 Ap o cop e
k ul-i k ul- go home-inf
k u u-ri k u u-r ho e-sg
p o g- o p o g- woman-sg
. . .
t o or-i t o or- far away-sg
w og-i w^og- tall-sg
D ag a a-r e D ag a a-r Dagaare-sg 18
Another diagnostic for stress is tone. It is well-known that tones are attracted to stress
in many languages, including English Hayes 1995 . In Dagaare, the phenomenon of tonal
p olarity itself can b e derived from tone-to-stress attraction: characteristically, lexical
tones, b oth H and L, asso ciate to the stressed p enult and nal syllables get whatever is
the contextually appropriate derived tone, i.e. the p olarity tone. This agrees in spirit
with Newman's 1995 suggestion that the proso dically prominent ro ot is more capable
of realizing lexical contrasts than the unprominent sux.
In conjunction with tone-to-stress attraction, p enultimate stress predicts that the
numb er of syllables should make a di erence in the placement of lexical tones. For con-
rming evidence, consider nouns which take a singular sux of the form -CVV instead
of the canonical -CV. Since a word- nal CVV can b e syllabi ed either CVV or CV.V,
we predict two distinct stress patterns and, given tone-to-stress attraction, two distinct
tone linearization patterns. This prediction is b orne out. In some CV.CVV words, lexical
tones show evidence for CV.CV.V with stress in the middle, others for CV.CVV with
initial stress. The choice seems tied to a particular sux and is thus morphologically
conditioned. The following data is organized by the familiar classes A, B, C, D. Classes
0 0
A and B are new. The stressed p enult is underlined.
36 a A e .b a a e r.ri `cro co dile sg./pl.'
. . .
.r a a or.ri `kind of b erry sg./pl.' o
. . .
0
A kp a.l a . a kpal.li `sling sg./pl'
. u kpar.ri `shirt sg./pl. kp a.ru
. . .
u.l e . e ul .li `branch sg./pl.
B m o.lu u m ol.li `kind of animal sg./pl.
. . . . .
.lu u kp ol.li `termite sg./pl.' kp o
0
. a vi e.li `go o d sg./pl.' B vi.l a
. . . .
.l a e `white sg./pl.' pi . a pi .li
. . . .
si.l a . a sil.li `hawk sg./pl.'
. . .
.l a a k u l.li `river sg./pl.' C k u
. . .
w a .l a a w al.li `antelop e sg./pl.
.
!
. r uu pii.ri `sheep sg./pl.' D pi
......
In A, the lexical H is attracted to the p enult = the rst syllable and the nal syllable
0
gets a p olarity tone. In A , the lexical H shifts b etween the rst and the second vowel
dep ending on syllabi cation. The initial L tone is lexical. In the singular it is realized on
the rst vowel, in the plural it gets trapp ed b etween the H tone and left edge, resulting in
0 11
a contour. B/B show the same e ect for lexical L tones. In C, there is no tonal evidence
either way: the words are toneless and get a derived [H] under b oth syllabi cations. In D,
!
all the exceptionally H-toned suxes are syllabi ed CVV; else wewould exp ect *pi . r u . u,
. . .
with the suxal H on the p enult plus a nal p olarity tone. As exp ected, these suxes do
not form falling contours under emphatic lengthening, but remain level.
The stress-sensitive tones are lexical. We are not aware of any cases where derived
tones, for example p olarity tones, would shift towards stress. The lexical status of the 19
shifting L tones is evident from the fact that they trigger p olarity. For the H tones,
wemay also concatenate the relevant ro ots with a H-toned adjective. The result is an
intervening downstep instead of H-suppression.
!
37 kpal- f a- a sling-bad-sg
!
kpar- f a- a shirt-bad-sg
!
ul- f a- a branch-bad-sg
Finally,we note a connection b etween H tone and emphasis. In Dagaare, as in many
other languages, certain lexical items come in emphatic/nonemphatic pairs; here we con-
sider pronouns and tense markers. Nonemphatic pronouns are all monosyllabic, hence
unstressed, and all b ear the default L tone. Their emphatic counterparts are all at least
disyllabic, hence stressed, and all b ear a H tone. The past and future tense markers
12
b ehave analogously.
38 Personal pronouns Bo domo 1997 :
Nonemphatic Emphatic Gloss
N, m a m a. a 1.sg
f u f u . u 2.sg
. . .
u u .n o 3.sg
. . .
1.pl .i ti ti.ni
. . . .
2.pl .i y e y e.ni
. . . .
b a b a.n a 3.pl.hum
a a.n a 3.pl.nonhum
39 Tense markers:
d a general past long ago past, past to day
d a. a 2{6 days ago
n a general future shall/will
n a. a b e going to
One alternativewould b e to prop ose that H tone is an underlying accent and L tone
absence thereof. This would mean that in yi.ri `house' the rst syllable would b e stressed
and in wi.ri `horse' unstressed. However, this is clearly the wrong analysis for Dagaare.
First, b oth lexical H and lexical L tones are stress-sensitive. Second, if H tone could b e
identi ed with stress, one would exp ect stress-related asymmetries b etween initial H and
L syllables. However, b oth p ermit the same syllable typ es CV, CVC, CVV, CVVC, b oth
participate in the same morphophonological length alternations n u u -ri/n u - o hen-pl/sg,
. . . . .
k u u-ri /k u- e ,hoe-sg/pl, b oth may take emphatic stress and b oth are equally common.
Thus, the contrast b etween H and L seems genuinely tonal and indep endent of stress. 20
2.6 Summary
Wenow summarize the diagnostics. For obvious reasons, test 3 do es not apply to L tones.
40
diagnostic lexical derived
1. Trigger p olarity yes no
2. Spread no yes
3. Undergo H suppression no yes
4. Trigger downstep yes no
5. Attracted by stress yes no
The same tone can b e sub jected to multiple diagnostics. Consider the H tone in
k unkun i `tortoise'. By all the applicable diagnostics, H is derived.
41 Testing the H tone:
Spreading, no p olarity k unkun- i *kunk un- i
!
No downstep k unkun iyiri *kunk un i yiri
!
Suppressed k unkun-f a a *kunk un- f a a
Not attracted by stress k un.ku.n i *kun.k u.n i
Wehavenow established empirically that Dagaare tones can b e assigned into two
mutually exclusive groups: lexical and derived. This is where the explanatory work
b egins. Why do these diagnostics work in the rst place? Why should these disparate
tonal prop erties cluster? This will b e the topic of the rest of this pap er.
3 Tonal corresp ondences
In this section, we derive the b ehavior of Dagaare nominal tones from optimal cor-
respondences between input tones, output tones and tone-b earing units. The idea
of de ning well-formedness in terms of declarative input-output constraints evaluated
in parallel was prop osed by Koskenniemi 1983 and given an optimality-theoretic in-
terpretation by McCarthy and Prince 1995 . The original idea has recently b een
extended to base-reduplicant McCarthy and Prince 1995 and output-output relations
Benua 1995 , Burzio 1996 . As McCarthy and Prince p oint out, the notion of corresp on-
dence is a very general one and straightforwardly extends to autosegmental asso ciations
such as tone{TBU asso ciations. This is of course not to say that they are the same rela-
tion. For example, input{output relations are further constrained by identity, tone{TBU
relations by synchrony. 21
3.1 Intro duction
We prop ose 42 as the optimal tone-TBU corresp ondence. These two constraints sub-
sume the core clause of most versions of the autosegmental Well-Formedness Condition
Goldsmith 1976 , Halle and Vergnaud 1982 , Pulleyblank 1986 .
42 Configuration Constraints
TBU a Every tone is linked to exactly one TBU = t,tbu
b Every TBU is linked to exactly one tone = tbu,t
T
It will b e immediatelyobvious that in most tone languages t,tbu and tbu,t
cannot b e surface-true. For example, b oth oating tones and tone spreading violate
t,tbu: in the rst case there is a tone without a TBU, in the second case the tone is
asso ciated to several TBUs. However, as so on as these two constraints are emb edded in
Optimality Theory Prince and Smolensky 1993 more reasonable results start to emerge.
The two p ossible rankings generate two kinds of languages: one with spreading and
oating tones, another with toneless TBUs and contours.
43 To o many TBUs. Solution: Spreading
tbu, t t, tbu
a tbu tbu
*
T
b tbu tbu
*!
T
44 To o many tones. Solution: Floating tone
tbu, t t, tbu
a tbu
*
T T
b tbu
*!
T T
45 To o many TBUs. Solution: Toneless TBU
t, tbu tbu, t
a tbu tbu
*!
T
b tbu tbu
*
T 22
46 To o many tones. Solution: Contour
t, tbu tbu, t
a tbu
*!
T T
b tbu
*
T T
These two simple corresp ondence constraints, emb edded in Optimality Theory, predict
four familiar kinds of tonal phenomena. This is a direct implementation of the intuition
that many tonal pro cesses result from a more or less successful alignment of the tonal
melo dy with the segmental melo dy. As Kenstowicz 1989 puts it:
West African systems with contour tones and additional pitch heights are probably
the surface re ex of a housing shortage{to o many high and low tonal atoms crowding
into the same syllabic space. Because of their greater size, Bantu words tend to have
the opp osite problem{a go o d deal of tonally uno ccupied space.
Four more tonal pro cesses need to b e addressed. In the traditional OT terminology,
tone deletion is a violation of ParseT, tone insertion a violation of FillT. Here we
state these constraints as corresp ondences b etween input tones and output tones. The
optimal corresp ondence is shown in 47. For the sake of notational clarity,we use an
arrow for input{output corresp ondence.
47 T output tone
T input tone
i
Two more constraints are now added to the system: t,t =every output tone corre-
i
sp onds to exactly one input tone and t ,t=every input tone corresp onds to exactly one
i
output tone. Besides tone insertion and tone deletion, these constraints militate against
twotyp es of multiple corresp ondence, less often discussed, but nevertheless predicted by
Corresp ondence Theory: tone fusion and tone ssion. In the segmental domain, these
pro cesses have b een dubb ed \Coalescence" and \Breaking" McCarthy and Prince 1995 ,
123-4.
48 Insertion Fusion Deletion Fission
a T b T c ; d T T
; T T T T
i i i i 23
Tone deletion and tone insertion are very common pro cesses. An example of b oth is
provided by Meeussen's Rule. As discussed earlier, one standard analysis is that a se-
quence of two adjacent H tones triggers the deletion of the second tone, yielding H;, which
through default insertion pro duces the surface sequence HL. Tone fusion is found for in-
stance in Kipare Bantu: Tanzania where adjacent underlying H tones merge into one H
tone due to the high-ranking OCP Odden 1995 , 462-3 and also in Shona Myers 1996 .
Convincing cases of tone ssion are harder to nd. One p otential example is tone copy-
ing instead of the usual asso ciation-cum-spreading Hyman and Pulleyblank 1988 . An
obvious explanation for the absence of ssion is the OCP.
49 Tone ssion unattested?
TBU TBU TBU
T T T
T
i
The four-constraint system is summarized in 50.
50 Summary of constraints
a t,tbu *Spread,*Float
b tbu,t *Contour,*Toneless
d t,t *Insertion,*Fusion
i
c t ,t *Deletion,*Fission
i
3.2 Some consequences
In Optimality Theory, a fresh set of constraints forces one to explore how they interact
under di erent p ossible rankings. In this case, the constraints make some quite sp eci c
predictions ab out the b ehavior of tones. While it is imp ossible to survey even a tiny
fraction of the relevant empirical material in a pap er on just one tone language, Dagaare,
we will b egin the work by sp elling out some of the more obvious consequences.
In the simplest situation where the numb er of tones and TBUs is identical and tone
linearization is not a ected byany factors b eyond the numb er of units, such as stress,
the result will b e a p erfect one-to-one mapping. After all, this is what all the constraints
strive for; there is no ro om for improvement.
51 TBU TBU TBU
T T T
T T T
i i i 24
Most tone languages deviate from this ideal in one way or another and, as a result,
various tonal pro cesses arise. We examine two p ossible deviations: i to o few tones ii
to o many tones. Our four-constraint system yields twenty-four p ossible tonal grammars
which di er minimally in how they treat anomalous situations.
In 52, two TBUs comp ete for one lexical tone. There are four p ossible repairs. Which
one is actually chosen dep ends on which of the four constraints ranks the lowest. The
corresp onding rankings are shown in 53.
52 a Insertion b Spreading c Toneless TBU d Fission
TBU TBU TBU TBU TBU TBU TBU TBU
T T T T T T
T T T T
i i i i
53 a f*Toneless,*Spread,*Fissiong*Insert
b f*Toneless,*Insertion,*Fissiong*Spread
c f*Spread,*Insertion,*Fissiong*Toneless
d f*Spread,*Insertion,*Tonelessg *Fission
In 54, wehave the converse situation: two lexical tones comp eting for the same TBU.
Again, four p ossible solutions arise out of the rankings in 55. The four constraints in 55
are the same constraints as in 53, only mnemonically relab elled.
54 a Deletion b Floating tone c Contour d Fusion
TBU TBU TBU TBU
T T T T T T
T T T T T T T T
i i i i i i i i
55 a f*Contour,*Float,*Fusiong*Deletion
b f*Contour,*Fusion,*Deletiong *Float
c f*Float,*Fusion,*Deletiong*Contour
d f*Float,*Deletion,*Contourg*Fusion
The system excludes certain tone con gurations as universally imp ossible. Since in OT
every universal claim is relative to some set of constraints { the Galilean syllable theory
of Prince and Smolensky 1993 b eing a prime example { we will call these predictions
soft universals. All else b eing equal they will hold, but indep endent higher-ranking
constraints may out them in well-de ned situations. 25
56 Soft Universal 1: Only lexical tones form contours.
* TBU
T T
T
i
57 Soft Universal 2: Derived tones don't oat.
* T
These soft universals derive from the fact that output tones may exist for two reasons:
to realize a lexical tone or to satisfy a TBU. Neither condition obtains here. In 56,
a tone has b een inserted simply to form a contour, but we know that contours are no
go o d and wewould b e b etter o without them. Thus, omitting the output tone can only
render the structure b etter. In 57, the output tone serves no purp ose whatso ever. All
it do es is violate the constraint*Float and by removing it we get a b etter i.e. empty
13
structure.
So far, wehave only considered T {T and T{TBU corresp ondences. The system
i
p ermits one more pair of constraints: t ,tbu = \each input tone corresp onds to exactly
i
one TBU" and tbu,t = \each TBU corresp onds to exactly one input tone". We will
i
assume that input tones can b e linked to TBUs only through an output tone; there is
thus no way to satisfy either constraint except by p ositing a mediating output tone. We
conclude this section by showing how these two constraints interact with the rest of the
grammar. Crucially, they will b e resp onsible for the lexical/derived distinction discussed
in the rst half of the pap er.
The fact that the same constraintworks against b oth no asso ciation \at least one
corresp ondent" and multiple asso ciation \at most one corresp ondent" predicts two
typ es of languages. The ranking tbu,t t,tbu predicts a language with oating
and spreading while t,tbu tbu,t predicts a language with contours and toneless
syllables.
58 Language typ e 1: Language typ e 2:
tbu,t t,tbu t,tbu tbu,t
TBU TBU TBU TBU TBU TBU
T T T T T T
T T T T T T
i i i i i i
Float Spread Contour Toneless TBU 26
The twotyp es of languages strikingly resemble prototypical tone languages and pro-
totypical stress languages, resp ectively. Floating and spreading are characteristically
\tonal" phenomena, rare in stress languages, thus one would generally exp ect true tone
languages to rank tbu,t t,tbu. On the other hand, b oth contours and toneless
TBUs are common in stress languages. For example, the English question tune M*LH
Lib erman 1975 , Hayes 1995 can b e mapp ed onto anything from a whole sentence as in
What's wr ong with your moth er? to a single TBU as in What!? . In the rst case, the
tones are far apart and the intervening syllables are phonologically toneless, in the second
wehave a contour. The hyp othesis is that the traditional descriptivetyp ology which
groups languages to tone vs. stress languages follows from the two p ossible rankings of
two corresp ondence constraints.
The following question now arises: since a language must inevitably cho ose one or the
other ranking, how can mixed systems exist at all? Yet we know there are languages with,
say, b oth spreading and contours, in fact Dagaare is one such language. In the context of
Optimality Theory, what this suggests is that one of the pro cesses is marked, forced by
some higher-ranking constraint, the other unmarked. Since Dagaare is a tone language,
we exp ect the ranking tbu,t t,tbu which predicts spreading and oating as the
unmarked options and contours only under sp ecial circumstances. There are strong hints
where to lo ok: i on theoretical grounds, we know that only lexical tones may form
contours; ii on empirical grounds, we know that only lexical tones do form contours in
Dagaare; iii of the six p ossible constraints, the twowehave not yet examined refer to
lexical tones. The problem is solved by ranking the constraint t ,tbu ab ove the rest.
i
59 Lexical tones don't oat, but form contours.
*Float = *Contour = *Float=
i
t ,tbu tbu,t t,tbu
i
a TBU
T T *
T T
i i
b TBU
T T *! *
T T
i i
Not only havewe explained why lexical tones may form contours, but also why lexical
tones do not spread. 27
60 Lexical tones don't spread.
*Spread = *Toneless = *Spread =
i
t ,tbu tbu,t t,tbu
i
a TBU TBU
T *
T
i
b TBU TBU
T *! *
T
i
The lexical tone will not spread since spreading an input tone is worse than having
a toneless TBU. The remaining solutions are spreading a derived tone none is present,
inserting a tone the actual solution in Dagaare, or leaving the TBU toneless in Dagaare,
b eing toneless is worse than insertion. Note that a derived tone will spread in the same
situation, another correct prediction.
61 Derived tones spread.
*Spread = *Toneless = *Spread =
i
t ,tbu tbu,t t,tbu
i
a TBU TBU
T *!
b TBU TBU
T *
Wenow sum up the argument. Being a tone language, Dagaare has the ranking
tbu,t t,tbu which predicts spreading in the case of overspace to o many TBUs,
to o few tones and oating in the case of underspace to o few TBUs, to o many tones. The
problem is that Dagaare o ccasionally solves the underspace problem by forming contours.
Our theory suggests that this has to do with the sp ecial b ehavior of lexical tones. The
constraintt ,tbu correctly predicts lexical contours and the fact that lexical tones do
i
not spread.
All six constraints require that there b e exactly one corresp ondent. It remains to b e
seen whether this is empirically to o restrictive. In that case, one could easily weaken the
theory by splitting each constraintinto two parts: `at least one' and `at most one'. `At 28
least one' is familiar from Goldsmith's 1976 Autosegmental Well-Formedness Condition,
restated in OT as ParseTone `Every tone has a TBU' and SpecTone `Every TBU
has a tone' Zoll 1996 or alternatively in terms of Max, Dep, Integrity and Unifor-
14
mity Myers 1996 , McCarthy and Prince 1995 . Making `at least one' and `at most one'
separate constraints gives one more descriptive exibility b ecause each half can b e ranked
indep endently. On the explanatory side, one will lose the prediction that lexical tones
b oth trigger p olarity and form contours. The observed connection would then b ecome
purely coincidental. The soft universals \Only lexical tones form contours", \Derived
tones don't oat" would continue to follow as they in no way dep end on whether the
constraints are split.
3.3 Tone and stress
Accounts of tone-accentinteraction usually p osit some principle such as the Tone-Accent
Attraction Condition of Goldsmith 1987 which guarantees that tones are preferably
asso ciated to accented syllables. In Dagaare, the strength of the attraction dep ends on
two factors: whether the tone is lexical or derived and whether it is high or low. Wehave
identi ed three typ es of stress-induced tonal e ects in Dagaare: default [H] tones, lexical
contours and tonal p olarity.We will now examine each one in turn.
We start by guring out the ranking of *Insertion by examining the toneless class
C where insertion is vital. The stressed syllable p enult is underlined.
62 Class C, toneless words
*Cont = *Ins = *Spr = [H] [L]
*Tless = t,t *Flo =
i
tbu,t t,tbu
a p o g o
. .
* * *
H
b p o g o
. .
* * *!
L
c p o g o
. .
*!* *
H L
d p o go
. .
*! * *
H
e po go
. .
*!* * *
Ranking *Insertion between *Toneless and *Spread yields the desired result.
Insertion is preferred over toneless TBUs as d and e show, but it must b e minimal: in 29
the optimal candidate only one tone is inserted and then spread. This contrasts with the
ungrammatical candidate c where b oth TBUs have their own derived tones.
We are left with the choice b etween a and b. This is no more a matter of nding
the optimal tone{TBU con guration: wemust decide b etween H and L tones. We prop ose
that stressed syllables universally prefer i lexical tones to derived tones and ii H tones
to L tones. It follows that lexical H is the universally b est and derived [L] the universally
worst choice for stressed syllables. The status of lexical L and derived [H] is set language-
sp eci cally; it is here that the two principles con ict.
63 Constraints:
a H = If stressed, then H.
b L = If stressed, then L.
c [H] = If stressed, then [H].
d [L] = If stressed, then [L].
64 Universal ranking: H [L]
Full ranking Dagaare H L [H] [L]
As long as the ranking *Contour [H] holds the correct result is guaranteed. The
opp osite ranking would predict a language where stressed syllables would always b ear a H
tone; if none were present lexically, a derived [H] tone would b e inserted even on pain of
contour. Since the data underdetermines the p osition of the subhierarchy[H] [L] within
the global ranking, we arbitrarily place it at the b ottom of the hierarchy. We further
assume that the tonal preference hierarchy for stressed syllables has a corresp onding
inverse hierarchy [L] [H] L H for unstressed syllables. Thus, all else b eing
equal, the optimal tone for unstressed syllables is [L]. For the purp oses of this pap er, the
inverse hierarchy is resp onsible for the assignment of default [L] to unstressed syllables.
Minimally, the constraint [L] and hence the rest of the hierarchy must rank b elow*Ins.
Ranking it ab ove*Ins would wrongly predict that toneless unstressed syllables get [L] by
insertion. In reality, the stress-induced [H] tone spreads on all syllables: /po g-o /`woman'
. .
15
! p o .g o.
. .
The following two tableaux show how stress plays a role in deriving contour
tones. We assume that lexical tones are underlyingly unasso ciated, asso ciation b e-
ing yet another consequence of constraintinteraction. In particular, we p osit no Uni-
versal Asso ciation Convention which seems correct Archangeli and Pulleyblank 1994 ,
Hyman and Ngunga 1994 , Tranel 1995 , Zoll 1996 . 30
65
H L *Spr = *Cont = *Ins = *Spr = [H] [L]
i
*Flo *Tless = t,t *Flo =
i i
t ,tbu tbu,t t,tbu
i
a kp a r u u
. .
L H L * * * *
L H
i i
b kp a ru u
. .
L H L *!* * * *
L H
i i
c kp a r u u
. .
L L H *! * * *
L H
i i
In kp aru u `guineafowl-sg', the lexical H tone is aligned with the stressed syllable which
. .
in turn forces a p olarity tone on the nal syllable. Candidate b motivates the ranking
*Contour L: lexical L tones are attracted to stress, but not at the exp ense of a
contour. Candidate c loses b ecause stressed syllables prefer H tones to L tones.
Tableau 66 shows the corresp onding plural which has fewer syllables. Given the
choice b etween stressed and unstressed, the lexical H tone cho oses the stressed syllable,
forming a contour. This con rms the ranking H *Contour: lexical H tones are
attracted to stress to the extent that they form contours if necessary. 31
66 The stress-sensitivity of lexical H tones
*Flo = H *Cont = L *Ins = *Flo = [H] [L]
i
t ,tbu tbu,t t,t t,tbu
i i
a kpar ri
.
L H L * * * *
L H
i i
b kp ar ri
.
L H L *! * * * *
L H
i i
c kp ar ri
.
L H *! * *
L H
i i
Perhaps the most characteristic phenomenon of Dagaare tonology, tonal p olarity, also
follows from stress. Tableau 67 shows in rather great detail why tonal p olarity exists. 32
67 Class A tonal p olarity
OCP *Spr = H *Cont = L *Ins = *Spr = [H] [L]
i
*Flo = *Tless = t,t *Flo =
i i
t ,tbu tbu,t t,tbu
i
a wi ri
L H * * * *
L
i
b wi L HL * **! * * * L i c wi ri H L * *! * * L i d wi ri L H * *! * * * L i e wi ri L * *! * * L i f wi ri L *! * * * * L i g wi ri H *! * * * * * L i h wi ri L L *! * * * * L i 33 The optimal candidate a wi-ri has simply acquired a p olarity tone on the toneless sux. In the ungrammatical candidate c *wi-ri the lexical L tone has migrated to the sux giving rise to \ro ot p olarity". This do es not happ en b ecause the stressed ro ot is more capable of realizing lexical contrasts than the unstressed sux. Another interesting failure is d *wiri where a derived [H] tone has spread onto the stressed syllable. While this satis es [H], it violates the higher-ranking *Contour; lexical tones may form contours, derived tones may not. Candidate g *wiri ignores the lexical tone altogether and p osits a derived [H] tone just as in the toneless word p og o. This fails b ecause lexical tones have . . to b e realized on some TBU. Finally, h *wiri fails b ecause of the OCP.For the time b eing, we will assume that the OCP refers to output tones and rank it at the top of the 16 hierarchy. In this section, wehave derived the following tonal phenomena: the insertion of [H] on toneless words; contours involving lexical H tones on stressed syllables; tonal p olarity. In all three, p enultimate stress turned out to play a crucial role. 3.4 Downstep and the OCP What happ ens when two lexical H tones collide due to morphological concatenation? As ! Class D shows, the result is an internal downstep H- H: 68 Class D internal downstep OCP *Flo = *Tless = *Ins = *Flo = [H] [L] H L i *Spr = *Cont = t,t *Spr = i i t ,tbu tbu,t t,tbu i ! a kp a a u . H LH * * * * * H H i i b kp a a u . H H *! * * * * * H H i i c kp a a u . H H *! * * * H H i i The oating bu er tone in a is a derived tone. While the Soft Universal 2 \Derived tones don't oat" do es hold with resp ect to the con gurational constraints, it is here sup erseded by an indep endent higher-ranking constraint: the OCP. This example also establishes the ranking OCP *Insertion.However, so far no conclusive evidence 34 for the derived status of the internal downstep has b een provided. In fact, b oth 69a and 69b were considered equally plausible: ! ! u 69 a kp a a - b kp a a - u . . H [L] H H L H At this p oint, this indeterminacy is resolved. If the bu er tone were lexical, it would ! not oat, but form a contour, resulting in *kp a a u due to the ranking *Float *Con- i . tour. Thus, the grammar selects the representation H-H complemented by [L] tone interp olation over HL-H whichwas prop osed as a p ossible alternative earlier. Example 70 shows that the OCP also triggers downstep p ostlexically, explaining the ! external downstep after H suxes. 70 The asso ciative construction: downstep H L OCP *Spr = *Tless = *Ins = i t ,tbu tbu,t t,t i i ! ! yi a kp a a- u - ri . H LH LH L ** *** H H H i i i ! b kp a a- u yi- ri H LH H L *! ** ** H H H i i i So far wehave b een assuming that the OCP applies at the level of outputs. The remaining examples show that this formulation is in need of revision. According to My- ers 1994 , 14, the OCP is sensitive to \parsed representational elements" whichinour framework translates to input tones with output corresp ondents. Unfortunately, this for- mulation will not work for our purp oses. Among other things, it would fail to trigger p olarityinwiri = L [H] where there is only one input tone with an output corresp ondent. We prop ose the following version of the OCP for Dagaare: 71 OCP Dagaare: A lexical tone may not b e followed by an identical tone. This formulation is more sp eci c than one would hop e note esp ecially its inherent directionality, yet it seems descriptively correct for Dagaare. First, it readily explains tonal p olarity which only o ccurs after lexical tones. Second, it captures b oth the presence of downstep in 70 and its absence in 72: 35 72 The asso ciative construction: no downstep OCP *Flo = H *Tless = L *Ins = i t ,tbu tbu,t t,t i i a wi ri yi ri L H H L * * ** L H i i ! b wi ri yi ri L H LH L * * ***! L H i i Third, Dagaare not only tolerates [L]-L sequences, but creates them freely in complex expressions. This further supp orts the hyp othesis that [T]-T do es not countasanOCP violation. In 73, p enultimate stress attracts the lexical L tone and the initial syllable receives [L] by default. The resulting [L]-L sequence thus do es not violate the OCP. 73 OCP *Spr = H *Tless = L *Ins = *Spr = [H] [L] i t ,tbu tbu,t t,t t,tbu i i a vi l a a . L L H * ** * * L i b vi l a a . L H L * *! ** * L i c vi l a a . L H * *! * * * L i 36 Another example comes from H tone suppression where the toneless ro ot pog- `woman' . is assigned a default [L] even when a lexical L follows. 74 p o g- o + t o o -ri ! p og-t o o-ri woman-far.away-sg ...... OCP *Spr = *Tless = *Ins = [H] [L] [L] H L i t ,tbu tbu,t t,t i i a p og- t o o- ri . . . . L L H * ** * * L i b p og- t o o- ri . . . . H L H * ** * * !* L i Unfortunately, the analysis do es not generalize to longer words where [H] is suppressed only in the last ro ot syllable, a problem we share with the Dissimilation Analysis. Thus, the descriptive generalization \no [H] tones on the TBU left of a N+A juncture" remains an explanatory challenge. 75 b ad e r-i +f a- a ! b ad e r-f a- a spider-bad-sg . . . 3.5 *L{L Finally,wehave solved an outstanding problem in the analysis of several Gur languages at least Dagaare, Dagbane and Mo ore: the systematic absence of L-toned nominals longer than one syllable. Thus, nouns like*s ar a ,*s ak ar a etc. are systematically missing. Consider *s ar a . The following p ossibilities arise: 1. The word is lexically toneless: /sara/. Being disyllabic, the word receives p enul- timate stress and a derived [H] tone. Result: s ar a . 2. The word has a single lexical L tone: /sara + L/. Due to p enultimate stress, the tone gets linked to the rst syllable and the second syllable b ecomes [H] by p olarity. Result: s ar a . 3. There are two lexical L tones, one on each syllable, prelinked or otherwise. In this case, the OCP interp olates a [H] tone. Result: sar a . There is thus no way to derive a noun like*s ar a . The analysis correctly predicts the p ossibility of L-toned monosyllables suchas b a `they' which are unstressed and for this reason receive a default [L]. 37 4 Conclusion In this pap er, wehave describ ed yet another language with b oth stress and tone. However, the theoretical signi cance of the Dagaare facts go es b eyond this discovery. Wehave shown that several pro cesses distinguish b etween lexical tones and derived tones and that this distinction can b e given a straightforward nonderivational interpretation in terms of Corresp ondence Theory. In a sense, lexical tones o ccupy the middle ground b etween stress and derived tones. Like stress, lexical tones do not spread, but shift to metrically strong p ositions and there is typically only one lexical tone p er word. Derived tones are more tone-like in that they dissimilate, spread and do not undergo metrically motivated shifts. Also, given the traditional idea that lexical tones are \earlier" than derived tones, the Dagaare facts are compatible with the following famous conjecture: I know of no facts that would contradict a weaker universal principle [...] that if the tonal phenomena in a language are governed by b oth kinds of rules [tone rules and accent rules, AA/AB], the accent reduction rules o ccur earlier in the grammar than the assimilations and dissimilations. If this conjecture is correct, then [. . . ] it will b e p ossible to sp eak of a language as having a pitch accent system up to some p oint in the ordering of its rules and having a tonal system from that p oint in the rules on. Languages could then b e classi ed according to how early in their grammars the p oint o ccurred at which they b ecame tone languages McCawley 1970 While many African tone languages seem to yield to analyses in terms of lexical or metrical accents, privative H tones, and other reduced systems, Dagaare has proved sur- prisingly rich with its stressed vs. unstressed syllables, lexical vs. derived tones and high vs. low tones. However, while conceptually indep endent, the three dimensions interact inaway which conforms to our exp ectations based on b etter known languages. In par- ticular, stress attracts lexical tones as in many African tone systems and high tones as in many stress systems. Wehave also derived the puzzling phenomenon of tonal p olarity from the interaction of p enultimate stress and lexical tones and shown that tonal p olarity and downstep are b oth OCP e ects and only o ccur in connection with lexical tones. Three kinds of tonal constraints have b een p osited: 1. Tone{TBU constraints. These simple and uniform constraints strive for a transparent one-to-one mapping b etween tones and TBUs. Many central tonal phenomena such as asso ciation, spreading, oating, tonelessness, contour forma- tion, insertion and deletion follow. Di erent rankings yield a set of p ossible tone languages. Among other things, the system predicts that spreading and oating vs. 17 toneless TBUs and contours should co-o ccur as default options. 2. Stress-to-tone constraints. In addition to pure con gurational constraints, the distribution of tone refers to prop erties of TBUs stressed, unstressed and tones themselves H, L. We prop osed H > L > [H] > [L] as the tonal preference hierarchy for stressed syllables. 38 3. OCP. The prop er understanding of this principle continues to b e the sub ject of active debate Myers 1994 , Leb en 1996 . For us, the OCP is crucial for downstep and tonal p olarity. Our main conclusion was that in Dagaare the OCP is sensitive to the lexical/derived distinction. We conclude by cross-classifying our predictions by tonal class and syllable structure. The four empty cells seem accidental gaps. 76 CVV CV.CV CVV.CV CV.CVV CV.CV.V CV.V CVC.CV toneless [H] [H] [H] [H] [H] f a. a p o.g o k u l.li k u .l a a . . . . . bad-sg woman-sg river-pl river-sg H H [L] H [L] H [L] H [L] [L] H [L] d a . a yi.ri n u o .ri e.b a a . . . . b eer-sg house-sg mouth-sg cro co dile-sg L L [H] L [H] L [H] L [H] [L] L [H] s a. a wi.ri vi e.li kp o.lu u vi.l a. a . . . . father-sg horse-sg good-pl termite-sg good-sg H-H H [L] H H [L] H H [L] H H [L] H H [L] H [L] ! ! ! i i. li z u. ri pi. r u u ...... head-pl horn-sg sheep-sg 39 App endix Singular Plural a d ankyini wall-sg d ankyim e wall-pl d aku r- a a male name-sg d aku r-ri male name-pl . . . k unku-n i tortoise-sg k unku-m o tortoise-pl k u m a asi Kumasi-sg name of a city k u m a asi-ri Kumasi-pl . . . . . b s aki -ri bicycle-sg s aki - e bicycle-pl . . . . t abu l-i table-sg t abu l- o table-pl . . . . spider-sg b ad er- e e spider-pl c b ad er-i . . . . . b angyi r- a a latrine-sg b angyi r-ri latrine-pl . . . t ak or- o window-sg t ak or-ri window-pl k ommi-ru u tomato-sg k ommi- e tomato-pl gb atir-u u kind of plant-sg gb atir- e e kind of plant-pl . . . . simi- e Bambara b eans-pl s ow ol- o kind of fo o d-sg s ow ol- e e kind of fo o d-pl ...... d ank ar a a Accra name of a city { { sipirii aspirin { { ay op oi seven-sg ay op onni seven-pl . . d ag ar- a a b eam-sg d ag ar-ri b eam-pl . d ag a a-ri Dagaare sp eaker-sg d ag a a-b a Dagaare sp eaker-pl . l ez ar e twenty-sg l ez e e twenty-pl . . di nd ori do or-sg di nd o e do or-pl ...... fi ntil e lamp-sg fi ntilli lamp-pl ...... k on o-ri farming-sg k on o- e farming-pl ...... n o oti -ri sho e-sg n o oti e sho e-pl ...... s aku u -ri scho ol-sg s aku - e scho ol-pl . . . . . k ar ansu -ni kerosine-sg k ar ansu -m o kerosine-pl . . . . sitim e steamer-sg sitim e-ri steamer-pl birigyi bridge-sg birigyi-ri bridge-pl d em okir a asi demo cracy-sg d em okir a asi-ri demo cracy-pl . . . . . e f umi-ni needle-sg f umi- e needle-pl f b ogti bucket-sg b ogti-ri bucket-pl b u os u bus-sg b u os u -ri bus-pl ...... k op u cup-sg k op u-ri cup-pl . . . . . shirt-pl -ri shirt-sg s e eti s e eti ...... tu r oz e trousers-sg tu r oz e-ri pairs of trousers ...... w a akyi watch-sg w a akyi-ri watch-pl . . . b ol o nti b elt-sg b ol o nti-ri b elt-pl ...... b ol o nti e b elt-pl . . . . w al ansi wireless set-sg w al ansi -ri wireless set-pl . . . l ati riki electricity-sg l ati riki-ri electricity-pl ...... 40 Notes An early version of this pap er was presented at the the Stanford Phonology Workshop, De- cemb er 5th, 1995. We esp ecially thank Edward Flemming, Vivienne Fong, Brett Kessler, Paul Kiparsky and Will Leb en for comments. All errors remain our resp onsibility. 1 The sp elling variant Dagaari is common in the literature. The dialect describ ed in this pap er is the Central dialect sp oken in and around Jirapa, northwestern Ghana. The native sp eaker judgments are those of the second author. An overview of the struc- ture of Dagaare is given in Bo domo 1997. Of other sources, wewant to mention es- p ecially Kennedy's 1966 eldnotes which constitute a brief, but valuable guide to the basic phonology of the Central dialect. Related languages/dialects of which phonologi- cal descriptions exist include Dagara Delplanque 1983, Wul eP enou-Achille 1982, Dag- bane Dagbani Wilson and Bendor-Samuel 1969, Hyman 1993, Olawsky 1996 and Mo ore Kenstowicz et al. 1988. In 1960, the numb er of Dagaare sp eakers was estimated to b e 205,000 Wilson and Bendor-Samuel 1969. Kennedy 1966 rep orts 201,680 sp eakers in Ghana and an additional 90,000 in Upp er Volta Burkina Faso. More recent gures include 311,000 in Ghana and 190,000 in Burkina Faso Grimes 1988 and 423,200 Grimes 1992. The latest estimate is 1,000,000 and over Bo domo 1997. This gure includes b oth Ghanaian and Burkina Dagaaba. The standard orthography do es not represent tone at all. For us, V = H tone, V = L tone, /i, e, o, u/ = [+ATR], /i ,e,o,u/ = [{ATR]. The vowel /a/ is [{ATR], but as it do es not . . . . have a [+ATR] counterpart we simply write /a/ without the dot. Underscore /a/ indicates nasalization. 2 The systematic absence of LL monotones only holds of nominals. Among verbs, LL is the hallmark of certain paradigms: b ongi `to know', g o lli `to go around', iri `to get up', fi e `to ...... whip', ogi `to vomit'. Even the derivational relationship b etween nouns and verbs is sometimes expressed tonally as LH p olarity vs. LL: 77 n a a chief, rich p erson n a a to get rich, imp ortant m u u dumb p erson m u u to get dumb n u o sweet, sweetness, joy n u o to grow fat . . . . ti e tree ti e to supp ort . . . . e to cut, slide op en e farm, clearing wi wi . . . . We note the following caveats: i The nominalizing sux -Ung `-ness' is inherently L. Combined with a L ro ot, this yields L{L as in vi el-u ng `go o d-ness'; ii The word t o ori {t ob o ...... `tobacco', apparently from the French tabac needs a sp ecial account; iii Finally, Kennedy 1966, 43 cites the forms b an a `lizards' and p a:r e `b ottom'. In the dialect we are describing, these words are pronounced b o nni and p ari . . . . The data also show another p eculiar characteristic of Dagaare morphology. The same sux e.g. -rImay mark either singular or plural dep ending on the stem. Thus, wehave k u u-ri ho e- sg k u- e ho e-pl, but ti- e tree-sg ti i-ri tree-pl.Even more strikingly, the same word may b ehave . . . . . di erently in geographically adjacent dialects: pii-ri ro ck-sg pi- e ro ck-pl Central dialect, pi- e ro ck-sg pii-ri ro ck-pl Southern dialect. 3 This ignores recent loans suchass e eti `shirt' and t e esi `test'. ...... 4 One comp ound typ e pre xes a CV-ro ot to a word. Some of the following examples come from Kennedy 1966, 44: 41 78 a l ajbi-ri small axe-sg < l a-ri +bi-ri axe-sg + seed-sg . nijmi-ri eye-sg < niN- e+ bi-ri face-sg + seed-sg s ojbi-ri path-sg < s o-ri+ bi-ri road-sg + seed-sg ! ! b bijny og- r o child snatcher < bi- e+ ny og- r o child-sg + snatch-impf bijt u u-r o child follower < bi- e+ t u u-r o child-sg + follow-impf We note that while adjacent lexical H tones usually induce intervening downstep, here we ! have l ajbi-ri instead of *l aj bi-ri . The tonal p erturbations in b also remain unanalyzed. The tonal phonology of Dagaare comp ounds awaits further study. An assortment of long words, b oth simple and comp ound, can b e found in the app endix. 5 The Dagaare pattern is exactly the opp osite of Bantu Goldsmith 1988. Proto-Bantu was most probably a free tone language where all the p ossible tone patterns were attested in disyllables: HH, HL, LH, LL. Later, the contrast b etween HL and HH was neutralized in many of its descendants, presumably a re ex of Meeussen's Rule HH ! HL which has precisely this e ect synchronically Goldsmith 1988, 84. Dagaare on the other hand retains the HL vs. HH contrast whereas the LH vs. LL contrast is absent. 6 Pulleyblank defends a subtle variant of the dissimilation analysis couched in Lexical Phonol- ogy. According to him, the p olarizing Margi clitics are underlyingly H, but [+extratonal]: if they o ccur next to H, they are deleted by Floating H Deletion an OCP-related rule, followed by default tone L insertion. Next to L, the H tone surfaces as is. Interestingly, Pulleyblank p. 210 notes that the ro ot tone never spreads onto the p olarity sux, not even when the sux is sandwiched b etween ro ot and another sux. His explanation is that \sub ject clitics are exceptions to bracket erasure and that tone-spreading do es not cross b oundaries p ost-lexically." However, he needs to assume that in the rst and third p erson sub ject clitics, bracket erasure do es apply p. 211. There are thus two parts to the explanation p. 212: i extratonality; ii failure of bracket erasure. If Margi is at all like Dagaare, we could p erhaps account for this b ehavior bysaying that lexical tones do not spread. 7 The nal [H] is optionally suppressed, yielding s a as a a `father's father', s a as a avil- a a . `father's go o d father' and s a avil- a as a a `go o d father's father'. . 8 ! It deserves to b e noted that Kennedy 1966, 46 do es not transcrib e downstep after - H suxes although he do es transcrib e downdrift after L toned suxes; in everything else we fully agree with his transcriptions. 9 We note the following V-initial words which are not generated by the prop osed template: a `de nite article', ab o o `which', ane `and', am e `these', an a `they', ang `who', aw ol a `how many' . grammatical markers, pronouns; ayi `2', at a `3', ay op oi `7' numerals; ammesiere `excellent!', . ab a `oh!', ai `no!' interjections; i `to do', eng `I'; u `he/she/it', u n a `he/she/it, emph.', o o `yes'. . . . . While most of these words are clearly semantically sp ecial, there are some V-initial nouns and ! verbs as well e.g. ii li `horn'. C-initialitythus seems a tendency at b est. .. . 10 In Dagara, similar segmental restrictions are observed: \En n, la plupart des opp osi- tions interviennent a l'initiale, du fait de la distribution lacunaire de nombreux phonemes, elle- m^eme due aux nombreuses neutralisations intervenantauniveau du mot" Delplanque 1983, 50. From the related Wul e, weeven have one native sp eaker rep ort of rst syllable ac- cent: \Ind ep endemment de la nature des sch emes tonals, la premi ere syllab e CV- du \mot phonologique" Wul e est toujours accentu e e. Cet accent met en valeur CV- par rapp ort a-Vou a -CV, il joue donc un r^ole culminatif " P enou-Achille 1982, 378-9. 42 Kennedy's segmental co-o ccurrence restrictions are p erhaps to o strong. In addition to the onset s listed by Kennedy,wehave found instances of /b/, /y/ and /t/: d o .b o `men', a.yi `2', 2 . . a.t a `3'. Recent loans are even more p ermissible: k o.pu `cup' ngmor.f o `gun', bi.ri.si `brick', tu r o.z e `trousers', w a a.kyi `watch', bi.ri.gyi `bridge', l a.ti.ri.ki `electricity'. Similarly,words ...... like b a al-b a `sick p erson-pl' show that the /gr, gl/ restriction, while p erhaps representative, do es not hold p erfectly even in the nativevo cabulary. 11 0 0 One p ossible analysis of the A/A and B/B distinction is that b oth syllabi cations, CV.V and CVV, are phonologically p ermitted, but morphology selects one over the other as follows: i Assume that the singular sux /-UU / b ears a lexical accent which attracts stress; If acc penultimate stress = a syllabic tro chee at the right edge is an inviolable constraint, we derive the syllabi cation kp a.r-u . u; ii In kp o.l-u u , the di erent sux /-UU/ is unaccented. Given . . that only ro ots and accented suxes are stressable, we derive initial stress. 12 The marker z a a `exactly past yesterday' do es not have a corresp onding \weak" memb er. 13 Both universals may b e violated in sp eci c conditions. Examples from Dagaare include ap o cop e see section 2.5 in cases like w og-i ! w^og- tall-sg. Here, the H part of the contour is a lexical tone, the L part a derived p olarity tone. In this case, an input TBU has b een suppressed on the surface, a sp ecial situation which o ccurs variably in p ostlexical phonology. The solution is to distinguish b etween input TBUs and output TBUs. As for oating derived tones, consider downstep: H [L] H. Here the sp eci c condition is the high-ranking OCP. 14 Our constraints are easily translatable into McCarthy and Prince's system: 79 Maxt,tbu =For all T, there's at least one TBU. =*Float Dept,tbu =For all TBU, there's at least one T. =*Toneless Integrityt,tbu =For all T, there's at most one TBU. =*Spread Uniformityt,tbu =For all TBU, there's at most one T. =*Contour Maxt ,tbu =For all T , there's at least one TBU. =*Float i i i Dept ,tbu =For all TBU, there's at least one T . =*Toneless i i i Integrityt ,tbu =For all T , there's at most one TBU. =*Spread i i i Uniformityt ,tbu =For all TBU, there's at most one T . =*Contour i i i Maxt ,t =For all T , there's at least one T. =*Deletion i i Dept ,t =For all T, there's at least one T . =*Insertion i i Integrityt ,t =For all T , there's at most one T. =*Fission i i Uniformityt ,t =For all T, there's at most one T . =*Fusion i i t,tbu = *Spread & *Float tbu,t = *Contour & *Toneless t ,tbu = *Spread & *Float i i i tbu,t = *Contour & *Toneless i i i t ,t = *Deletion & *Fission i t,t = *Insertion & *Fusion i 15 Indep endent evidence for default [L] in Dagaare comes from the optional p ostlexical pro cess of vowel insertion which breaks up consonant clusters. The ep enthetic vowel gets its tone by spreading from the tone on the left, if any, else [L] by default. 43 80 a b og.ti b o.g[ o].ti `bucket' b b or.f o b o.r[i].f o `English' . . . . . c mal.ki m a.l[i].ki `angel' . . . d tr o.z e t[u ].r o.z e `trousers' default [L] . . . . . As it stands, our analysis do es not predict the rightward spreading of lexical tones. The data suggests that *Spread ranks lower in p ostlexical phonology, p ermitting the spreading of i lexical tones. The remaining question is why leftward spreading is blo cked in favor of default [L]-insertion. So far, directionality of spreading has b een of no substantial consequence. The examples at hand may show that directionality is not generally reducible to tone{TBU and stress{tone constraints. For a recent discussion, see Zoll 1996. 16 Typ e yiri = Class B is simply the mirror image of Class A and is derived analogously. 17 Of these constraints, we did not discuss two: *Deletion = *Fission = t ,t =\For ev- i ery input tone, there is exactly one corresp onding output tone" and *Contour = *Toneless i i = tbu,t =\For every TBU, there is exactly one corresp onding input tone". To our knowl- i edge Dagaare never deletes lexical tones, thus *Deletion = *Fission is undominated. The constraint *Contour = *Toneless requires that every TBU have a lexical tone instead of i i a derived one. If undominated, this constraintwould among other things force the spreading of lexical tones, an unwelcome consequence. In Dagaare, this constraint ranks low enough to b e invisible. Bibliography Archangeli, D., and D. Pulleyblank. 1994. Grounded Phonology. Cambridge, Mas- sachusetts/London, England: The MIT Press. Benua, L. 1995. Identity e ects in morphological truncation. In J. N. Beckman, L. W. Dickey, and S. Urbanczyk Eds., Papers in Optimality Theory, 77{136. UMass, Amherst: GLSA. Bickmore, L. S. 1995. Tone and stress in Lamba. Phonology 12:307{341. Bo domo, A. B. 1997. The Structure of Dagaare. Stanford Monographs in African Languages. CSLI Publications. Burzio, L. 1996. Multiple corresp ondence. Pap er presented at the BCN Workshop on Con icting Constraints, Groningen, July 5, 1996. Dakubu, M. E. K. 1982. The tones of Dagaare. Ms., Language Centre, University of Ghana, Legon, Accra. Delplanque, A. 1983. Phonologie transformationnel le du dagara.Soci et ed' etudes linguistiques et anthrop ologiques de France SELAF. Goldsmith, J. 1976. Autosegmental Phonology. PhD thesis, Massachusetts Institute of Technol- ogy, Cambridge, Mass. Goldsmith, J. 1983. Accent in Tonga: An autosegmental account. In I. R. Riho Ed., Current Approaches to African Linguistics, Vol. 1, 227{234. Dordrecht: Foris. 44 Goldsmith, J. 1987. Tone and accent, and getting the two together. In BLS 13, 88{104. Goldsmith, J. 1988. Proso dic trends in the Bantu languages. In H. van der Hulst and N. Smith Eds., Autosegmental Studies on Pitch Accent, 81{93. Dordrecht: Foris. Grimes, B. F. Ed.. 1988. Ethnologue: Languages of the world. Dallas, Texas: Summer Institute of Linguistics, Inc. 11th edition. Grimes, B. F. Ed.. 1992. Ethnologue: Languages of the world. Dallas, Texas: Summer Institute of Linguistics, Inc. 12th edition. http://www.sil.org/ethnologue/ethnologue.html. Hall, E. 1977. Dagaare. In M. K. Dakubu Ed., West African Language Data Sheets, Vol I. The West African Linguistic So ciety. Halle, M., and J.-R. Vergnaud. 1982. On the frameworks of autosegmental phonology. In H. van der Hulst and N. Smith Eds., The Structure of Phonological Representations. Dor- drecht: Foris. Hayes, B. 1995. Metrical Stress Theory. Chicago and London: The University of Chicago Press. Ho mann, C. 1963. AGrammar of the Margi Language. London: Oxford University Press. Hyman, L. 1978. Tone and/or accent. In D. J. Nap oli Ed., Elements of tone, stress and intonation, 1{20. Washington DC: Georgetown University Press. Hyman, L. 1993. Structure preservation and p ostlexical tonology in Dagbani. In S. Hargus and E. Kaisse Eds., Phonetics and Phonology, Volume 4: Studies in Lexical Phonology, 235{54. San Diego, California: Academic Press. Hyman, L. M., and F. X. Katamba. 1993. A new approach to tone in Luganda. Language 69:34{67. Hyman, L. M., and A. Ngunga. 1994. On the non-universality of tonal asso ciation `conventions': evidence from Ciyao. Phonology 11:25{68. Hyman, L. M., and D. Pulleyblank. 1988. On feature copying: parameters of tone rules. In L. M. Hyman and C. N. Li Eds., Language, speech and mind, 30{48. Routledge. Kanerva, J. 1989. Focus and Phrasing in Chichewa ^ Phonology. PhD thesis, Stanford University, Stanford, California. Kennedy, J. 1966. Col lected Field Reports on the Phonology of Dagaari. The Institute of African Studies, University of Ghana. Kenstowicz, M. 1989. Accent and tone in Kizigua { a Bantu language. In P. Bertinetto and M. Lop o caro Eds., Certamen Phonologicum 1, 177{88. Torino: Rosenb erg & Sellier. Kenstowicz, M., E. Nikiema, and M. Ourso. 1988. Tonal p olarityintwo Gur languages. Studies in the Linguistic Sciences 181:77{103. Koskenniemi, K. 1983. Two-Level Morphology: A General Computational Model for Word-form Recognition and Production. PhD thesis, University of Helsinki. 45 Leb en, W. R. 1973. Suprasegmental Phonology. PhD thesis, Massachusetts Institute of Tech- nology, Cambridge, Mass. Leb en, W. R. 1996. The status of the OCP in the theory of tone. Pap er from ACAL 17, University of Florida, March 29-31, 1996. Lib erman, M. 1975. The Intonational System of English. PhD thesis, Massachusetts Institute of Technology, Cambridge, Mass. McCarthy, J., and A. Prince. 1995. Faithfulness and reduplicative identity. InJ.N.Beckman, L. W. Dickey, and S. Urbanczyk Eds., Papers in Optimality Theory, 249{384. UMass, Amherst: GLSA. McCawley, J. D. 1970. Some tonal systems that come close to b eing pitch accent systems but don't quite make it. In CLS 6, 526{32. Chicago: CLS. McCawley, J. D. 1978. What is a tone language? In V. Fromkin Ed., Tone: A Linguistic Survey, 113{131. New York: Academic Press. Myers, S. 1994. OCP E ects in Optimality Theory [1]. Ms., UniversityofTexas. Myers, S. 1996. OCP E ects in Optimality Theory [2]. Ms., UniversityofTexas. Myers, S., and T. Carleton. 1996. Tonal transfer in Chichewa. Phonology 13:39{72. Newman, P. 1995. Hausa tonology: Complexities in an "easy" tone language. In J. A. Goldsmith Ed., The Handbook of Phonological Theory, 762{781. Oxford: Blackwell. Odden, D. 1995. Tone: African languages. In J. A. Goldsmith Ed., The Handbook of Phono- logical Theory, 444{475. Oxford: Blackwell. Olawsky, K. J. 1996. An Introduction to Dagbani Phonology.Dusseldorf: Heinrich Heine Uni- versitat. P enou-Achille, S. 1982. Syst ematique du signi ant en Dagara: vari et eW ul e.Paris: Editions L'Harmattan. Prince, A., and P. Smolensky. 1993. Optimality Theory: ConstraintInteraction in Generative Grammar. Rutgers University, New Brunswick, and University of Colorado, Boulder. Pulleyblank, D. 1983. Accent in Kimatuumbi. In J. Kaye, H. Ko opman, D. Sp ortiche, and A. Dugas Eds., Current Approaches to African Linguistics. Dordrecht: Foris. Pulleyblank, D. 1986. Tone in Lexical Phonology. Dordrecht: Reidel. Rialland, A. 1985. Le ni/in ni ou l'armation/l'interrogation en moba langue voltaique parl ee au nord-togo. Studies in African Linguistics 258{61. Supplement9. Tranel, B. 1995. On the status of the Universal Asso ciation Conventions: Evidence from Mixteco. In BLS 21. 46 van der Hulst, H., and N. Smith. 1988. The variety of pitch accent systems: Intro duction. In H. van der Hulst and N. Smith Eds., Autosegmental Studies on Pitch Accent, ix{xxiv. Dordrecht: Foris. Wilson, W. A. A. 1970. External tonal sandhi in Dagbani. African Language Studies 405{16. Wilson, W. A. A., and J. T. Bendor-Samuel. 1969. The phonology of the nominal in Dagbani. Linguistics 56{82. Zoll, C. 1996. Con icting directionality. Ms., MIT/ROA-151. 47