Ross, Malcolm. 2010. "A preliminary analysis of East Kewa Tone." In Kenneth A.

McElhanon and Ger Reesink, A Mosaic of languages and cultures: studies celebrating the
career of Karl J. Franklin, 294-309. SIL e-Books, 19. [Dallas]: SIL International. http://
www.sil.org/silepubs/abstract.asp?id=52526

17
A Preliminary Analysis of East Kewa Tone
Malcolm Ross
ABSTRACT
This paper provides a preliminary analysis of the tone system in East Kewa. The domain
of tone in East Kewa is the morpheme, but like other New Guinea Highlands languages
tone sandhi applies such that the tone of each morpheme affects the tone of the following
morpheme. Like languages elsewhere in the world, East Kewa nouns display a larger
paradigm of tone choices than do its verb roots. Verb words, however, are made up of a
verb root and a desinence, each with their own tone, and the combination of the two
tones brings complications of its own.

1 Introduction
These notes are the beginnings of a description of tone in East Kewa (Engan
family, Trans New Guinea, located in the Southern Highlands Province of Papua
New Guinea). This preliminary analysis would have been more difficult, had it not
been helped on its way by the account of tone in West Kewa by Karl and Joice
Franklin (Franklin 1971, Franklin and Franklin 1962, 1978), but there are
apparently quite complex differences of detail between the tone systems of the
two dialects and no attempt is made here to say anything about these differences.1
Mark Donohue (1997) divides New Guinea tone systems into three broad types,
each rather diverse and divided from each other only by fuzzy boundaries:
1. syllable-level tone: the system assign[s] a separate tone to each syllable in
a word;
2. word-level tone: the domain of each tone is the word as a whole;

It is a pleasure to offer this paper in honour of Karl Franklin, whom I admire both as a scholar and as a brother.

294

East Kewa tone

295

3. pitch accent: there is one simply designated syllable that determines the
shape of the pitch pattern on the rest of the word.
On this categorisation East Kewa is a word tone language, although, as noted in
5, this is something of an oversimplification. In this respect it seems to be fairly
typical of Papua New Guinea Highlands languages of the Trans New Guinea
family (Ross 2005).
In the analysis of East Kewa tone a distinction between lexical tone (i.e., word
tone) and utterance tone is necessary. Lexical tone is the tone melody which a
lexical item bears in the lexicon. Utterance tone consists of the tones which
occur in the utterance as a result of tone-altering rules that apply to lexical items
in context. A full description of utterance tone would also include the
contribution of intonation.
The preliminary and partial analysis presented below is based on elicited data
from a single informant, Apoi Yarapea of Mugumapu village, who holds a Ph.D.
in linguistics and also helped to establish the tonal contrasts and to categorise
the data. The recorded items were analysed with the help of the software
application Praat.
Two systems are used to represent tone in this paper. Phonemic lexical tones
are usually written as H (high) and L (low). M is used for phonetic mid tone in
certain cases where H and L are neutralised. High tone is also written with an
acute accent, low tone with no accent. In general, H, L and M are used to
represent tone patterns in the abstract, and the acute accent is used
orthographically. No attempt is made to represent lexical and utterance tone
differently, as it is generally clear from context which is intended.

2 Non-tonal phonological basics

2.1 The phonological word
A phonological word consists of a simple or compound lexical word plus any
enclitics (1).
(1)

a . n=me ( FREE:1SG=ERG) I
b . s ground + yaa sky < syaa earth and sky

2.2 Interaction of prosody and vowel length

The minimal word is a monosyllable of two moras. In a curious asymmetry,
there are five underlyingly monomoraic vowels /i e a o u/ and one invariably
bimoraic vowel /aa/. However, when a phonological word is monosyllabic, its
vowel spans two moras, and the distinction in length between /i e a o u/ and
/aa/ is neutralised.

296

Malcolm Ross

Bimoraic /aa/ is always low central, and is lower than /a/ when the latter is
lengthened in a monosyllable.
A long vowel is written as a sequence of two like vowels. This allows a
contour tone, i.e., HL or LH, to be written using only the acute accent. Bimoraic
/aa/ and lengthened /a/ are not distinguished orthographically, as more work is
needed to check the distinction between them in individual items.

3 Tones and their manifestation

3.1 Tone and the phonological word
It is helpful to distinguish between a tone melody, whose domain is the
phonological word, and a tone, whose domain is the syllable. Thus kdo scar
has an HL melody consisting of two tones, H on the first syllable, L on the
second. In a monosyllable, both tones occur in sequence on the single
phonetically lengthened vowel: y bird. In a trisyllable, on the other hand, the
HL melody is spread across the three syllables. In lmogo pupil of eye the first
syllable has H tone, the last L, whilst the middle syllable is usually M.
The Franklins suggest that West Kewa has four tone melodies, H, L, HL, and
LH. The HL and LH melodies are imposed on the single syllable of a monosyllable
and across the syllables of a polysyllabic phonological word. This analysis works
well for East Kewa if it is restricted to root forms, but not when it is applied to
verb + desinence combinations, each of which is a phonological word, since
some such combinations in verb class 2 have an LHL melody.
3.2 Tone cues
The hearer identifies H and L tones from pitch movement cues. Utterance-initial
tone is generally neutralised to mid tone, but in East Kewa, neutralisation of H is
sometimes only at the beginning of a syllable, so that there is a distinction between
ip 3SG free pronoun and p 3DL free pronoun.2 The first syllable of p begins
at mid-level but rises to H, whereas the first syllable of ip does not rise.3
The initial mid-level sets a benchmark pitch. Thus when one of the adverbs
(k)abala yesterday or (k)bl before, first,4 occurs at the beginning of an
utterance, the former is roughly MML, the latter MMH.
I write roughly in the previous paragraph for two reasons. First, a
phonological word is stressed on its first syllable, with secondary stress on
alternate following syllables.5 The tone on intervening unstressed syllables
In West Kewa, utterance-initial syllables are mid tone (Franklin and Franklin 1978:18, 28), and there is no distinction
between the pronouns, where both are nip (Franklin 1971:34).
3
Where items differing only in tone melody are elicited in contrasting pairs, there also seem to be instances in table 2
where initial H tone is not neutralised. However, there is one instance where elicited ki hand (toneless) seems to have
initial H, and it may be that utterance-initial H (as opposed to MH over the first syllable) also represents neutralization.
4
Forms with and without initial k- occur. The difference appears to be dialectal.
5
There are exceptions to this generalisation, at least among verbs, but I do not have enough relevant data to sort them out.
2

East Kewa tone

297

appears to be indeterminate. Second, a variety of local phonetic factors influence

syllable pitch, among them syllable-initial consonants.6
Once a benchmark pitch is set, the cue for pitch recognition is relative pitch.
Table 1 shows approximate pitch relativities. Because these are relativities, a long
straight line simply says that both syllables have the same pitch: it says nothing
about absolute level.
Table 1: Pitch cues for tones
within a word
HH
HL
LL
LH

____ or __
__
____
__

Across a word
boundary
____ or __ or __
__
____
__

utterance
finally
____
__
__
n/a

Certain comments and generalizations can be made about table 1:

1. A sequence of like tones has roughly the same pitch, with three exceptions:
a. in an HH sequence,
i. word-internally the second H may be higher than the first;
this seems to be an instance of upstep, i.e., an
additional way of signaling H;
ii. across a word boundary one H may be higher than the
other: the higher H seems to mark focus (cf. 6);
b. in an utterance-final LL sequence the second L is lower than
the first and functions as a boundary tone.
2. There are two degrees of fall, which are in contrast only utterancefinally, distinguishing HL from LL; but the distinction is normally
redundant since the height of the first tone of the sequence has been
established and the second is L regardless of the degree of fall.
3. There is no ambiguity in the system provided that the height of the
first tone of the sequence has been established (e.g., within a word
HH and LH may both be marked by a rise, but its function in both
cases is to mark the second tone as H).
4. Utterance-final LH does not occur in my data set.
Example (2) contains a sequence of H tones across word boundaries,
and m taro is higher than both the preceding and the following
high-tone syllables. As indicated in the generalization (1aii), this is
apparently because the speaker treats it as the focus item of the utterance.
If this is so, then it is an instance of an intonation feature superimposed
on lexical tone.
6
For example, the Franklins comment regarding West Kewa that in an HH or LL sequence across either /p/ [] or /r/
[] the first tone displays a slight risefall (Franklin and Franklin 1978:18).

298

(2)

Malcolm Ross

kbl
m da-wa
First
taro see-3S. N R P A S T
First I saw the taro.

Citation-form utterances (both declarative and polar-interrogative)

display falling pitch movement, i.e., the benchmark around which H and L
vary falls, especially after the focus. This appears to represent declination,
marking the final part of an intonation unit. If the utterance is broken
into two parts as in (3), where there is a pause between subject + ERG and
the predicate, then pn-m begins at a somewhat higher pitch than it
would otherwise, representing a the start of a new intonation unit.
(3)

m=m
pn-m
F R E E : 3 P = E R G cut-3 P . N R P A S T
They cut (it).

Pitch rises and falls between tones are generally manifested phonetically
within the second of a sequence of two unlike tones. Thus on an LH
sequence, the pitch rise occurs in the course of the syllable on which H
occurs (or on an intervening unstressed syllable). Towards the end of an
utterance, however, a HL sequence may show an anticipatory drop during
the syllable on which H falls. Systemically this is perhaps attributable to
intonation rather than to tone. 7 Table 1 shows a curious asymmetry in the
interaction between tone and intonation. Most of my examples have
apparently declarative intonation. One would expect a final L boundary
tone. This occurs in utterance-final HL and LL sequences (LL sequences
displaying a fall which does not occur elsewhere), but it does not occur in
utterance-final HH sequences. On the other hand, focus may be marked by
an additional rise in HH sequences, as shown in (2), but apparently not in
any sequence which includes a low tone. 8

4 Lexical tone-melody classes

The tone-melody classes of nouns, personal pronouns and verbs are
considered in the three subsections below.

The Franklins make the following generalisations involving mid tone, but these follow from the description here,
which does not require M tone other than utterance-initially since tones are cued by relative pitch differences:
1. a phonemic H between two L is mid, i.e., LHL < [LML].
2. a phonemic L between two H is mid, i.e., HLH < [HMH].
3. a syllable between an H and L is mid, i.e., HXL < [HML].
8
A huge amount of work remains to be done on Kewa intonation. For example, citation forms of yesno questions show
the same tones as their declarative counterparts, but with an increased pitch range.
7

East Kewa tone

299

4.1 Tone-melody classes of nouns

Table 2 displays the nouns for which I have adequate tonal data. They are
arranged vertically in five tone-melody classes numbered 0, 1, 2, 3 and 4.
The table has eight columns, the fourth to the eighth showing the tone
melodies manifested by each item in five different contexts. Some items
show unexplained discrepancies; these may reflect incorrect analysis or
glitches in elicitation.
The fourth and fifth columns of table 2 show respectively the melodies
of lone citation forms and of instances occurring initially in longer
utterances. The two melodies match except in the case of class 1, where
the citation melody is M, the utterance-initial melody MH. This is
unsurprising, since both contexts are utterance-initial, but the lone citation
forms make up the whole utterance and are thus also utterance-final. Class
1 nouns have a floating H tone (see below) which docks on the final
syllable if the word occurs at the end of an i n t o n a t i o n u n i t .
I stated above that tone on utterance-initial syllables is neutralised to M.
However, certain nouns in table 2 display HL in the fourth and/or fifth
column where ML is expected. It is tempting to dismiss this as noise and
to label all examples as ML, but this would not be true to the data, and it
is too early in the process of analysis to be sure that the contrast is not
significant. The fact that ML and HL both occur in classes 0 and 4
suggests that there is no contrast, but the consistent difference between ki
HL hand and ki ML key in elicited sentence pairs suggests that there is.
However, elicitation of ki hand and ki key outside of paired sentences
does not obviously display this contrast, and it may result from the fact
that hand was elicited first in each pair, and the lower onset of key
reflects declination.
I have recorded citation forms, including probable minimal tonal pairs,
for many more nouns than are shown in table 2, but the neutralisation of
citation forms renders these partially unusable. Classes 0, 3 and 4 all have
ML (or HL) as citation forms. Only classes 1 and 2 are recognisable from
the citation form alone (as M and MH respectively).

300

Malcolm Ross

Table 2: Tone-melody classes of nouns

Class

noun

gloss

citation
melody

utteranceinitial
melody

utterancemedial
melody /H_

utterancemedial
melody /L_

verbinitial
melody

aa
ki
maa
pada

man
hand
taro
room

ML
ML
ML
ML

ML
HL
ML
ML

H
H
H
H

L
L
L
L

assim
assim
assim
assim

aa
adasa
ada
kapaa
mapua
karia
maa
taga
rogoma
rapalo
papa
tapa
yasa

leg
game meat
house
egg
dream
bamboo cane
neck
ashes
clay
arrow
kin term
platform
mushroom

M
M
ML
M
M
M
M
M
?
?
M
M
ML

MH
MH
MH
MHa
MHa
MH
MH
M(H)
MH
MH
MH
MH
ML

HL
HL
HL
H(L)b
HL
HL
HL
H(L)b
H(L)b
HL
HL
HL
HL

L
L
L
L
L
L
L
L
L(H)
LH
L(H)
LH
LH

H?
H
H
H
H
H
H
H
H
H
H?c
assim
assim

e
kaai
kau
puti
popo
yasa

garden
banana
lizard
mushroom type
steam
song

MH
MH
MH
MH
MH
MH

MH
MH
MH
MH
MH
MH

LH
LH
LH
LH
LH
LH

LH
LH
LH
LH
LH
LH

assim (=H)
assim (=H)
assim (=H)
assim (=H)
assim (=H)
assim (=H)

kabu

digging stick

ML

ML

LH

HL

assim

yaa
ki
kau
kodo
lemogo

bird
key
owl
scar
pupil of eye

ML
HL
HL
ML
MLL

ML
ML
HL
ML

HL
HL
HL
HL
HL

HL
HL
HL
HL
HL

assim (=L)
assim (=L)
assim (=L)
assim (=L)
assim (=L)

Utterance-initial there is a ML fall on second syllable, but the following verb is H. The fall seems to
represent an intonation unit boundary (declination), as there is a pause before the verb.
b
L is imperceptible, perhaps because it merges with following H.
c
Verb-initial H is sufficiently high to contrast with the syllables that follow it, but it is lower than final
a

If a word is utterance-medial, its tone melody in classes 0, 1 and 3 varies

according to the final tone of the previous word. The sixth and seventh columns
of table 2 show the tone melodies that occur when the word is preceded by,
respectively, a high and a low tone. Where the second tone of HL or LH is
parenthesised, this indicates that it is difficult to tell whether the tone is present
or not.
In the elicitation frames (examples below) a noun is always followed by a verb.
The eighth column shows the word-initial tone of a verb following each class of
noun.9 Following a class 1 noun, the verb-initial tone is always high. In the other
classes it assimilates to the final tone of the noun.
Table 3 summarises the tone classes displayed by the nouns listed in table 2
and discussed below. Each class, except class 0, is equated with the underlying
9

I dont know whether the word-initial tone of a word other than a verb is affected in the same way.

East Kewa tone

301

tone melody indicated in the first and second columns. Class 0 is underlyingly
toneless. The raised H of class 1 denotes a floating high tone.
Table 3: Summary of tone-melody classes of nouns (see table 2)
Class
0
1 = LH
2 = LH
3=H
4 = HL

toneless
low
rising
high
falling

citation
melody

utterance- utteranceinitial
medial
melody
melody /H_

utterance- verbmedial
initial
melody /L_ melody

ML
M
MH
ML
ML, HL

ML, HL
MH
MH
ML
ML, HL

L
L(H)
LH
HL
HL

H
HL
LH
LH
HL

assim
H
assim (= H)
assim
assim (= L)

Examples of the five classes are:

Class 0: pada (no tone) room

Class 1: yasa (LH)
mushroom
Class 2: yasa (LH)
song
Class 3: kabu (H)
digging stick
Class 4: kodo (HL)
scar

The behaviour of the toneless class 0 pada room is illustrated in (4). Both
syllables of pada assume the final tone of the preceding word (and the first
syllable of the following verb, adawa I saw, also assimilates to the final tone of
pada).
(4) Class 0: pada
a. bl pd da=wa
first
room see-1S. N R P A S T
I saw the room first.
b. abala
pada ada=wa
yesterday room see-1S. N R P A S T
I saw the room yesterday.
In class 1 (= low), the underlying L melody remains after a preceding L
tone, but becomes HL after a preceding H, i.e., the first tone agrees with
the final tone of the preceding word, illustrated by ada house in (5).
However, as indicated in the first column, the underlying melody of this
class is not simply L, but apparently L H , i.e., there is a floating H tone.
This H manifests itself in the utterance-initial melody MH and in the first
tone of the following verb. It also occurs sporadically on the final syllable
of the noun itself.10
10

It is not yet clear whether this sporadic appearance is conditioned in some way.

302

Malcolm Ross

(5) Class 1: ada

a. bl da
da=wa
first house
see-1S.NRPAST
I saw the house first.
b. abala
ada
da=wa
yesterday
house
see-1S.NRPAST
I saw the house yesterday.
Examples (6) and (7), respectively, show how in classes 2 (LH) and 4
(HL) the underlying contour tone remains unaffected by the preceding tone.
The LH tone is apparently resistant to perturbation in East Kewa in all
contexts. The first syllable of the following verb again assimilates to the
final tone of the noun, as in classes 0 and 3.
(6) Class 2: kaai
a. bl kaa
da=wa
first banana see-1S.NRPAST
I saw the banana first.
b. abala
kaa
da=wa
yesterday
banana see-1S.NRPAST
I saw the banana yesterday.
(7) Class 4: kodo
a. bl kdo ada=wa
first scar see-1S.NRPAST
I saw the scar first.
b. abala
kdo ada=wa
yesterday
scar see-1S.NRPAST
I saw the scar yesterday.
Class 3 (apparently high) has at present only one member, kabu digging
stick (8), and is therefore suspect. It is also suspect in terms of its tonal
behaviour, as its word-initial tone is the opposite of the preceding trigger,
LH after a preceding H tone, HL after a preceding L. This would, if correct,
confound any claim that synchronic tone perturbation is simple
assimilation. However, the following verb has the tone predicted by the
tone of bl/abala, as in class 0.
(8) Class 3: kabu
a. bl kab
da=wa
first digging.stick see-1S.NRPAST
I saw the mushroom first.

East Kewa tone

303

b. abala
kbu
ada=wa
yesterday
digging.stick see-1S.NRPAST
I saw the digging stick yesterday.
I have observed above that, just as the tone of a noun may be affected by
the last tone of the previous word, so the first tone of the following verb
usually matches (is usually assimilated to) the last tone of the preceding
word. This is indicated by assim in tables 2 and 3. The one class where
assimilation does not occur is class 1. Here, the following verb always has
initial H, attributed to the floating H tone of nouns in this class (9).
(9) Class 1: rapalo
a. bl rpalo
da=wa
first a r r o w
see-1S.NRPAST
I saw the arrow first.
b. abala
rapalo da=wa
yesterday
arrow
see-1S.NRPAST
I saw the arrow yesterday.
There are two unexplained cases listed under class 1 in table 2, tapa
platform and yasa mushroom. They have class 1 behaviour but cause
assimilation of the following tone (i.e., they seem to reflect an underlying
L melody but with no floating H). They do not appear to be elicitation
errors, and may prove to reflect yet another class of noun.
4.2 Personal pronouns
As I noted in the discussion of the tonal behaviour of nouns, the tone of (at
least) the first syllable of a verb is determined by the preceding word. In
many of the utterances I have elicited (and not infrequently in text), the
preceding word is a personal pronoun. These are listed in table 4.
Table 4: Personal pronouns
person
1
2
3

SINGULAR DUAL

n
ne
ip

s
pi
p

PLURAL

n
mi
m

There are two tonal minimal pairs among the pronouns: n 1s vs. ne 2s, and
ip 3S vs. p 3D.
A personal pronoun subject is followed by the ergative enclitic =me if the
verb is transitive. The enclitic becomes =mi if the final vowel of the
pronoun is /i/ or /u/, i.e., after ip, pi, p, mi or m. The enclitic is

304

Malcolm Ross

lexically toneless and acquires the tone of the last syllable of the pronoun.
Because the first, and sometimes also the second, syllable of an utteranceinitial pronoun is neutralised to mid tone, it is sometimes only the tone of
the ergative clitic which disambiguates the members of a tonal minimal
pair. That is, n=m (l S=ERG) is manifested a s nem , and ne=me
(2S=ERG) as neme.
There is a further complication with regard to pi (2D) and mi (2P) to
which I return in 4.3.
4.3 Tone-melody classes of verbs
Unlike nouns, verb words in East Kewa usually consist of a root and a
desinence. This raises a question. Does a verb word have a single tone
melody? Or do verb root and desinence have separate melodies which
combine to form a compound melody across the verb word? The
compound-melody alternative provides a simpler analysis, and I assume it
in the description below, returning to a discussion of the two alternatives in
4.4, where melody classes of verbs are compared with those of nouns.
The analysis in this section is based on elicited paradigms with the near past
and present progressive desinences shown in table 5. Two sets of desinences are
shown for each tense. The second set in each case is used with verbs with a
mono- or disyllabic root ending in /aa/. Note that trisyllabic roots in /aa/ like
pogolaa jump take the default desinences. Syllables marked with a grave accent
are always L and never perturbed to H.
Table 5: Verbal desinences
NEAR
person and
PAST
number
default after -aa

PRESENT
PROGRESSIVE
default after -aa

1S
2S
3S
1D
2/3D
1P
2/3P

-lo
-le
-la
-lep
-lep
-lem
-lem

-wa
-e
-a
-apa
-ape
-ama
-ame

-ripu
-ripi
-ripa
-pa
-pe
-rima
-rimi

-to
-te
-ta
-tep
-tep
-tem
-tem

Verb roots are set out in tone melody classes in table 6 in a format
similar to that used for nouns in table 2. The three classes correspond in
their tonal behaviour to noun melody classes 0, 1 and 2, except for the fact
that verb root class 1 is simply L: it shows no sign of a floating H tone.
Classes 0b, 1b and 2b are verb roots ending in /-aa/ which take the special
desinence sets shown in table 5. The first six columns of table 6 correspond

East Kewa tone

305

with columns in table 2, with the important difference that the hyphen in
the middle of the tone melodies in the fifth and sixth columns marks the
boundary between root and desinence. Parentheses in the tone melodies of
the desinences take account of the fact, seen in table 5, that some
desinences have one syllable, others two. The melodies of disyllabic
desinences include the parenthesised tones; the melodies of monosyllabic
desinences exclude them.
Table 6: Tone-melody classes of verb roots and desinences
Class

verb

gloss

citation
melody

utterancemedial
melody/H_

utterancemedial
melody /L_

FREE:
2D/P

0a

pona
na
epa
la
tya
waraa

cut
eat
come
speak
hit
touch

HL
M
HL
HL
HL
ML

HH-HH
H-H(L)
HH-H
H-H
H-H
HH-HL

LL-L(L)
L-L(L)
LL-L
L-L
L-L
LL-LL

L
L
?
?
?
L

ad a
ruma
kala
pea
ria
koda
aa
pogolaa
rumaa

see
climb up
give
do
carry
enter
stand
jump
share out

M
M
M
M
M
HL
M
M
ML

HL- L(L)
HL-L(L)
HL-L
HL-L
HL-L
HL-L
HL-LL
HLL-LL
HL-L(L)

LL-L(L)
LL-L(L)
LL-L
LL-L
LL-L
LL-L
L-LL
LLL-LL
LL-L(L)

L
L
?
?
?
?
?
H
H

ruba
rera
rogaa

throw out ML
cry
LH
tie
M

LH-(H)L
LL-H
LH-LL

LH-(H)L
LH-H
LH-HL

L
?
?

0b
1a

1b

2a
2b

The column headed FREE:2D/P in table 6 makes reference to a feature

for which I have no explanation. In table 4 above, the second person dual
and plural free pronouns are shown as pi and mi respectively. Because
their second tone is L, this provides the contextual tone for the following
word, which in (10a) is the class 1a verb ruma climb. Preceding certain
verbs of class 1b, however, the tone melody of these pronouns shifts from
HL to H, and they become p and m, inducing the tone pattern expected
after a contextual H. This is illustrated in (10b), where the class 1b verb is
rmaa share out.
(10) a. pi=mi
ruma=ape
FREE:2D=ERG climb-2D.NRPAST
You two climbed (it).

306

Malcolm Ross

b. p=m
rmaa=pe
FREE:2D=ERG shared-2D.NRPAST
You two shared (it).
Table 7 is a summary of the tone classes displayed by the verbs listed in
table 6. 11 There are several features in table 7 of which I am not yet able
to give a coherent account. This is partly because I have not yet collected
enough paradigms, partly because some have gaps because crucial
disyllabic desinences have not been collected. The gaps are indicated by
? in table 6. This is important in the cases of class 0, which is apparently
divided into two subclasses by the tonal behaviour of its desinences, and
of class 2, where no pattern can yet be recognised in the tonal behaviour
of desinences.
Table 7: Summary of tone-melody classes of verb roots and desinences
(cf. table 6)
Class

utterancecitation medial
melody melody /H_

utterancemedial
FREE:
melody /L_ 2D/P

LL-L(L)
LL-L(L)

0a
0b

toneless

HL

(H)H-HH, (H)H-H(L) (L)L-L(L)

HH-HL
LL-LL

L, ?
L

1a = L
1b

low

HL-L(L)
HL-L(L)

L, ?
H, ?

2a = LH rising
2b

LH

LH-(H)L
LH-LL

LH-(H)L
LH-HL

L, ?
?

Table 7 also shows the default behaviour of desinences. Their first or only
syllable is assimilated to the tone of the final root syllable, and their second
syllable, if any, is low. This default behaviour resembles that of a class 1 (low)
noun. However, there are two kinds of deviation from default behaviour.
Firstly, as noted in table 6, among the paradigms collected thus far just one
class 0 verb, pona cut, assimilates both syllables of its desinences, except those
with fixed L, to H if the root-final syllable is H. Contrast (11a), with an -HH
desinence, with (11b), where the final desinence is one the tone of which always
remains low (table 5).
(11) a. m=m
pn-m
FREE:3P=ERG
cut-3P.NRPAST
They cut (it).

11

Class O is the Franklins underlying H, Class 1 is their underlying L, Class 2 their underlying LH.

East Kewa tone

307

b. m=m
pn-lme
FREE:3P=ERG cut-3P.PRES
They are cutting (it).
Note, however, that in other class 0 verbs for which relevant data have been
collected a disyllabic desinence ends in a L tone as expected (see 13a).12
The second deviation from default (low) behaviour is that certain class 2
verbs do not assimilate the first or only syllable of a singular desinence to H when
we might expect them to. Usually, the first syllable of a desinence has the same
tone as the final syllable of the root, as in (12a), where n eat is a class 0 verb,
but the class 2 verbs ruba throw out and rogaa tie take a desinence with L tone
despite root-final H, as in (12b). The patterning here is not yet clear.
(12) a. n=m
FREE:IS=ERG
I a t e ( i t ) .

n=w
eat-1S.NRPAST

b . n=m
rub=wa
FREE:IS=ERG
throw.out-1S.NRPAST
I thre w (it) out.
4.4 Comparing the tone-melody classes of nouns and verbs
I have assumed in 4.3 that verb root and desinence have separate melodies
which combine to form a compound melody across the verb word, avoiding the
alternative possibility that a verb word has a single tone melody. The description
shows that with certain exceptions desinences have their own melody
corresponding roughly to that of class 1 (L) nouns. However, the non-singular
present progressive desinences in table 5 display a final L which defies
perturbation. This pattern has no parallel in the tonal behaviours of nouns, but it
appears to represent a desinence-specific tone melody.
If the available tone melodies of East Kewa are treated as a system, then there
are also systemic grounds for analysing verb roots as having separate tone
melodies. The tone-melody classes of nouns in table 3 and of verb roots in table
7 are numbered in the same way in order to facilitate comparison, and table 8
displays the two sets side by side. There are no verb classes 3 and 4, i.e., no
classes corresponding to the high and falling classes of nouns. The missing
classes correspond to the two noun classes which have a HL melody when the
final syllable of the previous word is L. As a consequence all verbs have initial L
when the preceding word ends in L.13

12
The Franklins (1978: 3537) treat the latter as a separate verb class (underlying HL), but its tonal assimilation is not
parallel to that of nouns in HL.
13
This apparent fact may be somehow related to the tonal behaviour of FREE:2D/P pronouns when they co-occur with
(some?) Class 1b verbs (4.3), but how they may be related is not clear.

308

Malcolm Ross

Table 8: Comparison of tone-melody classes of nouns and verb roots

Class

Utterance-medial melodies:
underlying Noun classes Verb classes
tone?
/H_ /L_
/H_

/L_

0
1
2
3
4

toneless
L
LH
H
HL

(L)L-L(L)
LL-L(L)
LH-(H)L

H
HL
LH
H
HL

L
L
LH
HL
HL

(H)H-HH, (H)H-H(L)
HL-L(L)
LH-(H)L

Although the parallel between noun and verb root classes is thus incomplete,
it is more convincing that the parallel between noun melodies and whole verb
words. This is because verb words of class 2 have an LHL tone melody, and
three-tone melodies have not been found among nouns. 14 Further, verb words
with an HL melody display different mappings of tone to syllable. In class 0 we
find HHL and HHHL, as in wr-rpa in (13a), but in class 1 H L L a nd H L L L ,
a s i n rma-leme in (13b).
(13) a. ip=m
wr-rpa
FREE:3S=ERG touch-3S.NRPAST
S/he touched (it).
b. m=m
rma-leme
FREE:3P=ERG
climb-3P.PRES
They are climbing (it).
If these are both manifestations of a single HL melody, then a diacritic feature
must be introduced into the analysis to account for the different mappings of the
melody onto sequences of four syllables. If roots and desinences are understood
each to have their own melody, this complication does not arise.

5 Conclusions
Perhaps the most obvious conclusion to be drawn about East Kewa tone is that
there is much more to be learned. Phonologists will probably find my account
unnecessarily conservative. This is partly a reflection of the paucity of my data
and of a desire not to extrapolate to analytic decisions that might not apply to a
larger data set, and partly a reflection of the fact that I am not a phonologist.
There are also HLH examples in the Franklins materials.

14

East Kewa tone

309

Certain things seem clear, however. The domain of tone in East Kewa is not
the word, but the morpheme. This is also true of other Trans New Guinea
languages of the Highlands of Papua New Guinea, although there are substantial
differences of detail between the tone system of, say, Fore and the tone system
of East Kewa (Ross 2005). East Kewa (and other Highlands languages too)
displays a quantitative difference between the paradigm of tones available to
nouns and the paradigm available to verb roots. The paradigm of verb-root tones
is smaller than the paradigm of noun tones, and this seems to reflect a tendency
across those of the worlds languages in which the tonal domain is the
morpheme and in which the verb displays greater affixation than the noun
(Larry Hyman, pers. comm., 6 November 2004). This tendency is also
manifested in Niger-Congo languages and in Tokyo Japanese,15 and reflects the
larger tendency for verbs to have larger paradigms than nouns (Rhodes 1987).

References
Donohue, Mark, 1997. Tone systems in New Guinea. Linguistic Typology 1:347
386.
Franklin, Karl J., 1971. A grammar of Kewa, New Guinea. Pacific Linguistics C-16.
Canberra: The Australian National University.
Franklin, Karl J., and Joice Franklin, 1962. Kewa (i): Phonological asymmetry.
Anthropological Linguistics 4(7):2937.
Franklin, Karl J., and Joice Franklin, assisted by Yapua Kirapesi. 1978. A Kewa
dictionary, with supplementary grammatical and anthropological materials.
Pacific Linguistics C-53. Canberra: The Australian National University
Hyman, Larry M., 2006. Word-prosodic typology. Phonology 23:225257.
Rhodes, Richard, 1987. Paradigms large and small. Berkeley Linguistics Society
13:223234.
Ross, Malcolm, 2005. Towards a reconstruction of the history of tone in the
Trans New Guinea family. In Shigeki Kaji, ed., Proceedings of the symposium
crosslinguistic studies of tonal phenomena: Historical development, tone-syntax
interface, and descriptive studies, 331. Tokyo: Research Institute for
Languages, Cultures of Asia, and Africa, Tokyo University of Foreign Studies.

15
Tokyo Japanese is usually classified as a pitch-accent language (rather than a tone language), but, as Hyman (2006)
argues, pitch accent languages do not form a coherent category. Tokyo Japanese can readily be analysed as a tone
language in which the tonal domain is the morpheme, albeit with a tonal organisation quite different from that of Kewa.