Lund University, Dept.

of Linguistics 241
Working Papers 42 (1994), 241-250

A Spectrographic Analysis of
Vowel Allophones in Kabardian
Sidney A . J. Wood
This article, a revised version of a paper given at the Maikop symposium of the Societas
Europeae Caucasiologica in 1992, reports some preliminary results from an invesdgation of
the spectra of vowel allophones in the NW Caucasian language Kabardian. It is concluded
that the environment for palatal vowel allophones consists only of the palatal semivowel and
not of other consonants with a palatal component, and that the unrounded palatovelar
environments need further investigation.

Introduction
The phonerruc analysis o f the Kabardian vowels is controversial and this
study is being undertaken i n order to clarify some of the problems
i n v o l v e d . Recent contributions on this topic have been p u b l i s h e d by
Anderson 1991, C h o i 1991, Colorusso 1988 (chapts. 3, 8), Smeets 1984 and
W o o d 1991a. The issue is central to phoneme theory since it concerns the
v a l i d i t y o f T r u b e t z k o y ' s 1939 doctrine that a l l phoneme systems must
consist o f a consonant system and a vowel system, and that any system must
contain at least three contrasting elements. The alternative is to disregard
that doctrine and push the analysis as far as the data w i l l allow, arriving at a
result that was anathema to Trubetzkoy and his followers, i.e. one or two
v o w e l phonemes contrasting syntagmatically with a l l consonants, or no
v o w e l phoneme at a l l . vowels i n speech being epenthetic. T w o typical
exponents of these approaches are illustrated in Figs. 1 and 2. Catford 1942
conforms to the standard vertical three-phoneme solution from Trubetzkoy
(i.e. a high phoneme versus a mid phoneme versus a low phoneme).
Kuipers 1960 disregards the Trubetzkoy doctrine and, f o l l o w i n g Jakovlev
1923, proposes a two-phoneme system w i t h mid and low vowels i n
complementary distribution. Jakovlev had identified the traditional 'short a'
as mid, and ' l o n g a ' as low, the timbre contrast being predictable from
quantity, whereas Kuipers was now proposing an underlying syntagmatic
contrast o f /ha/ vs. /a/ that surfaces after metathesis as [a] vs. [a].
242 SIDNEY A. J. WOOD VOWEL ALLOPHONES IN K A B A R D I A N 243

T a b l e 1. Kabardian consonants according to Catford 1942 and Kuipers

VOWEL ASSIMILATION ENVIRONMENT 1960, B a g o v et al. 1970 and Smeets 1984, retranscribed according to I P A
conventions with the guidance of notes i n the sources (from W o o d 1991a).
plain A s s i m i l a t i o n environments according to K u i p e r s are (2-4,5,6,7,8), and
palatal rounded
uvular elsewhere according to Catford / j / at ( 3 ) , and (5,6). Kuipers classifies j, w as hJ, h'^
j kw qw etc
q etc
respectively. The character / at (4) denotes a voiced palatal fricative.
high i u/u a i
glottal voiceless voiced sonorant
mid e ob A B P' P b m
f f V
low D a a
t' t d n r Catford interprets ts
ts' ts and (fe as f^-s and d+z
F i g u r e 1. A n example of the three-phoneme solution, based on the narrow
transcription o f the North wind and the sun i n C a t f o r d 1942. T h i s s z
particular solution does not admit complementary distribution between mid
allophones and low allophones. Z
(1) e'
J 3

VOWEL ASSIMILATION ENVIRONMENT (2) ^ ' a

( 3 ) c' c J j
palatal rounded P^^j" pharyngeal , , 5 j
cfjetc kwqwetc "^"'^^ h elsewhere (4)
^ qetc
xw
(5) ,
high i ti/u m 3 kw kw gW w

mid e o/o Catford has q, qh

(6) q ' q
low Q Q a qw and q"^- q'^'^ instead
% K
KW
F i g u r e 2. A n example of a two-phoneme solution (Kuipers 1960), showing
complementary distribution between mid and low allophones of a non-high (7) h
phoneme.
(8)7 h
7w
The variety o f v o w e l allophones arises by perseveratory lingual and
r o u n d i n g a s s i m i l a t i o n to p r e v o c a l i c palatal, palatovelar and u v u l a r
consonants, and by anticipatory lingual and rounding a s s i m i l a t i o n to
Background
postvocahc rounded palatovelars and uvulars. The consonants o f Kabardian
Phoneme analysis
are listed i n Table 1.
Trubetzkoy's 1925 criticisms of Jakovlev's solution were partly diachronic
A n analysis of vowel spectra cannot resolve the controversy concerning
(Kabardian ' l o n g a ' and 'short a' have different origins and thus represent
the status of m i n i m a l vowel systems. That is a theoretical issue and is a
different phonemes) and partiy synchronic (die distinction between ' l o n g a'
matter o f linguistic creed, depending on whether or not one accepts the
and 'short a' is no longer one of quantity and there is now a phonemic
Trubetzkoy doctrine of systems. A believer w i l l halt the analysis once a
opposition of 'sonorance' between them).
three phoneme system is reached. What spectral analysis can do is offer
physical correlates for identified allophones, especially evidence for or
against complementary distribution.
 244 SIDNEY A. J. WOOD VOWEL ALLOPHONES IN K A B A R D I A N 245

Spectral analysis of Kabardian vowels

(F2-F1)
C h o i 1991 measured the spectra of Kabardian vowels by three speakers of
2500 20O0 1SO0 1000 500 200 Hz
T—r-r —7- T r the Terek dialect who read a piece of continuous prose twice (taken from
ii^ri IT a i i 1 1 1 1
S a g i r o v 1967:182). A l l tokens were p o o l e d and sorted according to
(FU prevocalic environment, using Catford's analysis of the consonants and
- 30O vowels. H i s results (Figure 3) show a full complement of allophones at each
1?*
of the tiiree heights, similar to Catford's solution (Figure 1). Unfortunately,
the v o w e l spectra i n C h o i ' s material are subject to other influences i n
- 400
addition to the lingual and labial articulations being studied, and these

9 potential sources of bias are worth looking at more closely.

SCO (1) The numerical values of formant frequencies for a given timbre vary
between speakers due to individual differences o f vocal tract morphology
600 and can therefore only function contrastively within the speech o f the
person who uttered them. Pooling data from several speakers unnecessarily
700 adds between-speaker spectral variation to the witiiin-speaker variation
being observed. This can be avoided by reporting each subject individually.
800 (2) Spectra are also expected to vary with degree o f accentuation and
focus. Languages may certainly differ i n the extent of reduction related to
F i g u r e 3. C h o i ' s 1991 spectral analysis of K a b a r d i a n v o w e l allophones. accentuation, but the possibility of formant frequency averages being biased
T h e characters represent the average spectrum o f pooled tokens i n the
by pooling accented and unaccented tokens should not be overlooked.
respective environments: ' c ' postalveolars, ' k ' unrounded velars, ' k W
r o u n d e d velars, ' q ' u n r o u n d e d uvulars, ' q w ' rounded uvulars, ' h ' (3) Nasal contexts were included i n C h o i ' s material. T h e problem here is
pharyngeals, ' 4 ' tiie defauh allophones. that the introduction of antiresonance and nasal formants from nasal
coupling during vowels adjacent to nasals can mask the oral resonances
Catford 1977 p r o v i d e d experimental confirmation o f T r u b e t z k o y ' s related to the lingual and labial environments that are the object o f the
observation on the quantity contrast, by comparing the durations o f these study. The formant frequencies measured under nasal conditions w i l l not
two vowels and demonstrating there was no greater a difference between reflect the articulations being studied under non-nasal conditions. It w o u l d
them than w o u l d be expected from the regular correlation between duration be appropriate to reject the vowel tokens from the nasal environments, or at
and degree o f opening. C h o i 1991 repeated the same experiment w i t h least deal with them separately.
different data and came to the same conclusion. However, diat experimental (4) It is w e l l k n o w n that different consonant environments bias v o w e l
approach is not effective against Kuipers's solution since he is not offering a formant frequencies, but these contextual effects do not always give rise to
surface contrast of quantity between 'long a' and 'short a ' but an underlying the experience of different vowel allophones. In the case o f Kabardian the
syntagmatic contrast o f /ha/ vs. /a/. K u i p e r s ' s solution can only be refuted vowels are said to be assimilated into identifiable allophones that can be
by either denying die generative model (which is a matter o f creed again) allotted appropriate characters from the phonetic alphabet. T h e formant
or b y attacking the justification for the postulated /ha/ and metathesis frequency average captures the contextual bias from the consonant, but it is
(which is a matter of whether the language data allows tiiat interpretation). not an indicator of the independence of an allophone from its neighbouring
K u m a x o v 1984 queries the vahdity o f the ha>ah>a interpretation, and C h o i vowel allophones. A scatter plot of the original tokens is a better indicator
quotes Catford's arguments on the same theme. T h e issue remains as of how far an allophone is distinct from or confused witii its neighbours.
controversial as ever. These effects are taken into account i n the study reported here.
 246 SIDNEY A. J. WOOD VOWEL ALLOPHONES IN KABARDIAN 247

reached a m a x i m u m , as experience from w o r k i n g with X - r a y films and

F2 2000 1500 1000 Hz synchronised sound showed that this was the point on w h i c h the articulator
gestures for a v o w e l converge, before the mandible begins to close again
for die next consonant ("Wood 1979, 1989, 1991b).
Linear prediction w o u l d have been preferred but die L u n d I L S system
was inaccessible f o l l o w i n g the sudden and untimely demise o f the m a i n -
frame lab computer, and replacement desktop systems based on Macintosh,
P C and Sun had not yet been installed. That is also the reason w h y so littie
of the material was analysed at that time, just sufficient to satisfy one's
curiosity and whet die appetite for more. The L u n d lab is n o w back i n
business again and processing w i l l be resumed.

Results and discussion

T h e formant frequencies obtained are summarised i n F i g u r e 4 for
c o m p a r i s o n w i t h C h o i ' s resuhs i n F i g u r e 3. Note that C h o i follows
Ladefoged's practice of plotting F 2 - F 1 ratiier than just F 2 . Figure 4 shows
tiiat die tokens, for each environment form distinct allophone islands, except
that the palatal allophones ( ' p ' ) o f the close phoneme are h a r d l y
F i g u r e 4. Frequencies of F l and F 2 o f accented Kabardian vowels i n the
distinguishable from the default allophones ( ' i ' ) , by F l and F 2 at least, and
present study. The characters represent different consonant environments:
' p ' palatal (2,3,4 i n Table 1), ' i ' default high allophone, ' u ' rounded (5,6), hardly seem to waiTant the status o f an independent allophone. A t the same
'Q' default mid allophone (Jakovlev's 'short a ' ) , ' o ' rounded (5,6), ' a ' time, these tokens tend to accumulate i n the upper half o f the F 2 range of
default open allophone (Jakovlev's 'long a'). The lines A A and B B divide the default allophone, and w o u l d thus have a higher average F 2 .
high from mid from low. The frequencies are compressed to the M e l scale. The Kabardian consonants include rounded velars and uvulars (see [5,6]
i n Table 1) that constitute environments for rounding vowels and velarising
Procedure or uvularising them, yielding [u] or [o] as the case may be. There are clear
Recordings were made from nadve speakers o f K a b a r d i a n i n M o s c o w , examples of these i n Figs. 3 and 4. C h o i also has examples of a low rounded
M a i k o p and T b i l i s i (residents or visitors). The subject reported here is a allophone [D] (there were no candidates for this i n m y limited data).
male speaker o f K u b a n Kabardian, a postgraduate student at the M o s c o w The high, mid and low vowels are distinguished by l o w , middle and high
Academy of Sciences. F l frequencies, w h i c h is expected from acoustic theory (both widening the
The material consisted of the hst of words given by K u m a x o v 1981:44 as anterior end of the vocal ti-act and narrowing the posterior end raises F l ) .
exemples of the v o w e l phonemes i n a variety of consonant environments The data available here only permit a more penetrating discussion o f the
(but not all). The words were ordered randomly and read both in isolation palatal allophones and what constitutes the palatahsing environment in
and framed i n focal position i n a carrier sentence. One reading i n isolation Kabardian.
by one subject is reported here.
Accented vowels were analysed, excluding those i n nasal contexts. The Palatal allophones
formants were measured by locating a cursor i n a broadband spectrogram, The available descriptions have us expect an [i]-like allophone for the high
based on F F T slices, displayed on the screen o f a P C computer (Ternsttom phoneme i n the palatal envirnoment, i.e. an F2 o f at least 2000 H z , but diis
1987, 16-bit A r i e l D S P , at Umea) and reading die coordinates o f the cursor has not materiahsed. F 2 o f these tokens i n Figure 4 ranges 1400-1750 H z
position. The formants were measured at die point i n die v o w e l where F l
 248 SIDNEY A. J. WOOD VOWEL ALLOPHONES IN KABARDIAN 249

and the coiTCsponding set i n Figure 3 ('e', C h o i ' s postalveolar environment) T h e results i n F i g s . 3 and 4 agree that many of the tokens i n the
has an average F 2 - F 1 of 1400 H z (i.e. mean F 2 around 1750 H z ) . T h i s respective palatalising environments for the high v o w e l have F 2 m u c h
result can be partly explained by our having included different consonants lower than expected for an [i]-hke timbre, so low i n fact that they coincide
i n this environment. with the tokens i n the default environment. This indicates that we should
M y palatal set ' p ' i n Figure 4 theoretically comprises consonants from reconsider what constitutes die palatahsing environment i n Kabardian, and
[2,3,4] i n Table 1, i.e. laterals, unrounded palatovelar stops and fricatives that laterals, alveopalatals, palato-alveolars and unrounded palatovelars are
(said to be produced as palatal affricates and fricatives) and the palatal not part of it, despite the palatal component i n their articulations. That
semivowel, but there were i n fact no tokens with the s e m i v o w e l i n the leaves just the palatal semivowel, as i n Catford 1942.
present set o f data. Catford 1942, on the contrary, reports only the semi- The default environment consists of the labials and alveolars that remain
v o w e l yielding allophones [i,e], the rest producing tire default allophones. after removing die palatal, velar and uvular assimilation environments. The
F i g u r e 4, w h i c h shows F 2 ranging 1400-1700 H z i n m y lateral and common feature of tiie laterals, alveopalatals and palato-alveolars, diat must
unrounded palatovelar environment ' p ' , suggests he is correct i n counting now also be included i n the default environment, is that they are coronal,
only the semivowel i n the palatalising environment. like the alveolars already diere. The palatalising environment is simply die
C h o i ' s palatal set (that he calls postalveolar) does not include the laterals non-coronal palatal, i.e. just / j / .
or palatahsed velars, but instead it includes alveopalatals /<?', c. ?/ and What about the plain palatovelars, that are said to be so palatalised? They
palato-alveolars / / , 3/ from (1) in Table 1. H i s data does include examples certainly did not palatalise vowels i n m y data, and C h o i never had them as
with the semivowel / j / , which presumably explains why the average F 2 for palatalising environments anyway. D o they belong to the default environ-
this group is as high as 1750 H z . 1 would now expect that his tokens i n the ment tiiat yields [i] or do they belong to a velarising environment diat yields
semivowel environment were indeed [i]-like witii F 2 around 2000 H z , while [ui], along with the unrounded uvular environment? T h i s requires further
the alveoplatals and palato-alveolars yield the default allophone rather than clarification.
the palatal allophone (which consequently pulled his average F 2 - F 1 down
for the palatal context). References
The present material has no example of / j / witii a mid vowel, that yields Anderson, J . M . 1991. 'Kabardian disenivowelled. again". Studia Linguistica
[e], but C h o i ' s does. Once again, his average F 2 - F 1 is very l o w , 1100 H z . 45, 18-48.
(i.e. mean F 2 around 1550 H z ) . Something hke 1800 H z w o u l d be expected B a g o v . P. M . , B . X . B a l k a r o v , T. X . Kuaseva, M . A . K u m a x o v & G . B .
for an [e]-like timbre, and the discrepancy presumably once again reflects R o g o v a (eds.). 1970. Gratnmatika kabardino-cerkesskogo literaturnogo
the fact that alveopalatals etc. are not part o f the palatalising environment. jazyka; 1 fonetika i morfologija [Grammar of literary K a b a r d i n o -
C h o i also has a palatal set among the low vowels. N o n e o f the sources, Circassian. part 1, phonetics and morphology]. M o s c o w : N a u k a .
including Catford, reports a palatal allophone of a low vowel (compare e.g. Catford. J . C . 1942. 'The Kabardian language'. Maitre Phonetique 3rd
Figs. 1 and 2), and the absence of a l o w palatal allophone was part o f Series 78, 15-18.
Jakovlev's argument for complementary distribution between mid and low
Catford, J. C . 1977. ' M o u n t a i n o f tongues: the languages of the Caucasus'.
vowels - his default 'long a' is [a], default 'short a' is [a] and i n the palatal
Annual Review of Anthropology 6, 283-314.
environment they are [e(:)]. A proponent o f the three-phoneme solution
C h o i , J . - D . 1991. ' A n acoustic study of Kabardian v o w e l s ' . Journal of the
might instead argue that there is an accidental gap i n the lexikon so tiiat / j /
International Phonetic Association 21, 4-12.
never occurs with /a/. So what is C h o i ' s palatal environment with /a/?
Colarusso, J. 1988. The Northwest Caucasian languages, a phonological
Presumably tiie same alveopalatals etc. again, which now seem not to belong survey. N e w Y o r k : Garland.
there. Jakovlev, N . 1923. Tablitsy fonetiki kabardinskogo jazyka. Moscow:
Oriental Institute.
250 SIDNEY A . J . WOOD 251

Kuipers, A . H . 1960. Phoneme and morpheme in Kabardian. T h e Hague:

Mouton.
K u m a x o v , M . A . 1981. Sravnitelno-istoriceskaja fonetika adygskix
(cerkesskix) jazykov. [Comparative and historical phonetics of the A d y g e
(Circassian) languages]. M o s c o w : Nauka. The Department of Linguistics and
K u m a x o v , M . A . 1984. Ocerki obscego i kavkazskogo
[Studies i n general and Caucasian linguistics]. N a l ' c i k :
jazykoznanija
Izdatel'stvo
Phonetics 1993-94
El'brus.
Sagirov, A . K . 1967. 'Kabardinskij j a z y k ' . In E . A . B o k a r e v & K . V . Staff
Lomtatidze (eds.), Jazyki narodov SSSR, V o l 4, Iberijsko-kavkazskie General linguistics
jazyki, 165-183. M o s c o w : Nauka. Mats Eeg-Olofsson Lecturer
Smeets. H . J. 1984. Studies in West Circassian phonology. Leiden: H a k u c h i Barbara Gawroriska Research assistant
Marianne Gullberg Research student
Press.
Gisela Hakansson Lectiirer
Ternstrom, S. 1987. Swell soundfile editor. Stockholm: Soundswell M u s i c Research assistant
Kristina Hansson
Acoustics. Arthur H o l m e r Research student
Trubetzkoy, N . 1925. R e v i e w o f Jakovlev 1923. Bulletin de la Societe de Merle H o m e Lecturer
Linguistique de Paris 26, 277-286. Robert Jarvella Lectiirer
Christer Johansson Research student
Trubetzkoy, N . 1939. Grundziige der Phonologic. Prague: Travaux du
Ann Lindvall Research student
Cercle Linguistique de Prague V I I . Lectiirer
U l r i k a Nettelbladt
W o o d , S. A . J. 1979. ' A radiographic analysis of constriction locations for Anders Nordner Research assistant
vowels'. Journal of Phonetics 7, 25-43. Thore Pettersson Lecturer
W o o d , S. A . J. 1989. 'The precision of formant frequency measurement Bengt Sigurd Professor emeritus
Jan-Olof Svantesson Lectiirer
from spectrograms and by linear prediction". Speech Transmission
Revaz Tchantouria Research assistant
Laboratory Quarterly Progress Report STL/QPSR 1/1989, 91-93.
Karina Vamhng Research feUow
Stockholm: R o y a l Institute of Technology.
W o o d , S. A . J. 1991a. 'Vertical, monovocalic and other impossible v o w e l Phonetics
systems: a review of the articulation of the Kabardian v o w e l s ' . Studia Gosta Bruce Professor
Eva Girding Professor emeritus
Linguistica 45, 49-70.
Kerslin Hadding Professor emeritus
W o o d , S. A . J. 1991b. ' X - r a y data on the temporal coordination o f speech
D a v i d House Research fellow
gestures'. Journal of Phonetics 19, 281-292. Kurt Johansson Lecturer
E v a Magnusson Lectiirer
B c r t i l Malmberg Professor emeritus
Duncan M a r k h a m Research student
Y a s u k o Nagano-Madsen Research assistant
Kerstin Naucler Lecturer
Paul Touati Research fehow
M e c h t i l d Tronnier Research smdent
Sidney W o o d Lectiirer