Post on 10-Sep-2020
•
41
VOWEL CONTRAST REDUCTION IN TERMS OF ACOUSTIC SYSTEM CONTRAST
IN VARIOUS LANGUAGES *
Tjeerd de Graaf**
and Florien J. Koopmans-van Beinum
1 . INTRODUCTION
The aim of this paper is to demonstrate that the measure for Acous-·
tic System Contrast ASC (Koopmans-van Beinum, 1980) can be used not
only for description and comparison of vowel contrast reduction in
various speech conditions of one language, but also to compare the
degree of vowel contrast reduction in various languages, whether these languages involve many vowels in their vowel systeir. or only a few. We might hypothesize that systems involving fewer vowels
wo�ld display more contrast reduction than richer vowel systems
since the acoustic vowel space is less filled. Next we want to show that the ASC measure can also be used to com-
pare the degree of contrast between the set of long vowels and the set of short vowels in those systems that contain quantitatively
paired voweln.
Finally some problems will be discussed with regard to the descrip
tion of vowel contrast reduction in languages involving a substan
tial number of nasal vowel s in their vowel system.
2. COMPARISON OF VOWEL SYSTEMS
Several studies on systematics in the distribution of vowels within
* This is an extended version of a paper read at the lOth International Congress of Phonetic Sciences, Utrecht, August 1983.
** Iy·�titute of Phonetic Sciences, Groningen University •
I
43
monly reported difference of quality between long and short vowels
of corresponding position is centralization of the short vowels
(Lehiste, 1970, pp. 30-33). If so, then the resulting ASC values
for the short vowel subsystems must be smaller than for the corre
sponding long vowel subsystems.
For vowel systems containing nasal vowels the following universal
property holds: the number of nasal vowels in such a .v:owel syst;em
is equal to or sma l ler .than the number of oral.vowels (Ruhlen,
1975). In his sample Crothers (1978) has 50 languages (24%) with
nasal vowels. In our study we take Polish (6 oral vowels and 2
nasal vowels) and French (12 oral vowels and 4 nasal vowels) as representative examples. Only the oral· vowels are included in this
study. Fig. I displays the various vowel systems as used in this study. The vowels are represented in stylized diagrams according to the
articulatory dimensions high-low and back-front, in the sar:e way as
used for the first time by Hellwag (l 78i). For Polish and French
the nasal vowels are indicated separately, whereas for Hungarian
and Frisian we give the short and corresponding long vowels also
in a parallel way.
3. SPEECH NATERIAL AND MEASUREMENTS
The speech material in this study consisted of vowels spoken in
isolation, vowels from isolated words, and vowels from a context.
In most cases free conversation is used for this context whereas
some rare vowels in a many-vowel system are se.lected from sen-� '. .
tences read- from a�specific text. For each vowel in each of the
speech conditi�ns the formant values Ft and F2 are determined as
the average over a number of tokens from the speech material. Since
the distinction between stressed and unstressed syllables is not
equall y clear in <iil languages� we did not involve that di.stin.ction in our speech material apart from the fact that totally un
stressed syllables have been omitted . For the Dutch data � which
were the result of earlier measurements (Koopmans-van Beinum, 1980)
42
the vowel systems have demonstrated or confirmed that the natural
languages of the world for the greater part display an acoustical
ly and perceptually highly balanced dispersion of vowels in their
vowel system (Liljencrants and Lindblom, 1972; Disner, 1980;
Koopmans-van Beinum, !983).
In order to attain our object as given in the Introduction we ex
amined languages involving few vowels in their vowel system (Japa
nese, Polish), many vowels in their vowel system (Dutch, French,
Hungarian, Frisian), languages with quantitatively paired vowels in
their system (Hungarian, Frisian), and languages involving nasal
vowels (Polish, French) of which only the oral vowels are used here.
In a survey of vowel systems Crothers (1978) indicates that nearly
half of his 209 sample languages have contrasting long and short
vowels. In most cases (70%) the vowels of the two systems are equal
in number and arrangement. Two languages in our investigation
(Hungarian and Frisian) can be considered as representative examples
(in the case.of Hungarian the two vowels [a:] and[:,] are taken as a
pair, although their qualities differ considerably). The most corn-
Japanese u Fig. a e l. Vowel systems for
a the languages used in this study.
Polish u � 0 e 0 e
a
Dutch u y o: - �: e:l
j ce E u a:
French u y 0 Ii.a e
j oe " 5 re E et a a
Hungarian u y u: y: l: 0 !/> e o: �: ...
:I a:
Frisian u y LI! Y: l: 0 al'! 1 o: v\: e:
:) " :>: E: Cl a:
F
44
we averaged the vowel data of the stressed and the unstressed syl
lables from free conversation.
For the Dutch material and for part of the French speech material
formant frequencies (F1 and F2) were measured by spectral analysis
(Wempe, 1979). For the other languages formant frequencies were
measured by LPC analysis. In both methods a more or less stable
central ·se'gment of the vowel was determined as being characteris
tic for 'the whole vowel part, and used for measurement.
Figs. 2 td 7 display �he mean FI - F2 position per vowel per speech
condition, given in a logarithmic Fl - F2 vowel chart, for one
speaker per language. The C-symbol indicates the speaker centroid,
being the overall mean of the measured log Fl and log F2 values
per speaker (the average value is calculated over all non-nasal
vowels and all speech conditions).
F2 ___ _
600 I 800 1000 1200 I I t l ! 1 ! I
300 I ( �;o �
• .---. \ 500 I G O:o�, 800 '
900
Iii VOWELS IN ISOLATION
V VOWELS IN WORDS
F1
F2 ___ _
600 800 1000 1200
300 Hz ' u
' I ! ! !
400 I ..____, I '8 500 f-600 � 0�
700�! 800 �
900 "-'
1000
c
16,00 2000
...
l ( \ e
! I
!I VOWELS IN ISOLATION
V VOWELS IN WORDS
3000 Hz
/''°' VOWELS If\! FREE CONVERSATION e VOWELS IN FREE CONVERSATION
Fig. 2. Mean formant frequencies per Fig. 3. Mean formant frequencies per vowel per speech condition (connected per vowel) in
Polish (speaker P2).
vowel per speech condition (connected pervowel) in Japanese (speaker Jl). C indicates the �peaker centroid.
I I I I
F2 __ _ sf 600 1000 1200 16.00 2900 30,00 Hz II I.' ! I I 1--t-r I lJ1 l 'l
300 r � JJ ::: [_ u� �Bx � 600 \ \. � � � Q.�\'V.... �
Ill VOWELS IN ISOLATION � VOWELS IN WORDS � VOWELS IN FREE CONVERSATION
Fig. 4, Mean formant frequencies per vowel per speech condition (connected per vowel) in Dutch (speaker DJ).
F2 __ _
I'll VOWELS iN ISOLATION v VOWELS IN WORDS
@I VOWELS IN FREE CONVERSATION
Fig. 6. Mean £ormant frequencies per vowel per speech condition (connected per vowel) in
Hungarian (speaker Hl).
F
Ill VOWELS IN ISOLATION "' VOWELS IN WORDS
3000 Hz ' 1' I � �
G VOWELS !N FREE CONVER!::AT!ON
Fig. 5. Mean formant frequencies per vowel per speech condition· (connected per vowel) in French (speaker F2).
F2 ___ _
soo eoo 1000 1200 h--;;l---r ! I I I ; � i u"'"-."'f.... 300 � · -� Hz i
1600 2000 3000 Hz II ly.I '·']I
I ' ·I. 1•·1 1' �"/ ·�71 \ � "{ ff I \'f, JJ._f ! I
���� ·:t �e: ii
� ar:f; Jr; ti a c.,. 11 7 0� � .. /�
800 'X1 Cl . ,/ 900 � "'
"-t � I � Iii t: I I : I ·I li e--!
1000 ' '1 '*--t---+---+--ii--i--t-+-H� 1200
II! VOWELS JN iSOLAT!ON
� VO'NELS IN WORDS ® VOWELS IN FREE CONVERSATION
Fig. 7. Mean formant frequencies per vowel per speech condition (connected per vowel) in Frisian (speaker Fnl).
46
4 . ACOUSTIC SYSTEM CONTRAST
In accordance with the convention used by Koopmans-van Beinum (1980) all formant values were transferred to the logarithmic expressions
LF. 100.101og F. l_ 1.
with i = l or 2.
From these values the Acoustic System Contrast (ASC) was calculated
which is equal to the mean square of the distances in the Fl - F2
plane between the different formant values for the vowels and the
centroid according to the formula
N j " rv. -+) 2 ASC = N .6 \ 1 - c
3=l .J
in which V. is the two-dimensional vector for the vowel J in the J
FI - F2 plane and � is the vector for the centroid C.
In this way we acquired the dispersion or total variance of the
ASC
200
100
� I 1 I' I
n
0 IW.et!on B .,_d,. � CC-nllf:fS3tlon
J1 12 Ja P1 P2 01 02 03 04 F1 F2 Ht H2 Fn1 Fn2
Fig. 8. Histogram illustrating the vowel contrast reduction for three Japanese speakers (Jl, J2, J3), two Polish speakers (PI, P2), four Dutch speakers (DI, D2, D3, D4), two French
speakers (F l, F2), two Hungarian speakers (H 1, H2) and two Frisian speakers (Fnl, Fn2).
47
whole vowei system in each of the speech conditions, or the disper
sion of subsets of the vowel system, e.g. long vowels versus short
vowels, expressed in values comparable among themselves.
Table 1 gives the ASC values for each of the speakers and speech
conditions used in this study, whereas in Fig. 8 the same data have
been displayed in a histogram.
Table l. ASC values in various languages
T Language I Speaker Vowels in Vowels in Vowels in
nr. isolation l words conversation
.
Japanese l 613 I 383 206 2 I 50i 363 254
3 540 322 i49 • i
Poiish I I 744 I 311 273
2 608 405 I 214 I
Dutch 426 418 272
2 400 310 178 3 447 374 197 4 634 529 319
French 692 442 272
2 463 306 157
Hungarian 675 515 341
2 720 484 375
Frisian 621 408 317
2 624 518 256
•
600
I F1
'ol Hz 400
' 500 600
F2
800 10001200 ! , I !°"11 I
•, 'e.
o: / \ Iii
1500 :woo 3000 Hz i I I
I I 'i I ' :y;! : I i i � ' ' ii· ;I: �1:JJ
� ./"'- ' ! ll ' !J • ! i I e. i I c I "
e -H ! I i
I I , : i -+-H " .21:
i I ! j ' .+m. I I ! : I
m VOWELS IN ISOLATION '*' VOWELS !I'll WORDS tll VOWELS IN FREE CONVERSATION
Fig. 9. Mean formant frequencies per vowel per speech condition (connected per vowel) in
Hungarian (long vowels, speaker H2).
600
300 Hz I 400 I
500 500
F2 ___ _
800 10,00 1?00 I i
e:
rnoo 2000 3000 Hz I I l I Y:i I i:i ! I
ITT ' I�: i I <) +1 l I �e; j i
�,(!: +i c t) i i I! +-! I . _;1: � I I
tt1 � 1000 �
1 I i+H-1 12 o o *-�-r-1 --+--1-t1-+--+1 �1-+-! .L..ji I
fill VOWELS IN ISOLATION
T VOWELS IN WORDS 9 VOWELS IN FREE CONVERSATION
Fig. 11. Mean formant frequencies per vowel per speech condition (connected per vowel) in Frisian (long vowels, speaker Fn 1 ) .
48
600 I I
F1 JOO L I I Hz I I 400 ! I
500 [ t !
600
F2 ___ _
aoo 1000 1200 i ! i i ! I
1200
1600 2000 3000 Hz
.l)J!� 4c � I 1 �e 11 ii tH ;m ! '
I
!!! VO'NELS IN ISOLATION .., VOWELS IN WORDS ® VOWELS !N FREE CONVERSATION
Fig. J 0. Hean formant frequencies per vowel per speech condition (connected per vowel) in
Hungarian (short vowels, speaker H2).
F1
F2 ___ ,..
ra VOWELS !N ISOLATION v VOWELS IN WORDS tli VOWELS IN FREE CONVERSATION
Fig. 12. Mean formant frequencies per vowel per speech condition (connected per vowel) in Frisian (short vowels, speaker Fnl).
49
5. COMPARISON OF THE RESULTS IN VARIOUS LANGUAGES
As can be seen in the vowel diagrams (Fig. 2 to Fig. 7) all lan
guages in this study display a centralizing shift if one goes from
vowels pronounced in isolation to vowels in conversation. The cal
culated values of the acoustic system contrast ASC give the possi
bility to compare the degree of reduction between the various
speech conditions and languages as well as between the various
speakers (Table and Fig, 8). Thi::re is no indication for a system-
atic difference between vowel systems with a small number of vowels
and those with a large number of vowels, as was hypothesized in the
Introduction. Although we did not yet perform any statistics on
these data, and the number of speakers per language is very low, it
seems to be justifiable to conclude that vowel contrast reduction
occurs as a universal property of all languages, the differences in
the amount of reduction being mainly speaker dependent. This suppo
sition is reinforced by the fact that speaker 3 of Japanese, spe'.lker 2 of Dutch, and speaker 2 of French, are the very speakers with
the smallest ASC values in free conversation, and were judged by
their interviewers to be the speakers least careful in pronouncing.
Table 2 . ASC values for the subsystems of long and short vowels for two languages.
Languag8 Speaker\ . Vowels in Vowels in I Vowels in l I nr. isolation I words conversation
long short long short long short
I 428 255 Hungarian I 1 768 583 695 336 I
2 859 532 I 596 372 I 430 32i I f
Frisian l 680 562 460 355 354 280
2 655 593 563 l 473 289 223 J
I
ASC 800
n
H1
50
Fn 1
OJ lsolailon CTJ words � Conwt&lliion
Fig. 13. Histogram illustrating the vowel contrast reduction for two Hungarian and two Frisian speakers, where the ASC values are indicated separately for the long vowels (left) and the short vowels (right).
In the case of languages with equal numbers of long and short vowels
(Hungarian and Frisian) the vowel diagrams for the long and short
vowel s are given separately in Fig. 9 to Fig. 12, whereas the sepa
rately calculated ASC values are given in Table 2 and in Fig. 13 in a histogram. Here we find for the long vowels in each speech situa-
tion a larger value of the ASC than for the short vowels, which il
lustrates the fact that in general long vowels a.re more peripheral
than the corresponding short ones.
6 . VOWEL SYSTEMS WITH NASAL VOWELS
In our initial attempt to include in this study some languages in
volving nasal vowels in their vowel system (cf. Polish and French),
we came across several problems, some of them of methodological
nature.
5!
In the first place part of the recorded nasal vowels turned out t o
yield very unreliable measurements, o r even measuring was impossi
ble. This made our data on nasal vowels very incomplete.
If measuring was possible, a second problem arose in the applica
tion of the ASC algorithm, because of the special, 'nasal1 low for
mant frequency, beside Fl and F2 frequencies. Are we allowed to
leave these nasal formants out of consideration, and can the Fl and
F2 of nasal vowels be compared with those of oral vowels without
further ado? These two problems decided us to exclude the nasal vow
els from the present study. But this decision caused another method
ological problem: If indeed a vowel system is acoustically and per
ceptually highly balanced, then the nasal vowels will make up an
essential part of this balance and therefore must not be excluded.
In the case of the Polish vowel system we believe the loss is not
too serious, since both nasalized vowels ( 0 and e ) are more or
less in balance in the vowel system. The problems for the French
vowel system are much more serious, since exclusion of the nasalized
F2 ___ _
eoo aoo 1000 1200 1600 2000 1---,---l-t -.-�;-,-�·�..,....--t-_JI I !.·
300 Hz 400 �
I i ' I
,. -s• I if ,..#� ' 500 -
600
70�1 SOO� 900 �· 1 000
• 1 I I ! 1200 · . I i '
1=: Gl VOWELS iN ISOLATION 2= � VOWELS IN WORDS
sooo Hz . , I i I i )� I, ! I I . I I �n I
3,. e VOWELS IN FREE CONVERSATION
Fig. 14. Shift of the centroid for two French speakers.
52
vowels ( 3, re, a and E ) might disturb the balance. Indeed, re
sults concerning the French oral vowels display deviations from
the other results: whereas in all other cases the centroids per
speech condition stay stable, we can establish a clear shift of these
centroids into the direction of 'high front' for both French speak
ers (Fig. 14). Nieboer (forthcoming) who performed the measurements
of the French speech material, suggests a phenomenon of 'anterior
is�'. being the very French articulation base, instead of the cen
tralizing reduction tendency in other languages when going from
vowels pronounced in isolation to vowels in free conversation.
However, which is the role of the nasal vowels in this respect? In
any case a role we had to leave out of this study but one that earns
more attention in future research on vowel systems.
. -
53
REFERENCES
Crothers, J. (1978). Typology and Universals of Vowel Systems.
Universals of Human Language. Vol. 2: Phonology.
Edited by J.H. Greenberg. Stanford University Press,
Stanford, California.
Disner, S.F. (1980). Insights in vowel spacing: results of a lan
guage survey. UCLA Working Papers 50: 70-92.
Hellwag, C.F. (1781). Dissertatio inauguralis physiologico medica
de formatione loquelae. Tiibingen. Facsimile with a
translation in Dutch, Institute of Phonetic Sciences,
Arns terdam l 96 7.
Koopmans-van Beinum, F.J. (1980). Vowel Contrast Reduction: An a
coustic a.rid perceptual study of Dutch vowels in
various speech conditions. Diss., University of
Amsterdam.
Koopmans-van Beinum, F.J. (1983). Systematics in vowel systems. In:
M.P.R. van den Broecke� V.J. van Reuven and W. Zonneveld
(Eds.), Sound structures: Studies for Antonie Cohen.
Faris Publications. Dordrecht, i59-171.
Lehiste, L. (1970). Suprasegmentals. MIT Press, Cambridge, Mass.
Liljencrants, J. & Lindblom, B. ( J 972). Numerical Simulation of
Vowel Quantity Systems: The Role of Perceptual Con
trast. Language 48, 839-862.
Nieboer, G.L.J. (forthcoming). Kl:i.nkercontrastreduktie in bet Frans.
To appear as IFA-report 77, Institute of Phonetic
Sciences, Amsterdam.
Ruhlen, M. (1975). A Guide to the Languages of the World. Language
Universals Project. Stanford University.
Wempe, A.G. (1979). An experimental segment spectrograph based on
some notes on frequency analysis of speech segments.
Proceedings of the Institute of Phonetic Sciences
Amsterdam 5, 44-102 .