A comparative acoustic study of English and Spanish vowels
Ann R. Bradlow a)
Cornell University, Department of Modern Languages and Linguistics, Ithaca, New York 14853
(Received 30 November 1993; accepted for publication 14 November 1994)
Languages differ widely in the size of their vowel inventories; however, cross-linguistic surveys
indicate that certain vowels and vowel system configurations are preferred. A cross-linguistic
comparison of the acoustic vowel categories of two languages that differ in vowel inventory size,
namely, English and Spanish, was performed in order to reveal some of the language-specific and/or
universal principles that determine the acoustic realization of the vowels of these two languages.
This comparison shows that the precise location in the acoustic space of similar vowel categories
across the two languages i determined, in part, by a language-specific base-of-articulation property.
These data also suggest hat the relatively crowded acoustic vowel space of English may be
expanded with respect o the relatively uncrowded acoustic vowel space of Spanish; however, this
effect is variable depending on the syllable context of the English vowels. Finally, the data indicate
no difference in the tightness of within-category clustering for the large versus the small vowel
inventory.
PACS numbers: 43.70.Hs, 43.70.Kv
INTRODUCTION
Surveys of segment inventories indicate cross-linguistic
preferences for certain vowels and for certain vowel inven-
tory configurations. For example, in a survey of 317 lan-
guages, Maddicson (1984) finds that vowel inventories in
this sample vary from having three to fifteen distinct vowel
qualities, with two-thirds of the languages having between
five and seven distinct vowel qualities. Additionally, the spe-
cific vowels that comprise these statistically preferred vowel
inventories tend to be the same. For example, five-vowel
systems tend to have/i,e,a,o,u/, seven-vowel systems tend to
have these five vowels plus/e/and/o/, and six-vowel sys-
tems usually have/i,e,a ,o,o,u/. Furthermore, the vowel in-
ventories of the vast majority of the world’s languages in-
clude the three vowels that define the extremes of the general
vowel space, namely/i,a,u/. Accordingly, these three vowels
are known as the \"point vowels,\" and have been afforded a
special status in theories of vowel systems.
These cross-linguistic endencies have led to the hypoth-
esis that there are constraints on possible speech sounds and
their cooccurrence, which have their source in general lin-
guistic, or physical (i.e., auditory and articulatory) con-
straints. However, the exact nature of these constraints and
their interaction that produces the observed inventories is not
yet fully understood. This study is thus motivated by a gen-
eral interest in the effect of inventory size on the acoustic
vowel spaces of different languages, and in how this effect
might reveal some of the universal and/or language-specific
constraints leading to the observed patterns in sound inven-
tories of the languages of the world. Specifically, this paper
explores the effect of inventory size on the acoustic realiza-
tion of vowels in a language with a relatively large vowel
inventory, General American English, and in a language with
a relatively small vowel inventory, Madrid Spanish.
a)Current affiliation: Speech Research Laboratory, Psychology Department,
Indiana University, Bloomington, IN 47405.
I. GENERAL APPROACHES TO THE STUDY OF
VOWEL INVENTORIES
Previous work has led to the development of several
theoretical positions regarding the structure of vowel sys-
tems. Dispersion theory (DT) claims that speech sounds are
selected via constraints that are based on a principle of suf-
ficient perceptual contrast. In this theory the vowels of a
given language are arranged in the acoustic vowel space so
as to minimize the potential for perceptual confusion be-
tween the distinct vowel categories. Using computer pro-
grams to generate the optimal configurations for vowel sys-
tems of various sizes, this approach to vowel inventories has
proved fairly successful (Liljencrants and Lindbiota, 1972;
Lindblom, 1975, 1986; Disnet, 1984). However these inves-
tigations of DT focus exclusively on intercategory distance
as the determiner of vowel system configuration in a univer-
sally defined acoustic vowel space. As a result, this approach
fails to account for the observation that certain languages,
such as Swedish with nine vowels and Danish with ten vow-
els, crowd their vowels into a small corner of the entire
vowel space rather than dispersing them throughout the
available space (Disnet, 1983). In more recent developments,
the dispersion principle has been expressed as a principle of
sufficient, rather than maximal, contrast (Lindbiota, 1989,
1990). Furthermore the theory has been extended to account
for within-speaker variation. For example, Moon and Lind-
blom, 1989) show that under circumstances that require clear
speech, a speaker’s vowel space will be expanded relative to
his or her casual speech vowel space.
In a study that addresses the prediction of DT that vow-
els will be maximally dispersed in the acoustic space, Jong-
man et al. (1989) compared the relatively crowded vowel
spaces of English (with 11 monophthongs) and German
(with 14 monophthongs) with the relatively uncrowded
vowel space of Greek (with just 5 monophthongs). These
authors found that the crowded vowel spaces of English and
German are expanded relative to the uncrowded vowel space
1916 J. Acoust. Soc. Am. 97 (3), March 1995 0001-4966/95/97(3)/1916/9/$6.00 ‚ 1995 Acoustical Society of America 1916

of Greek. 1 Jongman et al. plotted the vowels of these lan-
guages in the auditory-perceptual space proposed by Miller
(1989). This representation scheme is designed to normalize
the data poi.nts for both inter- and intraspeaker differences by
representing speech sounds in terms of log ratios of the fun-
damental frequency and the first three formants. The results
of this study support the hypothesis that the acoustic realiza-
tion of vowel categories is dependent on inventory size and
suggest hat some version of the dispersion principle does
indeed hold true across languages: the larger the inventory,
the more \"expanded\" the acoustic vowel space.
The quantal theory of speech (QTS) (Stevens, 1972,
1989) suggests an alternative approach to vowel systems.
This theory is based on the observation that for certain pa-
rameters of the articulatory domain, there is a nonmonotonic
relation between variation in the articulatory configuration
and its acoustic onsequences. Similarly, certain changes in
the acoustic signal over some part of the range of a particular
parameter are nonmonotonically related to the corresponding
auditory response of the listener. In other words, according to
this theory, there are certain regions of stability in the pho-
netic space. In particular, it is claimed that there are stable
regions corresponding to the point vowels/i/,/a/, and/u/.
Thus this theory predicts that the point vowels should be in
approximately the same locations across all languages, re-
gardless of vowel inventory size. Furthermore, QTS predicts
that, since the point vowels are in phonetically stable re-
gions, they should show less within-category variability than
nonpoint vowels.
Evidence that certain changes in the articulatory domain
are nonmonotonically related to their acoustic consequences
comes from the observation that there are acoustic properties
which are relatively insensitive toarficulatory perturbation.
For example, in the case of nonlow front vowels which have
a fronted tongue body, the locations in the frequency spec-
trum of the second, third, and fourth prominences are rela-
tively insensitive to perturbation n the tongue-body position
along the anterior-posterior dimension (Stevens, 1989). For
this range, F2 is at a maximum and is within a few hundred
hertz of F3. In contrast, the frequency of F1 varies mono-
tonically with the size and position of the articulatory con-
striction. This result is presented by Stevens as evidence of a
stable region in this corner of the vowel space, which corre-
sponds to the point vowel/i/. Furthermore, in an articulatory
study of/i,a,u/ in General American English, Perkell and
Cohen (1989) find that for each of these vowels there is both
an articulation-to-acoustic saturation effect, and a muscle
contraction-to-displacement saturation effect. In other words,
these authors find that over a range of changes in the articu-
lation of these vowels, the acoustic output is relatively stable.
DT and QTS both propose general universal principles
to account for the observed cross-linguistic tendencies re-
garding vowel inventory size and structure. In contrast to
these approaches, the notion of a language-specific base-of-
articulation is presented as an account for the observation
that similar sounds across two languages can differ due to a
consistent, language-specific adjustment of the articulators.
This notion has been a part of the traditional phonetic litera-
ture over the ages: Disnet (1983) cites its origin as the work
of John Wallis in 1653 (set: Kemp, 1972). However, within
the tradition of generarive phonology, the idea of a language-
specific articulatory setting has often been considered outside
of the area of interest of theoretical linguistics. For example,
in The Sound Pattern of English, Chomsky and Halle (1968)
consider this aspect of speech as extragrammatical, nd thus
as part of the performance aspect of language, rather than
part of the grammatically determined competence aspect.
Nevertheless, several investigators have pointed to the
importance of the notion of a base-of-articulation forprovid-
ing insightful analyses of both phonological and phonetic
observations. In studies that have tested the predictions of
DT and QTS, phonetic differences between similar segments
of different languages have often been observed. For ex-
ample, Lindau and Wood (1977) investigate the vowels of
three related Nigerian languages, Yoruba, Edo, and Ghotuo,
all of which have phonemically equivalent seven-vowel sys-
tems, and find that the vowel spaces of Edo and Ghotuo are
very similar. Holyever, contrary to the prediction of the dis-
persion principle, the vowel space of Yoruba deviates from
the structure of the other two seven-vowel systems and is not
maximally dispersed. Similarly, Disnet (1983) finds that the
seven-vowel systems of Yoruba and Italian differ from each
other in their locations of the seven vowels in the acoustic
space. Disnet documents additional cases of systematic dif-
ferences across the vowels of several Germanic languages;
for example, she finds that the vowels of Danish are system-
atically articulated with a higher tongue position, as it is
reflected by F1, than the vowels of English. Disnet claims
that these data demonstrate the role of a language-specific
base-of-articulation property in the phonetic realization of
vowel phonemes. In particular, this type of language-specific
effect is seen in across-the-board shifts of the vowels of one
language relative to similar vowels in another language.
In light of the theoretical claims and experimental evi-
dence discussed above, the Fresent study was undertaken as a
direct means of assessing the contributions of language-
specific and general univers:d principles in the acoustic real-
ization of vowel categories across languages with relatively
large versus small vowel inventories. English and Spanish
were chosen for this study because of the large difference
between the sizes of their vowel inventories: English has
more than double the number of stressed monophthongal
vowels than Spanish. Additionally, the five-vowel system of
Spanish is statistically very common, whereas the 11-vowel
system of English is uncommonly arge (Maddieson, 1984).
Furthermore, the vowel systems of these two languages are
similar in that they vary along the same dimensions (neither
language has contrasfive rounding, length, or nasalization).
Consequently, the principal difference between these vowel
systems is in the number of vowels. Thus this English-
Spanish comparison represents a comparison of an unusually
large vowel inventory with a smaller cross-linguistically
common vowel inventory. T e expectation is that this differ-
ence between the two vowel inventories will highlight both
the differences and similarities attributable to language-
specific and/or universal aspects of vowel production.
In this comparison, vowel formant measurements from
each of the two languages are evaluated in terms of the pre-
1917 J. Acoust. Soc. Am., Vol. 97, No. 3, March 1995 A. Bradlow: Comparative study of English and Spanish vowels 1917