Effectof final-syllable position onvowel'duration in infant babbling D. K. Oiler Mailman Centerfor Child Development,Universityof Miami, Miami, Florida 33152
Bruce L. Smith Child Developmentand Mental Retardation Center, Universityof Washingtion,Seattle, Washington98195 (Received 24 January 1977; revised 27 May 1977)
•
Previousresearchhas suggested possiblespeech-productionand/or speech-perception-oriented causesfor the temporal phenomenonreferred to as final-syllablevowel lengthening,yet little conclusiveevidencehas been adduced for either possibility. The present study representsan attempt to provide additional information concerningthe nature of this phenomenonby approachingthe issue from a developmental perspective.A much smaller amount of final-syllablevowel lengtheningwas observedin the premeaningful vocalizationsof a number of very young infants than in phoneticallycomparableutterancesproducedby adult speakersof English. It is concludedon the basis of this preliminary evidencethat extensivefinalsyllable lengthening observedin the productions of adult speakersof English seeminglyconstitutesa learned behavior.
PACS numbers: 43.70.Gr, 43.70.Bk
INTRODUCTION
Utterance-final
that unlearned production properties were the source. syllables in English are substantially
longer in durationthan comparablenonfinalsyllables.•'•' According to data from a wide variety of studies employing both real and nonsense inventories, the ratio of the duration of vowels in utterance-final syllables to comparable vowels in nonfinal position is characteristically
between 1.4 and 2.0 in English (i.e.,
final vowels are
40%-100%longer).•' A numberof other languagesalso appear to show substantial
duration increm6nts
for final-
syllablevowels,e.g., German,xRussian• 3,4andSwedish.s'6 Onthe otherhand,somelanguages seemto reveal little (if any) final-syllable vowel lengthening, e.g.,
Finnish,•.,7-9Estonian,8-x0andJapanese. TM(It is not known whether such languages evidence final-syllable
consonantlengthening. )
It hasbeensuggested by someinvestigators, •.,5,6,x•. that final-syllable "lengthening"may be the result of unlearned, inherent properties of the speech production mechanism, but it has also been proposed that this phe-
nomenonis learnedandservesa perceptualfunction.2,x3, x4 Because some languages show considerable final-syllable vowel lengthening, while others reveal at most only minimal amounts, it is difficult to maintain a strict production-oriented explanation. On the other hand, small
If no lengthening occurred in the productions of young children, its later occurrence could probably be attribi•ted to learned factors of perception. Although the issue is undoubtedly more complicated than the simple model suggests, this study presents preliminary findings concerning developmental aspects of final-syllable lengthening and attempts to interpret them on the basis of these simplified notions.
Between5 and 12 monthsof age (i. e., prior to the onset of meaningful speech) normal infants produce vo-
calizationsgenerally referred to as "reduplicatedbabbling." Theseutterancestypicallyconsistof a string of several phoneticallysimilar syllables (CVCV...); the reduplicated consonant is usually an unaspirated labial or alveolar stop or nasal and the vowel generally
hasa qualitytranscribedas /a, e, or ae/. Thesetypes of utterances
involve
acoustic
characteristics
which
generally provide relatively clear evidence of opening and closure of the vocal tract, and the spectral changes between the consonants and vowels allow for relatively accurate segmentation with high interobserver reliability. I. PROCEDURES
ably do not serve a perceptual function; this implies a
Table I presents information about the six subjects whose reduplicated utterances were analyzed. From as few as seven to as many as 40 separate reduplicated ut-
production-orientedsourcein suchcases.x•.,•3Thus, at
terances
this time the source of utterance-final-syllable
terances ranged from two to six syllables in length.
amountsof vowel lengthening(as in Finnish, etc.) prob-
lengthen-
ing, as well as the role of production- and perceptionoriented
factors
remains
rather
unclear.
were
considered
for the individual
infants.
Ut-
About45%were two syllables, 25%three syllables, 20% four syllables, andonly 10%were longerthanfour syllables.
One question which often arises in discussions regarding such issues is whether the effects of final-syllable lengthening are observable in the speech of very
infant interacted with a parent and/or an experimenter.
young children.
An attempt was made to maintain a relatively constant
A simple model would suggest that if
final-syllable lengthening were detected in their productions, it would probably be reasonable to conclude 994
J. Acoust. Soc. Am., Vol. 62, No. 4, October 1977
High-fidelity
audio recordings were made while the
mike-to-mouth distance and yet not restrict the infant physically. Utterances selected for analysis were charCopyright ¸ 1977 by the Acoustical Society of America
994
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 193.0.65.67 On: Wed, 17 Dec 2014 10:34:59
995
D. K. Oiler and B. L. Smith: Vowel duration in infant babbling
995
thefirst,.second, andthirdsyllables, i.e., CI•CV,
TABLE I. Experimental subjects, ages, and number of reduplicated utterances
CVCI•,CVCVCV,CVCI•CV, andCVCVC•.
considered. Number
In order to maximize comparability of infant and adult utterances, a second inventory was included. Inventory
of
Age
utterances
Subject
(in months)
considered
I1 12 13 14 15 16
10 8 8 12 11 8
Two consisted to two- and three-syllable words of the same segmental form as those in Inventory One, but
withrelativelyequalstressoneachsyllable-CI•CI•and
40 22 11 7 7 21
CIdCIrCle.Theinstructions to thespeakersmadeno reference to duration of syllables.
If speakers found the
task difficult (e.g., if they persisted in producinga single primary stress on a given syllable of the utter-
ance)theywere told to try to sound"monotonous, like
108
babies do in reduplicated babbling."
acterized by reasonably good signal-to-noise
ratios.
Broadbandspectrograms (Kay 6061-B) were made of each of the utterances.
Although infant reduplicated utterances show minor segmental variation from syllable to syllable and although the problems of segmentation are somewhat more difficult with infant utterances than with adult productions, the basic results of the present work did not seem to be substantially affected by measurement problems. The primary difficulty in segmenting the consonantvowel sequences was to determine the t½•-•nination of the final vowel of the utterance. In many cases the harmonic structure
observed
for
the final
vowel
evidenced
two dis-
tinct properties. There was a "voiced" portion of the
segment,andan "unvoiced,"or breathy, portion. It is our feeling that the more meaningful measure of vowel duration should include/•oth the voiced and the breathy portions. For the sake of comparison, however, two
analyses were performed: (1) a conservative measure of vowel duration, which included only that portion of the vowel characterized by voiced harmonic structure;
and(2) a nonconservative measure, whichincludedboth the voiced and the breathy portions of the vowel.
As
previously mentioned, there was generally a relatively high degree of interobserver agreement. By comparing two sets of independently measured spectrograms, it was determined that the average error of measurement for adult vowels was approximately 4 msec and for in-
fantvowelsabout8 msec,,whetherthe conservativeor nonconservative
criterion
was used.
In analyzing the adult data from Inventory One, final stressed vowels were always compared with nonfinal, stressed vowels and final, unstressed vowels were compared only with nonfinal, unstressed vowels. In the equally stressed utterances, all syllables were compared since all were supposed to be of comparable stress. Utterances
stress
ducing equally stressed syllables difficult and that the English-speaking phonetic judges of the productions
found their task similarly difficult.) For both Inventory One and Inventory Two there were six speakers, who produced four repetitions
of each stress type from a
randomized list. As with the infant data, Kay Sonagrams were made from high-fidelity tape recordings of each utterance; the segmentation criteria employed were
also
utterance
the
same
as those
used
for
the infant
s.
II. RESULTS
Utterances from the adult subjects of the present study showed strong evidence of final-syllable vowels being
longer than nonfinal vowels (Table II). Consid•eringthe data of Inventory One (only one syllable bearing primary
stress), the average ratio of final to nonfinalvowel durations for all conditions was 1.49 when using the conservative segmentation criterion and 1.67 using the nonconservative criterion. Thus, final-syllable vowel increments were comparable to those established by other
investigators•'•"5'•' for final-syllable vowels in closed syllables.
The vast majority of the infant utterances in our sam-
which did not seem to have primary
on all syllables were discarded. (It shouldbe acknowle•lgedthat these English speakers foundthe task of pro-
All six subjects showed final lengthening for
both the conservative
and nonconservative
criteria.
ple evidencedopen (CV) final syllables. Since the adult Considering the data from Inventory Two (Table III) in
utterances in most previous research on the effect of position-in-utterance contained ½l•$½d final syllables,
which all syllables bear primary stress, the average
it was decided
ratios were some•whatlower--1.31 for the conservative
to obtain
further
information
from
a
group of adults pronouncing reduplicated nonsense utterances which would be comparable with the infant sample. All the adults were asked to pronounce reduplicated two- and three-syllable sequences consisting of nasal or stop consonants and mid or low unrounded
vowels(e.g., baba, dada, nene); noneof the utterances had a final consonant. Six adults, all of whom were native speakers of English and all but one of whom were
unaware of the experimental hypotheses, were asked to pronounce Inventory One, which included utterances in which primary stress was varied systematically across
criterion
and 1.42
for the nonconservative
criterion.
Again, all six subjects showedfinal lengthening for both sets of criteria. A t-test did not reveal significant differences between lengthening of final syllables by adults in Inventory One versus Inventory Two, either for the
conservativecriterion (t = 1.39, dr= 5, • > 0.10) or for the nonconservativeone (t = 1.86, df= 5, p > 0.10). (While it is clear that final-syllable lengthening did occur when stress was relatively equal for all syllables, the lack of
a significant difference between the results for Inventory One and Inventory Two should not be interpreted as
J. Acoust. Soc.Am.,Vol.62, No.4, October i977 Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 193.0.65.67 On: Wed, 17 Dec 2014 10:34:59
996
D.K. Oiler and B. L. Smith:Vowel durationin infant babbling
TABLE
II.
Final
to nonfinal
vowel durations
TABLE
for
of stressed
vowels
IV.
Ratios
of final
996
to nonfinal
vowel
durations
for
in-
fant subjects. A ratio greater than 1.00 represents final-syllable lengthening.
adult subjects- A ratio greater than 1.00 represents œinal-syllable lengthening. Average of ratios
•
and unstressed Nonconservative
vowels (Inventory One).
c riter
Nonconservative
Conservative
criterion
criterion
Subject
Conservative
ion
c rite r ion
Ratio
era
228/194= 1.18 267/217=1.23
Ratio
oa
Ratio
206/125=1.65 260/114=2.28 215/141 =1.52 196/123= 1.59 271/194=1.40 252/158=1.59
20 20 17 11 18 16
178/125=1.42 235/114=2.06 202/141 =1.43 161/123= 1.31 244/194=1.26 233/158=1.47
16
320/258
18
258/239 = 1.08 321/339=0.95 195/210=0.93 X=1.10
12 12
18
Ratio
70
62
= 1.24
104
76 133
64
er
173/174=0.99 240/210=1.14 305/248=1.23 181/239 = 0.76 267/329=0.81 195/210=0.93 X=0.98
53 96 94 76 123
64
18
aStandarddeviation of vowel durations: This figure represents the average of the standard deviation for final and nonfinal aStandard
deviation
of vowel durations:
vowels.
This
figure represents the average of the standard deviation
for final
stressed
and unstressed
and nonfinal
vowels
in both
conditions.
data from Inventory One, t-test comparisons of the adult and infant data showed significantly greater lengthening
strong evidencethat no such difference might ultimately exist, however, since the sample size in this experi-
ment was relatively small. ) Table
jects.
IV
summarizes
the data for
the six infant
sub-
Based on the conservative criterion, two of the
ratios for the adults, whetherthe conservative(t = 3.69, df = 5, p < 0.05) or nonconservative(t = 4.12, df = 5, p