ArticlePDF Available

Figures

Content may be subject to copyright.
A Contrastive Analysis of Schwa in English and Portuguese
Adriana Marusso1, Thaïs Cristófaro Silva2
1Federal University of Ouro Preto, 2Federal University of Minas Gerais, Brazil
amarusso@uai.com.br, thaiscristofaro@gmail.com
1. Introduction
The nature of representations is an important topic in the current phonological literature.
Results indicate that languages use important and detailed phonetic information to categorize
its sounds (Pierrehumbert, 1999). Furthermore, when the relationship between perception and
production is addressed the notion of a single underlying representation is deeply questioned.
This paper intends to be a contribution to this current debate by analyzing the phonetic and
phonological properties of schwa in Portuguese and English. The varieties to be analyzed are:
Minas Gerais Brazilian Portuguese (BP) and Southeastern British English (RP). The main
question to be addressed is whether or not the schwa has similar properties when related to a
set of vowels, as in English, and when it relates to a single vowel, as in Portuguese. Similar
properties to be investigated are the quality of schwa depending on the nature of adjacent
vowels and also the durational patterns of schwa in both languages. We first present a quick
review of the literature in order to set the relevant theoretical background. Then, we indicate
the major objectives of the paper and we describe the methods used in the experiment.
Finally, we discuss the results indicating the major conclusions to be drawn from the research
presented in this paper.
2. Background
Variability in the articulatory and acoustic realization of segments has been studied from
different approaches. In terms of articulatory constraints on variability, there are essentially
two descriptions of the differences between vowels with respect to the amount of contextual
variation. One of these is the target undershoot approach and the other one is the gestural
approach to coarticulation. The former approach suggests that the degree of coarticulation
depends on the distance between, and the time available to reach segmental targets (Stevens
& House, 1963; Lindblom, 1963). The second approach understands coarticulation as
regulated by the degree of articulatory compatibility between segments (Recasens, 1986,
1991; Browman & Goldstein, 1992).
Different authors have studied the variability of the neutral vowel in English, namely
the schwa. Recasens (1991) claims that schwas are the vowels with the highest degree of
variability. The author attributes this to “the inherently low requirements on the tongue
configuration during the production of this vowel” (p.187). Keating (1988:30) suggests that
featural “underspecification may persist into phonetic representation” (p.30). If a segment is
phonetically unspecified for a certain feature, instead of assimilating the feature values of
neighboring sounds, it is simply interpolated through. In the same line, Bates (1995) shows
that medial schwa is interpolated in RP English. However, Bates shows that a final schwa in
BE does not suffer interpolation. This indicates the different behavior of a schwa in medial
position and in final position. Browman & Goldstein (1992) also investigated English schwa
and proposed that schwa has a target, but “its target is the mean of all the vowels, and is
completely overlapped by the following vowel” (p.54). This paper investigates this claim with
regards to final utterance schwas.
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 351 -
Taking into account that in English basically any full vowel may reduce to schwa, it is
possible to think that this phonological fact is phonetically realized by a neutral vowel that
has a target which is the mean of all full vowels. This is thus the main claim of Browman and
Goldstein (1992). If their proposal is correct then, a different phonetic behavior from the
schwa in English could be expected for the Brazilian Portuguese schwa, which is exclusively
the unstressed variant of /a/. This paper intends to contribute to a better understanding of these
facts. That is, whether or not a schwa has similar properties when related to a set of vowels, as
in English, and when it relates to a single vowel, as in Portuguese.
On the other hand, vowel quantity or duration is another relevant aspect of vowel
reduction. In Portuguese, Cagliari & Abaurre (1986) state that the final syllable in an
utterance is always the longest one, no matter whether it is stressed or not, presenting thus a
reduced vowel with significant duration. Massini-Cagliari (1992) shows that accented words
are the most prominent in an utterance at all levels: pitch, duration, vowel quality, etc. and the
least vulnerable to compression. At word level, the stressed syllable is the most prominent and
the one that does not suffer vowel compression in contrast to unstressed vowels. However, it
would be interesting to investigate whether such a strong vowel, as it is the case of a stressed
one, may interfere on the production of other adjacent vowels.
This is somewhat what Turk & Sawusch (1997) carried out for English. They showed
that accentual lengthening goes beyond the stressed syllable. In fact, in two-syllable words
with the stress pattern stressed-unstressed, that is (s w), not only the primarily stressed
syllable but also the post-tonic one is lengthened. This paper intends to investigate Turk &
Sawusch (1997) with regards to Brazilian Portuguese and verify this claim for Southeastern
English.
Crosswhite (1999) proposes a phonological analysis for vowel reduction in several
languages and states that a schwa is non-moraic and therefore it is never longer than 40ms.
We will also consider in this paper Crosswhite’s claim with regards to BP and BE.
Considering the facts exposed above this paper investigates mainly the following
issues: a) the quality and quantity of a final schwa in BE and BP, b) whether an adjacent
vowel affects a schwa and c) Crosswhite’s (1999) claim that a schwa is never longer than
40ms. The following section describes the major characteristics of the experiment.
3. Method
3.1 The data
A specific experiment was designed to answer the three questions posed above. In order to
obtain comparable testing material in both languages, each oral vowel of BP was equated to a
vowel in BE and almost homophonous words were found in both languages. Examples of
these words are listed in Table 1.
Table 1. Sample words investigated in the experiment
Brazilian Portuguese Bristish English
Cida [sid] cedar [sid]
sêca [sek] sicker [sk]
peca [pk] packer [pæk]
paca [pak] parker [pk]
cola [kl] collar [kl]
Lola [lol] lawler [ll]
luta [lut]
looter [lut]
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 352 -
Table 1 lists sample words with each stressed vowel in both languages. There were
about ten different words for each stressed vowel in each language. Each word presented two
syllables with penultimate stress, namely a (s w) pattern. Table 1 shows that seven primarily
stressed oral vowels in BP /i e a o u/ and seven primarily stressed oral vowels in BE /i æ
  u/ were considered.
Participants were presented with a printed frame which showed a question to be read:
Did he say cedar? for BE and Ele diz Cida? for BP. Notice in the examples that the tested
words were accented and appeared in the final position in the utterance. Words also had a
CV.CV syllable pattern in order to minimize coarticulation. Nasal consonants were avoided
due to potential nasalization in Brazilian Portuguese. Thus, all data presented only oral
vowels in both languages.
Brazilian Portuguese participants were presented with a printed frame containing
Brazilian Portuguese sentences. English participants were presented with a printed frame
containing English sentences. Further information as to how the experiment was carried out is
provided in the following sections.
3.2 Elicitation and recording
Four native speakers of Brazilian Portuguese and four native speakers of British English
participated as volunteers in the experiment. All participants were female making thus data
comparable for acoustic measurements. None of the participants reported any hearing or
speaking difficulties.
Participants were presented with a six pages printed frame which contained
instructions and the list of sentences to be read. The introductory page provided a set of
instructions indicating to participants how they should carry out the reading. Instructions
requested them to read each sentence, as naturally as possible, without pausing between
words. It was also suggested that participants read sentences individually, pausing briefly
between them, as to avoid reading the sentences as if they appeared on a list.
The following pages of the frame included the testing sentences and also additional
sentences which were not analyzed in the experiment. These additional sentences were
intermixed with the others so participants could not identify which sentences were to be
analyzed.
The experiment was monitored for errors. Thus, if a mistake was detected, such as
misreading of a word, or the lack of attention as to which was the next sentence to be read,
then the participant was requested to read the relevant sentences again.
In each language, for each tonic vowel, there were about ten different words (66 words
for Brazilian Portuguese and 63 words for British English). Each speaker read each word
once. The total number of words to be analyzed in both languages was 516. Each of the four
Brazilian speakers produced a total of 66 utterances; therefore, the total number of words to
be analyzed for Portuguese was 264 utterances. Each of the four English speakers produced a
total of 63 utterances; therefore, the total number of words to be analyzed for English was 252
utterances.
Recordings were made in the sound-attenuated recording studio at the Universidade
Federal de Minas Gerais, Brazil; using a DAT TCD-D8 – Sony and a ECM-T15/t115
microphone. For the English data, recordings were made within a sound proof recording
studio at the University of Edinburgh, using a Soundtrac 200B studio Console, a Sony
PCM2700A DAT recorder and a AKG Blue Line electric condenser microphone (SE300B
amplifier + CK98 mic. capsule).
The data were recorded to Digital Audio Tape at a sampling frequency of 44.1 kHz
and a recording level of -10db this level did not fluctuate more or less than 2db. The
recordings were then down sampled to computer. Default sampling mode is mono recording
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 353 -
(single channel) at 16kHz sample rate, giving a 3 dB bandwidth of 7.88kHz. File format is
ESPS, the proprietary file format for the Entropic ESPS/xwaves+ software package. Then
these files were converted to .wav files so that they could be opened in a PC. The data were
analyzed using PRAAT 4.0.2. © (Boersma & Weenink, 1992-2008) and were previously
converted to a 11kHz sample rate which is more appropriate for vowel quality analysis in
female speakers. In the following section we indicate how the analysis proceeded.
3.3 Acoustic measurements
Acoustic measurements were made using a temporal window at PRAAT which included the
oscilogram, the spectrogram and the formant tracts for the first five formants. The tonic and
the post-tonic vowel in each word were analyzed.
In order to know whether the preceding stressed vowel had any influence on the
quality of final schwa, both vowels were measured in terms of the first three formants: the
stressed vowel and the final unstressed vowel (schwa)1. In order to organize the data
systematically a 20ms part of the vowel center were selected and the software automatically
provided the mean for the selected portion of the vowel. The 20ms portion of the vowel was
preferred to a single selected point in order to have more comprehensive data of formant
values.
Duration of final schwa was measured selecting as the starting point the beginning of
the harmonic pattern, which typically characterizes vowels, and the last observed period in the
vowel which was at the end of a sentence, thus followed by silence.
4. Results
The results to be presented considered two analyses. The first one comprises an assessment of
descriptive statistics which considered mean values, standard deviation and dispersion
graphics for each set of vowels. The second analysis presents ANOVAs results which allowed
us to determine the reliability of observed differences.
In the following sections results will be presented separately for each language and
then a comparative analysis will be discussed. Results for vowel quality are presented first
and then results of duration are given. We will first present the data for Brazilian Portuguese
and then the English ones.
4.1 Brazilian Portuguese results
The chart below presents the results of the mean values (by-subjects) with respect to F1 and
F2. The two rightmost columns list the results for the schwa, which occurred in the final
position of the word. The two leftmost columns list the results for the full vowel which
appeared in the stressed position in the word which contained a (s w) pattern, where the
schwa occupied the final position.
1 Even if the third formant was taken into account so as to check the characteristic 1000Hz spacing between
formants, all the analysis is based on the first two formants which are the most appropriate to describe vowel
quality.
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 354 -
Table 2. Mean values, by-subjects, for Portuguese vowels
Full vowels schwa
[a] F1
[a] F2
[] F1 [] F2
935
1601
625
1615
[i] F1
[i] F2
[] F1 [] F2
491
2567
601
1709
[u] F1
[u] F2
[] F1 [] F2
461
858
583
1669
[] F1
[] F2
[] F1 [] F2
725
2336
604
1710
[] F1
[] F2
[] F1 [] F2
725
1071
624
1543
[e] F1
[e] F2
[] F1 [] F2
554
2513
606
1777
[o] F1
[o] F2
[] F1 [] F2
557
973
622
1570
Mean of all
schwas 609
1656
Standard
deviation 15,40254
84,04449
The data in Table 2 indicate that schwa F1 is slightly lower when preceded by a high
vowel [u,i] or when preceded by a mid-high vowel [e,o]. However, schwa F1 is lower when
preceded by the mid-low vowel [], than when the schwa is preceded by the mid-high vowels.
It is important to notice that in Brazilian Portuguese a final schwa is a bit more opened
(mean of all schwas: 609Hz) than expected in the literature (F1 550Hz for females). The
standard deviation is low, showing that there is little variability and, therefore, F1 for a schwa
has a target value which is well defined in this position.
F2 values for the schwa are slightly lower when preceded by a back vowel, i.e., when
preceded by a vowel with a low F2 such as [o,] and higher when preceded by a front vowel,
i.e., when preceded by a vowel with a high F2 such as [,e,i]. However, F2 values of a schwa
are higher after [u] than after [a], which shows that F2 of [u] has no influence on schwa.
The mean for F2 of all schwas is around what is expected (1656Hz), according to the
relevant literature. However, the standard deviation shows greater variability of schwa in
terms of F2 than of F1. Therefore, we may conclude that a final schwa is also specified for
F2. We have not identified any possible generalization which indicated that the full stressed
vowel influenced on the nature of the schwa in Brazilian Portuguese. We argue then that a
schwa does have a target of itself in BP.
Consider now Figure 1 that indicates the values for each realization of schwa by each
speaker.
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 355 -
Di sp er s io n of f i nal s chwa
0
100
200
300
400
500
600
700
800
050010001500200025003000
F2 i n H z
após [u]
após [i]
após [ó]
após [a]
após [é]
após [ê]
após [ô]
Figure 1. Dispersion of final schwa in Portuguese.
Figure 1 shows that there is greater variability of schwa in terms of F2 than in terms of
F1. The difference between the lowest and highest F1 values is about 300Hz while the
difference between the lowest and highest F2 values is about 1000Hz. Despite this variability
the greatest number of realizations is around the target values which corroborate our claim
that a schwa does have a target in BP.
The ANOVA analysis showed that one specific speaker, i.e., Inf.4 (R), influenced the
results both for F1 and F2. This may account for some of the variability observed in the
results. The tests also showed that only the preceding vowel [u] influenced schwa F1 and the
back vowels [o,] influenced schwa F2. However, the lack of systematic influence of the
preceding vowel on the quality of the reduced vowel points to the fact that final schwa has a
target. We will now present the duration means for schwa after each vowel for each speaker.
Consider data in Table 3.
Table 3. Duration mean values (by-subject) for schwa
Stressed
vowel INF. 1 (F)
INF. 2 (J)
INF. 3 (L)
INF. 4 (R)
(Mean 1)
(SD 1)
[a] 93
126
134
123
119
17,94436
[i] 98
127
150
128
125,75
21,32878
[u] 92
117
147
131
121,75
23,31487
[] 88
120
139
99
111,5
22,63478
[] 91
120
115
106
108
12,72792
[e] 104
152
162
141
139,75
25,3295
[o] 120
131
137
142
132,5
9,469248
(Mean 2) 98
127,5714
140,5714
124,2857
122,6071
11,19603
(SD 2) 11,03026
11,81605
14,7293
16,46931
In the first row in Table 3 (Mean 1) indicates the average duration of a schwa per
vowel, by subjects and (SD 1) corresponds to the standard deviation. In the leftmost column
(Mean 2) indicates the average duration of schwa per speaker, independently of the quality of
the preceding vowel and (SD 2) corresponds to the standard deviation.
When one compares the data for all four speakers, one observes that INF 1 has the
lowest mean, INF 3 the highest; and INF 2 and 4 have very similar results. The standard
deviation for the four speakers is not so different; therefore, we may conclude that the degree
of variability of schwa is not too high.
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 356 -
The same holds for the general mean of approximately 123ms (across vowels and
speakers). This result shows that a schwa may be longer than it has been suggested in the
literature, since Crosswhite (1999) claims that no schwa is longer than 40ms. In order to
verify whether the durational pattern, which showed to be longer than expected, was a
characteristic of any schwa in BP, we measured a number of other schwas which occurred in
the additional sentences of the experiment (not word/utterance final nor in the accented word
of the utterance).
These extra measurements of schwa indicated that other schwas in weak prosodic
positions, mainly in structural words, are very short having an average of 30ms. Our results
indicate that a schwa is longer in utterance final position (mean 123ms), than in other
prosodic ones (average of 30ms).
We have shown that a schwa in BP does have a target although variability may also be
observed. It is the lack of generalizations which indicates that the full stressed vowel bears no
influence on the nature of the schwa in Brazilian Portuguese. When considering durational
values, we observed the lengthening of the final syllable of an utterance with a schwa
confirms Cagliari & Abaurre’s (1986) claim for BP that final unstressed vowels are longer
than other vowels. Our results also challenge Crosswhite’s (1999) proposal that a schwa is
never longer than 40ms. We will now consider the data for British English.
4.2 British English results
The chart below presents the results of the mean values (by-subjects) with respect to F1 and
F2. The two rightmost columns list the results for the schwa, which occurred in the final
position of the word. The two leftmost columns list the results for the full vowel which
appeared in the stressed position in the word which contained a (s w) pattern, where the
schwa occupied the final position.
Table 4. Mean values, by-subjects, for English vowels
Full vowels schwa
[] F1
[] F2
[] F1 [] F2
674
1275
614
1516
[i] F1
[i] F2
[] F1 [] F2
369
2707
635
1729
[u] F1
[u]2 F2
[] F1 [] F2
363
2141
623
1599
[æ] F1
[æ] F2
[] F1 [] F2
932
1624
626
1600
[] F1
[] F2
[] F1 [] F2
568
1160
636
1495
[] F1
[] F2
[] F1 [] F2
450
2318
596
1637
[] F1
[] F2
[] F1 [] F2
452
878
603
1522
Mean of all schwas 619
1585
Standard deviation 15,38397
82,35059
2 Our results have been corroborated by Paul Boersma (personal communication): “I just looked into the 2000
version of Gimson’s book, edited by Alan Cruttenden. He explicitly states that /u/ has been fronted and /æ/ has
fallen during the last 30 years (pages 83, 99). The formants that he shows are quite close to the ones that you
found”.
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 357 -
The data in Table 4 indicate that F1 does not seem to be affected by the quality of the
preceding vowel in English. This is because no correlation could be found between the
preceding vowel and the variability of schwa in our data. In fact, as the standard deviation
shows, there is little variability with a schwa. Therefore, we have to assume that F1 value for
a schwa is well defined in this position, which suggests that final schwa in English does have
a target.
In fact Jones (1976) mentions that final schwas in English, when final in the utterance,
are opener (and longer) than other schwas. This can be confirmed by the mean of all schwas
in Table 4 which is 619Hz. With regards to F2 one observes that the values are slightly lower
when preceded by a back vowel, i.e., when preceded by a vowel with a low F2 such as
[,,], and F2 values are higher when preceded by a front vowel, i.e., when preceded by a
vowel with a high F2 such as [i,,y3,æ]. The mean for F2 of all schwas is a bit below what was
expected (1585Hz), however, that mean shows that final schwa is also specified for F2. This
leads to suggest that a schwa has a target in English. Consider now Figure 2 that indicates the
values for each realization of schwa by each speaker.
Dispersion of final schwa
0
100
200
300
400
500
600
700
800
050010001500200025003000
F2 in Hz
F1 in Hz
após [u:]
após [i:]
após [O]
após [ae]
após [a:]
após [o:]
após [I]
Figure 2. Dispersion of final schwa in English
Figure 2 shows that there is greater variability of schwa in terms of F2 than in terms of
F1. The difference between the lowest and highest F1 values is about 300Hz while the
difference between the lowest and highest F2 values is about 700Hz. Despite this variability
the greatest number of realizations is around the target values.
The ANOVA analysis showed that one specific speaker, i.e., Inf. 3 (R), influenced the
results for F1, but no influence of any preceding vowel was verified. F2 was influenced by all
speakers due to the fact their production was heterogeneous. The results also showed that the
preceding front vowels [æ,,i,y] influenced schwa F2. However, the lack of systematic
influence of the preceding vowel on the quality of the reduced vowel, i.e., a schwa, points to
the fact that final schwa does have a target in English.
We will now present the duration means for schwa after each vowel for each speaker.
Consider data in Table 5.
3 This symbol is used here to better represent the phonetic quality of /u/, i.e., a high front rounded vowel.
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 358 -
Table 5. Duration mean values (by-subject) for schwa
Stressed
vowel INF. 1 (F)
INF. 2 (J)
INF. 3 (L)
INF. 4 (R)
(Mean 1)
(SD 1)
[] 169
187
142
157
163,75
19,03287
[i] 172
191
156
170
172,25
14,3846
[u] 182
182
153
160
169,25
14,99722
[æ] 143
170
146
134
148,25
15,37043
[] 146
159
150
148
150,75
5,737305
[] 166
187
145
166
166
17,14643
[] 189
175
151
174
172,25
15,73478
Mean (2) 166,7143
178,7143
149
158,4286
163,2143
10,9253
SD (2) 17,10472
11,38294
4,898979
13,80649
In the first row in table 5 (Mean 1) indicates the average duration of a schwa per
vowel, by subjects and (SD 1) corresponds to the standard deviation. In the leftmost column
(Mean 2) indicates the average duration of schwa per speaker, independently of the quality of
the preceding vowel and (SD 2) corresponds to the standard deviation.
When one compares the data for all four speakers, one observes that INF 3 has the
lowest mean and INF 2 the highest. The means of the four speakers are heterogeneous;
however, the standard deviations are low. INF 3 presents the lowest SD across vowels. When
considering all the vowels, the most homogeneous results are found for vowel [] with a SD
of 5,737305 by speakers. The general mean is 163ms (across vowels and speakers). In order
to know whether this long duration was a constant characteristic of these speakers or a more
general result concerning reduced vowels, we measured other schwas which occurred in the
additional sentences of the experiment (not word/utterance final nor in the accented word of
the utterance). These extra measurements showed that other schwas in weak prosodic
positions, mainly in structural words are very short indeed (about 30ms.). In a similar fashion
as the results obtained for Brazilian Portuguese our results for British English contradict
Crosswhite’s proposal that a schwa is never longer than 40ms when it occurs at the end of an
utterance.
We will now present a general discussion for both languages with the aim of
answering our research questions: a) what are the major quality and quantity characteristics of
a final schwa in BE and BP?, b) do adjacent vowels affect the nature of a final schwa? and c)
does Crosswhite’s (1999) claim that a schwa is never longer than 40ms hold?
5. General discussion and conclusion
In this section we will contrast our results for Brazilian Portuguese and British English. Table
6 shows the major characteristics of vowel quality of a schwa.
Table 6. Vowel quality of final schwa
Means of F1 of all schwas Means of F2 of all schwas
BP: 609Hz (sd 15.40254) BP: 1656Hz (sd 84.04449)
RP: 619Hz (sd 15.38397) RP: 1585Hz (sd 82.35059)
In terms of vowel quality, the mean of all schwas for F1 is very similar in both
languages (leftmost column). The mean of all schwas for F2 is just slightly different in each
language. General results indicate that a Portuguese final schwa is more anterior than the
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 359 -
English one. This indicates that schwas are very similar in both languages although it appears
that it is fine phonetic detail which characterizes the specificity of a schwa in each language
(Pierrehumbert, 1999).
Notice also that the difference between the lowest and highest F1 values is about
300Hz in both languages while the difference between the lowest and highest F2 values is
about 700Hz in English and 1000Hz in Portuguese. Therefore, Portuguese exhibits greater
variability in the horizontal dimension. Some unsystematic influence of the preceding vowel,
especially in terms of F2, was observed which did not allow us to figure out any
generalization. Thus, despite the observed variability our results indicate that a schwa has a
target in both languages. This suggests that an adjacent vowel does not affect the nature of a
final schwa in either language contradicting Browman and Goldstein’s (1992) proposal.
An evaluation of durational patterns shows a great influence of the prosodic
environment on the length of final schwa in both languages: general mean in BP 123ms and
163ms in BE. Therefore, the fact that the analyzed words where accented and final in the
utterance was crucial for vowel quantity: a longer than expected schwa was attested. These
results confirm Turk & Sawusch’s (1997) findings in English that the accentual lengthening
affects both the primary stressed vowel and also the following vowel (unstressed). We
observe that the same holds for Portuguese offering thus an open invitation to research to be
carried out in other languages, in order to understand better the relationship between
durational values and the effects of accent.
Finally, durational patterns questions Crosswhite’s (1999) claim that a schwa is never
longer than 40ms. At least for final utterance position a schwa may be much longer than she
has predicted.
Summing up, we have answered the proposed questions of this paper. Regarding the
quality and quantity characteristics of a final schwa in BE and BP a contrastive analysis
showed that in both languages a schwa presents similar acoustic characteristics both in terms
of quality and durational patterns in utterance final position. However, a schwa has its own
target in each language. We suggest that it is fine phonetic detail that plays an important role
in characterizing formant values of a schwa in either language (Pierrehumbert, 1999).
Concerning the influence of adjacent vowels in the nature of a schwa our results show
that it does not hold for utterance final schwas. Thus, Browman and Goldstein (1992) claim
that a schwa target is the mean of all English vowels does not hold for this prosodic
environment. Thus, the phonetic characteristics of a schwa are similar either as the reduced
variant of different vowels - as in English -, or as the reduced variant of a single phoneme - as
in Brazilian Portuguese.
Our results also support Bates (1995) proposal that final utterance schwas are different
from medial schwas in English. It might be the case that final schwas have an independent
target whereas medial schwas may undergo interpolation. Further research on this issue is still
necessary.
Finally, durational patterns questioned Crosswhite’s (1999) claim that a schwa is never
longer than 40ms. At least for utterance final position a schwa may be much longer than she
has predicted.
The main conclusion to be drawn in this paper is that the investigation of acoustic
properties of segments must take into consideration the prosodic patterns involved. Thus,
prosodic properties of reduced vowels may be similar in different languages, although fine
phonetic detail may contribute to the phonological interpretation of these vowels.
_____________________
New Sounds 2007: Proceedings of the Fifth International Symposium on the Acquisition of Second Language Speech
- 360 -
Acknowledgements
Adriana Marusso would like to acknowledge support from CAPES (Committee for
Postgraduate Courses in Higher Education) through PICDT and PDEE grants. Thaïs
Cristófaro Silva would like to acknowledge support from CNPq (National Council for
Scientific and Technological Development), grant number 30.33.97/2005-5.
References
Bates, S. (1995). Towards a definition of schwa: An acoustic investigation of vowel reduction in
English. Unpublished Ph.D. dissertation, University of Edinburgh.
Boersma, P., & Weenink, D. (1992-2008). Praat 4.0.2. A system for doing phonetics by computer.
Retrieved from www.praat.org.
Browman, C., & Golstein. L. (1992). Targetless Schwa: An articulatory analysis. In G. Docherty, & D.
R. Ladd (Eds.), Papers in laboratory phonology II: Gesture, segment, prosody (pp. 26-67).
Cambridge: Univ. Press.
Cagliari, L., & Abaurre, M. (1986). Elementos para uma investigação instrumental das relações entre
padrões rítmicos e processos fonológicos no português brasileiro. Cadernos de Estudos
Lingüísticos, 10, 39-57.
Crosswhite, K. (1999). Vowel reduction in Optimality Theory. Ph.D. Dissertation, UCLA.
Jones, D. (1976). An Outline of English Phonetics. 9th ed. Cambridge: University Press.
Keating, P. (1988) Underspecification in phonetics. In Working papers in Phonetics 69, University of
California Linguistics Club.
Lindblom, B. (1963). Spectrographic study of vowel reduction. Journal of the Acoustical Society of
America,, 35(11), 1773-1781.
Marusso, A. (2003). Redução vocálica: Estudo de caso no português brasileiro e no inglês britânico.
Unpublished doctoral dissertation, Universidade Federal de Minas Gerais, Belo Horizonte, Brazil.
Massini-Cagliari, G. (1992). Acento e ritmo. Ed. Contexto, São Paulo.
Pierrehumbert, J. (1999). What people know about sounds of language. Studies in the Linguistic
Sciences, 29(2), 111-120.
Recasens, D. (1986). An acoustic analysis of V-to-C and V-to-V coarticulatory effects on Catalan and
Spanish VCV sequences. In: Status Report on Speech Research, Sr-86/87, Haskins Laboratories.
Recasens, D. (1991). Fonètica descriptiva del Català. Institut d’Estudis Catalans: Barcelona.
Stevens, K., & House, A. (1963) Perturbation of vowel articulations by consonantal context: An
acoustic study. Journal of Speech and Hearing Research, 6, 111-128.
Turk, A., & Sawusch, J. (1997). The domain of accentual lengthening in American English. Journal of
Phonetics, 25, 25-41.
... Similar to English, word stress in focus position in Brazilian Portuguese corresponds to the most prominent syllable, which does not suffer vowel reduction (Massini-Cagliari, 1992;Cristófaro-Silva & Marusso, 2008). Vowel reduction in unstressed syllables of BP is characterized by so little F1 lowering and so little F2 backness that these do not dislocate unstressed vowels from the full vowel category poles in the acoustic space (Nobre & Ingemann, 1986). ...
Thesis
Full-text available
Although word stress is an inherent property of each word, it is not yet well understood how word stress is encoded in the representation of a word, or how and when it is retrieved during word processing. In this thesis, I considered the processes of L1 and L2 spoken word recognition, word production, and printed word naming, to assess the role that word stress plays in each of these. In this context, I reviewed several psycholinguistic word retrieval models, pointing out that these often do not specify the conditional relationships between syllables, syllabic structure, syllable position, and word stress. This literature review sets the stage for a sketch of this thesis’ experimental studies on word stress in bilingual and monolingual word retrieval.
Article
Full-text available
Os fenômenos da aquisição fonético-fonológicos específicos a homens e mulheres são, relativamente, pouco explorados. Os estudos fazem, frequentemente, médias com os dados de produção fonética de homens e mulheres, com o fim de neutralizar o efeito das diferenças de sexo. Este estudo pretende investigar a aquisição da duração vocálica do inglês americano como produzida por falantes mulheres e homens falantes nativos de português brasileiro. Os resultados mostraram que as mulheres produzem vogais mais longas do que os homens tanto no grupo controle monolíngue quanto no grupo experimental bilíngue. As análises estatísticas indicam que os efeitos relativos às diferenças entre sexo e os efeitos relativos à aquisição de L2 não estão correlacionados. Tais resultados são interessantes para argumentamos em favor de que estudos em aquisição de L2 e fonética devem sempre considerar as diferenças de aquisição e performance entre os sexos, a fim de que importantes fenômenos específicos ao sexo não sejam ignorados.
Article
Full-text available
Segmental as well as prosodic features have been the focus of much of the studies on second language phonology. Still most of the studies on non-native speech are found to be focusing heavily on vowels. This is because vowels form the most difficult part of perception and production of non-native speech. Due to difficulties faced while pronouncing some vowels, the L2 speakers may use native sounds in the place of difficult L2 sounds. This pattern is followed in the speech of Assamese speakers of English. While both Assamese and English have front and back vowels, the presence of central vowels in English makes the English vowel inventory quite different from Assamese. In addition to this, the presence of centering diphthongs pose a different level of problem for Assamese speakers pronouncing words involving centering diphthongs. Replacement and simplification are the two strategies that Assamese speakers of English employ to tackle this problem. The results of this production experiment reveal that the English central vowel [ə] is replaced by the Assamese vowels [e], [o], [u] and [a]. The process of substitution of English [f:] by Assamese [a] is also found to be a ubiquitous feature. The centering diphthongs in English are realized sometimes as monophthongs and sometimes as a sequence of two vowels in Assamese speakers' pronunciation of English. This study tries to make a detailed analysis of this kind of replacements based on acoustic comparisons between English central vowels and centering diphthongs, and their realizations in Assamese speakers' English speech.
Article
Full-text available
V-to-C and V-to-V coarticulatory effects in F 2 frequency are studied for Catalan and Spanish VCV sequences with vowels and consonants involving different degrees of articulatory constraint on tongue-dorsum activity. The findings reported in this paper indicate that coarticulatory effects decrease with the degree of articulatory constraint, for the following groups of consonants and vowels: DJ>[iJ; [.r]>[rJ; [8], [oJ>[yJ; [a]>[iJ. Differences in anticipatory vs. carryover coarticulation were also found to be strongly dependent on the degree of articulatory constraint associated with the intervening consonants and vowels. Overall, results suggest that coarticulatory effects are deeply related to the control mechanisms involved in the production of articulatory gestures.
Article
Full-text available
It is often assumed in work on the phonological underspecification of segments that while representations may at first be underspecified, they end up fully specified. Various kinds of rules are posited to ensure that in output forms all features have values for all segments. In this paper I will consider an alternative view: that underspecification may persist into phonetic representations. I begin by reviewing some of the relevant phonological phenomena and mechanisms, and some of the history of underspecification in phonetic studies. I then show how phonetic data may be taken to reflect the presence or absence of feature values in surface forms. Finally, the extent to which surface specification depends on segmental contrasts is examined. The aim of this paper is to present some phonetic phenomena that are potentially relevant to any theory of underspecification.
Article
Full-text available
Measurements of formant frequencies and duration are reported for 8 Swedish vowels uttered by a male talker in three consonantal environments under varying timing conditions. An exponential function is used to describe the extent to which formant frequencies in the vowels reach their target values as a function of vowel‐segment duration. A target is specified by the asymptotic values of the first two formant frequencies of the vowel and is independent of consonantal context and duration. It is thus an invariant attribute of the vowel. The results suggest an interpretation in terms of a simple dynamic model of vowel articulation.
Thesis
This thesis reports a single speaker acoustic study of vowel variability in connected speech. Over eight thousand vowel tokens taken from a corpus of read sentences are examined. The aim of the thesis is to achieve a better understanding of the nature of vowel reduction in English. Three questions are addressed. The first of these concerns the phonetic characterisation of schwa, the central or 'reduced' vowel. Schwa's contextual variability is assessed with reference to the question of whether or not it has an independent phonetic target. The second question concerns the role of stress in conditioning vowel reduction. Patterns of variability for sententially stressed and unstressed vowel tokens are examined in order to determine how far stress and context effects interact to influence vowel quality. The final question concerns the potential ditf erences between vowels with respect to inherent variability, that is, whether some vowels are inherently more susceptible to coarticulatory effects than other vowels. Maximal context-dependency for schwa strongly supports the hypothesis that it is completely unspecified for tongue position. The data indicate that it is also highly unspecified for jaw position. Evidence that schwa is targetless and can occupy almost any position in the vowel space depending on context, argues against the traditional concept of vowel reduction as an independent process of articulatory and/or acoustic centralisation. Greater context sensitivity for sententially unstressed vowels compared with their sententially stressed counterparts also supports an account of vowel reduction in terms of contextual assimilation. The results also indicate a continuum of underspecification. This ranges from-the more peripheral vowels /i, a, ɑ, ɒ, Ɔ/ which show the least contextual variability and which may be thought of as the most narrowly specified vowels to the more central vowels /I, ε , з, ᴧ, Ʊ/ and, in the present data, /u/, which show greater overall context dependency. It is proposed that greater acoustic stability for the more peripheral vowels reflects quantal acoustic properties and tighter articulatory and/or perceptual constraints on variability.Overall, the results support the view that vowel reduction represents a means of economising on articulatory effort. Schwa, the endpoint of the reduction process represents minimal articulatory effort insofar as it represents the straight-line interpolation between consonants and hence minimal resistance to coarticulatory effects. Shorter durations, greater context dependency and, in the case of the peripheral vowels, less extreme formant values for sententially unstressed compared with sententially stressed vowels reflects a reduction in articulatory effort and consequently less displacement from neutral. In view of the greater contextdependency observed for the more central vowels generally compared with the more peripheral vowels, the tense/lax alternation in phonological vowel reduction can also be interpreted as a saving on articulatory effort. A principal advantage of an account of English vowel reduction in terms of phonetic underspecification is that phonetic and phonological vowel reduction may be accounted for by the same mechanism.
Article
The experiments presented in this paper were designed to determine which linguistic unit, if any, serves as the domain of the durational effects of accent. Results from both experiments suggest that the durational effects of accent are bounded by a linguistic unit. Results from Experiment 1 suggest that the durational effects of accent extend beyond the syllable nucleus or rime. Results from Experiment 2 show that in disyllabic words, word-final, but not word-initial, unstressed syllables are lengthened under accent. These results suggest that the domain of accentual lengthening begins with an accented syllable and can include at least one following unstressed syllable within a word.
Article
Thesis (Ph. D.)--University of California, Los Angeles, 1999. Vita. Includes bibliographical references (leaves 434-447).