Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
1. Introduction
The paper reports the results of a corpora-based study of vowel reduction in English and Polish
across two speech styles, read and fully spontaneous. The study addresses the issue of vowel
reduction which, following the centralization theory (Koopmans-van Beinum 1980), consists in
diminishing the vowel’s distinctiveness, i.e. its quality and quantity. Given the shape of the vocal
tract, the process of vowel reduction modifies a vowel by pulling or dragging it to the most
centralized point of the vowel space. As Carr (2008) notes, “in many languages, unstressed
vowels are often reduced to schwa, which involves no deviation from the neutral position of the
tongue. A word such as personal has reduced vowels in its unstressed syllables, both of them
schwas: [ pʰɜːsənəl], but in the word personality, the syllable with primary stress has a non-
Numerous proposals have been made to identify and explain the factors triggering
reduction. Lindblom (1963), for instance, views vowel reduction as a result of undershoot, or
failure to reach a more peripheral tongue position in the vowel space due to time-related pressure:
“undershoot in the formant frequencies relative to the bull's-eye’ formant pattern” (Lindblom
1963: 1779). Others claim that vowel reduction is a function of stress (Crosswhite 2003; Barnes
2006), whereas Harris (2005) understands vowel reduction as information loss: ”[r]eduction
follows two apparently contradictory routes in vowel space, yielding either centralised values (the
‘centripetal’ pattern) or the corner values a, i, u (the ‘centrifugal’ pattern). What unifies these
vowels is the relative simplicity of their acoustic spectra compared to those of mid peripheral
1
vowels. Spectral complexity can be taken as one measure of the amount of phonetic information
present in a speech signal at a given time. On this basis, centripetal and centrifugal reduction can
both be construed as resulting in a loss of phonetic information” (Harris 2005: 119). Analyzing
vowel reduction in the setting of casual speech, Bybee (2009) claims that “vowels move closer to
the neutral position and schwas grow shorter and shorter until they are simply skipped […] as
changes in the timing of the articulatory gestures” (Bybee 2009: 30). In addition, a vowel
undergoes reduction if it is in a second or subsequent token of a word (Fowler and Housum 1987,
Baker and Bradlow 2009). Regardless of the theoretical stance, typically, lax vowels in
prosodically weak positions are subject to reduction (Cruttenden 2014). In a similar vein,
consonants also exhibit a parallel process of reduction, although on a smaller scale: lenition of
voiceless stops in intervocalic position may serve as an example here (Watson 2006).
The very term reduction deserves additional explanation in the context of vowels as its
effect may be partial (reduction of a lax sound to a more centralized one) or complete (schwa
elision). By the same token, a need arises to draw a distinction between phonetic and
phonological reduction. The former stands for obliteration of lexical distinctions that results from
neutralisation of phonetic contrast between two or more vowels (Crosswhite 2003; Barnes 2006).
of speaking style or tempo and can be represented by schwas in the dictionary entries. The latter,
by contrast, is a universal, cross-linguistic and naturally occurring process (Farnetani and Busa
1999; Vayra et al. 1999; Nikolaidis 2003; Flemming 2005; Jaworski 2007). According to
schwa when they occur in connected speech” (Lindblom 1963: 1781). Phonetic reduction may be
then taken as any approximation of vowels towards schwa vowel that ensues from an increased
speaking rate or decreased articulatory effort, e.g. vowel reduction present in grammatical words.
2
This is the sense in which the term vowel reduction is used in the present paper, as a general
tendency to centralize vowels and an approximation to schwa, rather than its full realization.
The study reported here takes into consideration vowel reduction across speech styles.
Speech style, in turn, may be understood as “the variation that occurs in the speech of a single
speaker in different situational contexts” (Cheshire 1992: 324). The variation, evoked in the
definition, is governed by a number of factors such as audience design (Bell 1984; Bell 2001) or
the amount of attention paid by the speaker (Labov 1994). Thus, a distinction can be made
between formal vs. informal speech style, where the level of attention increases with the level of
(Ernestus and Warner 2011); due to a specific design of one of the corpora used in the study (cf.
section 2), however, having a recourse to formal speech for style comparison was not possible.
Instead, the study compares read to fully spontaneous speech style, the former serving as a proxy
for formal style. In fact, a number of languages have been reported to vary across the
read/spontaneous distinction in the vowel space. Much of the published literature on vowel
reduction has found that both duration and formants are reduced (shortened and centralized) to a
Table 1: Comparison of speech styles across languages (adapted from Spilkova 2014).
3
Bondarko et Russian spontaneous speech (dialogues vs greater variability of formant values
al. (2003) read speech (read version of the for peripheral vowels /a/, /i/ and /u/
spontaneous speech transcript) in spontaneous speech
Moon and English citation forms (i.e. normal less formant displacement due to
Lindblom reading) vs. clear speech context in clear speech
(1994)
With regard to languages examined in the present study, two typologically unrelated
languages were selected: English and Polish. Apart from different rhythmic properties, vowel
inventories differ considerably between Polish and English, in both quality and quantity:
A comparison of Figures 1 and 2 reveals that all Polish vowels are of equal length and that
Polish, unlike English, has no schwa and the mid area of the vowel space is unpopulated with the
only exception of the /ɨ/ sound. As Sobkowiak puts it, “Polish vowels are never reduced so
4
thoroughly to a mid-central quality of schwa” (Sobkowiak 2008: 132). He also observes that
“most English vowels come between Polish vowels located nearby” (Sobkowiak 2008: 130). On
the other hand, Nowak (2006) has found that “the “soft” consonants have a very strong impact on
Polish vowels in both the F1 domain and the F2 domain (although the impact of many other
consonants is also non-negligible) and that this impact wanes as the duration of the vowel
model to the variability of 373 Polish vowels and we believe that it is appropriate to describe this
Thus, the present paper attempts to verify the claim that phonetic vowel reduction is
universal (Farnetani and Busa 1999; Vayra et al. 1999; Nikolaidis 2003; Flemming 2005) in two
typologically unrelated languages and across two speech styles of various degree of formality. In
this paper, two specific research questions are addressed. First, the study investigates the extent to
which vowels are reduced in spontaneous speech in comparison with read speech in Polish and
English. The second research question pertains to the role of speech rate in vowel reduction, as is
assumed in previous scholarship (Jurafsky et al. 1998, Shockey 2003). It is then hypothesized that
(1) vowels exhibit stronger reduction in fully spontaneous speech in comparison with read speech
in the two languages (2) vowel reduction in spontaneous speech is more robust in English and
considerably less so in Polish due to typological differences (3) speakers with high speech rate
reduce vowels to a greater extent than slow speakers. Therefore, in addition to providing more
data on differences between read and spontaneous styles for vowel reduction in Polish and
English, the results might inform the debate on the role of speech rate in reduction of vowels.
5
2. The methodology of the study
The data (both read and spontaneous) come from two corpora: Phonologie de l’Anglais
The Phonology of Contemporary English: usage, varieties, and structure is a project gathering a
series of corpora whose purpose is “to attain an effective and factual description and comparison
of various accents of English” (Durand and Pukli 2004: 2). Based on a common protocol and
2014, the corpus was composed of 31 surveys around the English-speaking world, 289 interviews
For the purpose of the present study, the Lancashire corpus was used which comprises
recordings of 9 female speakers, aged 23-83. The corpus contains 4 sets of data: formal
interview, informal interview, a text passage, the word list. The word list includes 127 lexical
items representing English vowels and 65 lexical items representing English consonants. The data
selected for the present study come from the word list as well as the informal interview (cf.
The Greater Poland Speech Corpus is collected within the project Internetowy Korpus
(The Internet corpus of contemporary Polish standard and vernacular spoken in the area of
Greater Poland). It must be stressed that the current project, though different from PAC in many
6
The Corpus is collected with the view to registering spontaneous speech data, as opposed
from the province of Greater Poland. The corpus collecting procedure consists of three stages: 1)
The high-quality of the recording is ensured by using a professional Roland R-26 recorder
and the lapel lavalier Rode microphones. The files are saved as a WAV format. The subjects are
recorded in a quiet room during the spontaneous conversation session and in a sound-proof room
In order to ensure the highest degree of informality, the speakers are recorded in a 2+2
interview format, i.e. two interviewees who know each other hold a conversation with two
interviewers. The interview lasts for about 40 minutes and covers a range of topics such as
studies and student life in Poznań, living, culture and entertainment in Poznań, as well as the
Internet use. The spontaneous speech session is followed by a reading task. In this task, subjects
are asked to read a set of 182 carrier phrases containing the test words. The sentences are
Upon the completion of the project the electronic version of the corpus (transcripts and
The aim and hypotheses. The aim of the study was to analyse the process of vowel reduction in
two typologically different languages, namely English and Polish, in two styles, reading vs
7
spontaneous speech. Two criteria of vowel reduction were measured: vowel duration and the
Hypothesis 1: Vowels exhibit stronger reduction in fully spontaneous speech in comparison with
Hypothesis 2: Vowel reduction in spontaneous speech is more robust in English and considerably
Hypothesis 3: Speakers with high speech rate reduce vowels to a greater extent than slow
speakers.
The subjects. The subjects of the study were nine female speakers of Lancashire English and nine
(four male and five female) speakers of Polish. The English subjects were aged 23-83. The
speakers of Polish were aged 20-22 and came from the area of Greater Poland.
The material and procedure. The hypotheses were tested on the basis of three vowels: English
KIT, TRAP and FOOT and Polish /ɨ/, /a/ and /u/ as in byty ‘entities’, baty ‘whips’ and buty
‘shoes’. For the study of English, the authors investigated 1 token of each vowel from the word
list and 5 words per vowel extracted mostly from the informal interview (occasionally from the
informal interview). In Polish, 3 tokens of each vowel from the word list (i.e. carrier phrase) as
well as 3 words per each vowel per speaker from the interview were examined. Altogether in
English, the authors obtained 27 tokens from scripted speech and 135 tokens from spontaneous
speech. For Polish, 81 tokens from scripted speech and the same number from spontaneous
As regards the selection of words for analysis, the following criteria were considered: in
grammar words such as could were avoided; in Polish (relatively) high-frequency (mostly) two-
syllable words were selected where the focus vowel appeared in a stressed position. Another
8
important criterion in the word selection process was the phonetic environment of vowels. In
order to facilitate the process of acoustic analysis, the selected vowels occurred in the direct
neighbourhood of obstruents. The table below presents all lexical items whose vowels were
analysed acoustically.
Upon extracting the words from the transcripts, the vowels underwent manual annotation in Praat
(Boersma and Weenink 2014).
Speech rate. Speech rate was operationalized as articulatory rate, i.e. without hesitation pauses
and filled pauses and expressed as syllables per second. Local rate was not considered in this
9
article on purpose on the grounds that the study, among other things, seeks to verify whether the
fastest speakers exhibit the highest reduction degree, rather than intending to perform an in-depth
correlation of speech rate with a variety of factors where such a fine-grained method of
calculating speech rate would be more appropriate (cf. section 4). For assessing the connection
between vowel reduction and speech rate, Pearson correlation was used (hypothesis 3). Size
effects for comparison of reduction across two languages (hypothesis 2) were given with the aid
3. Results
In order to tease apart vowel reduction from speech style and cross-linguistic differences, the
results are arranged according to the temporal and spectral domains of vowel production.
It was hypothesized that vowels exhibit stronger reduction in fully spontaneous speech in
comparison with read speech in the two languages. The comparison of the means, obtained in the
current study revealed that there were statistically significant differences for both Polish (read
M=84.88, SD=11.7, spontaneous M=49.01, SD=13.89) and English (read M=131.68, SD=13.17,
spontaneous M=76.54, SD=18.52) between read and spontaneous speech with regard to duration
of vowels. As for the Student t-test, there were 162 vowel tokens per each language, t (160) =
1.975, the p value (one-tailed) of the obtained data is 0.04, indicating a satisfactory degree of
statistical reliability.
10
131,6833333
140
120
76,54740741
100
[miliseconds] 80
60
79,56111111
40
20 50,45153325 English
0
Polish
read
spontaneous
According to Figure 4, the ratio of means of vowel duration in spontaneous speech was 58 per
cent in English and 63 per cent in Polish, relative to read speech (p<0.05). This is a considerable
comparison with the read speech style. For distribution of reduction across the KIT, TRAP and
FOOT vowels in English and /ɨ/, /a/ and /u/ in Polish, the analysis indicated that there was a
certain variability among them and between the two languages considered:
11
160,00 146,25
140,00 120,59 128,21
120,00
[miliseconds]
100,00
80,00
95,95
60,00
92,94 74,66
40,00 59,04
67,77
20,00 77,97
0,00 38,70
ɨ/ɪ 65,91
a/æ
read u/ʊ
read 46,74
read English
ɨ/ɪ
Polish
spont a/æ
spont u/ʊ
spont
100%
90%
75%
80% 71%
70% 60%
57% ɪ/ɨ
60% 49% 51%
50% æ/a
40% ʊ/u
30%
20%
10%
0%
spont to read English spont to read Polish
As it emerges from Figures 5 and 6, the three vowels, considered in the study, did not display
similar patterns for temporal reduction. Thus, in Polish, the low central vowel /a/ was reduced to
12
the lowest degree (as its realization in spontaneous speech was 71 per cent of its realization in
read speech), whereas the FOOT vowel was the most reduction-resistant in English (p<0.05).
These differences stem from different arrangement of vowels and density of population of the
Turning to the spectral aspect of reduction, the outcomes for differences between read and
spontaneous speech are presented below. Note that English female speakers were compared to
Polish female speakers, whereas no English male speakers from the PAC corpus were available
F2 [Hz]
3000 2500 2000 1500 1000 500 0
0
100
200
300 read eng
F1[Hz]
400 spont eng
ɪ ʊ
500
600
700
800
æ 900
13
F2 [Hz]
2500 2000 1500 1000 500 0
0
100
200
300 read pl
ɨ u
F1 [Hz]
400 spont pl
500
600
a
700
800
900
ɪ æ ʊ
F1 F2 F1 F2 F1 F2
read eng 554.7754 2451.198 820.21 1500.19 536.0963 1044.226
spont eng 439.3162 2035.824 775.52 1618.92 488.5739 1312.172
SD 81.64195 293.7137 31.59855 83.96058 33.6034 189.4659
ɨ a u
F1 F2 F1 F2 F1 F2
read pl 493.3909 1967.504 771.65 1461.09 438.0125 1157.88
spont pl 333.764 1548.953 590.70 1591.08 262.5527 989.158
SD 112.8733 295.9606 127.9543 91.9207 124.0688 119.3045
The analysis of differences between vowel space in read and spontaneous speech styles confirms
the findings for duration and provides support for the first hypothesis, stipulating that vowels in
14
3.2. Results for the second hypothesis
The second hypothesis put forward in this study is related to the typologically-based assumption
that overall, vowel reduction is robust in English and considerably less so in Polish due to
differences in the vowel inventories. In order to provide further insight into the relative
differences between reduction in the two languages, a between-group differences for duration and
formants was calculated by means of a one-way Anova. For these calculations, means of the three
vowels were considered, following the idea of centralization (see section 1) and comparing the
shrinking of the vowel space in spontaneous speech since it was subject to reduction, relative to
read speech (Figures 5-8). The number of analyzed vowel tokens for spontaneous speech was 216
As far as duration of vowel is concerned, the difference between spontaneous Polish and English
languages for F1 was found: F1, F (1,216) = 51.669, p = 0.000. The comparison of F2 between
spontaneous Polish and read English was also significant ((F (1,216) = 42.636, p = 0.000).
To sum up, as hypothesized, a one-way Anova revealed a main effect of reduction, both
in its temporal and spectral aspects, suggesting that English vowels in spontaneous speech were
15
3.3. Results for the third hypothesis
The third hypothesis verified the role of speech rate in vowel reduction. In particular, it stipulated
that speakers with high speech rate reduce more than slow speakers. Below, articulatory rates for
It must be observed that rates for Polish are significantly higher than for English. This can be
accounted for in terms of age differences: Polish speaker’s age ranged from 20 to 22, whereas the
age for English speakers ranged from 23 to 83. Indeed, previous scholarship has found that
younger speakers tend to speak faster (e.g. Verhoeven et al. 2004; Raymond et al. 2006; Jacewicz
et al. 2009).
As for correlation between reduction and articulatory rate, it was run separately for
16
100%
90%
80%
reduction (duration)
70%
60%
50%
English
40%
Polish
30%
20%
10%
0%
0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00
Rate [sps]
Surprisingly, for English, the correlation was negative and non-existent (r=-0.06). This implies
that fastest speakers did not necessarily reduce their duration of vowels to the highest degree. By
contrast, certain correlation, albeit feeble, was found in Polish (r=0.58), providing support for
hypothesis three.
Figure 10 attested to the lack of correlation between articulatory rate and F1 which runs
counter to the hypothesis (r=0.14 for English and 0.02 for Polish). It appears that with respect to
17
100%
90%
80%
70%
reduction (F1) 60%
50%
English
40%
Polish
30%
20%
10%
0%
0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00
Rate [sps]
Analysis of the link between F2 and articulatory rate reveals two opposing trends: a weak
correlation was found in Polish (r=0.41), whereas in English, no such correlation was established
(r=-0.27). Furthermore, the correlation found for English was negative which stands in stark
100%
90%
80%
70%
reduction (F2)
60%
50%
English
40%
Polish
30%
20%
10%
0%
0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00
Rate [sps]
18
4. Discussion
In light of the obtained results, the two first hypotheses were positively verified, whereas the third
one produced mixed results. The study establishes a significant difference in vowel reduction
across two speech styles, read and fully spontaneous across two unrelated languages (Figures 5, 7
and 8). This is highly consistent with the results, reported in Table 1. All vowel tokens, with the
notable exception of certain tokens of the /u/ vowel in Polish, were shorter and centralized in
spontaneous speech, relative to their duration as well as placed in less peripheral positions than in
read speech. These tendencies were observed in Polish and English alike, with the exception of
the extent of reduction process. It has been shown that reduction in English is considerably
stronger than in Polish (Table 4) which is in agreement with certain aspects of Polish phonology
(lack of schwa, Jassem 2003). The very fact of undergoing the process of centralization in Polish,
observed in the current study, is compatible with the findings of Nowak (2006). Thus, the
analysis of 324 vowel tokens from two corpora attests to the effects of speech style on vowel
reduction and provides support to the claim that phonetic reduction, found in the study, is present
speech rate and reduction, the findings of the current study did not provide a definite answer. To
a certain extent, the correlation between rate and duration was found in Polish (r=0.58) but not in
English (r=-0.06). This outcome may be partly explained by the fact that the population of Polish
subjects was very homogenous age-wise. In comparison, English speakers were significantly
more diversified as an age group. Considering previous studies on age and rate which point to
higher speech rate among younger speakers (Verhoeven et al. 2004, Raymond et al. 2006,
Jacewicz et al. 2009), the result for English is surprising. To complicate matters further, a
19
different picture emerges when the correlations between rate and F1 and F2 are taken into
account. The correlations, spanning the range from -0.02 to 0.41, suggest that articulatory rate
exerts no influence on vowel reduction in its spatial aspect. It must be noted here that most of the
literature, reporting a relationship between rate and reduction (e.g. Jurafsky et al. 1998, Fosler-
Lussier and Morgan 1999) has been based upon studies of duration alone, disregarding formants.
Since speech rate has been shown to affect duration, there is no reason to take duration as a proxy
for overall reduction of vowels, which in addition to length, consists in formant changes. Perhaps
it would be necessary to draw a line between spatial and temporal reduction in search of any link
between rate and reduction of vowels. The paper tentatively suggests that these two, the spatial
and the temporal aspect, may be trending in different directions. Alternatively, the results
obtained here may also lead to a suggestion that lexical times of high frequency (dużo ‘a lot’,
wszyscy ‘everybody’, tutaj ‘here’, czasu ‘time’ (Gen. Sg), this, have, good) may be undergoing
the process of lexicalization in the speaker’s phonology and as such, might be growing
impervious to the effects of articulatory rate. Nevertheless, the issue whether fast speech indeed
fosters vowel reduction as it is commonly assumed, remains an open question, well worthy of
further investigation: “casual speech need not to be fast; some speakers [...] use a quite informal
speech even at fairly slow rates of speech, while others [...] give the impression of great precision
The aim of the present paper was to study the process of phonetic vowel reduction in two
typologically different languages (English and Polish) in two speech styles (sentence reading and
spontaneous speech). The results of the study provided evidence that the topic of vowel
reduction, especially in Polish, is worth pursuing further and deserves a more exhaustive
20
treatment. Firstly, the database should be enlarged by analyzing the speech samples from more
speakers (which does not pose a problem for the Polish database, as the corpus includes
recordings of 70 speakers; the Lancashire corpus is at present limited to 9 speakers, however, the
analysis could be extended to other accents of English as well). Secondly, the analysis should be
extended to cover more vowels. Finally, it is worth investigating the correlation between the
vowel reduction and the reduction of consonants and consonant clusters. The question to be
answered is Do speakers who reduce vowels more preserve the consonants in their speech and
vice versa? In other words, is there a compensation strategy (preserving vowels or consonants)
that speakers apply to ensure that the message is intelligible and comprehensible to the listener?.
Acknowledgements:
The author gratefully acknowledges the financial support of the National Science Center (grant
number 2012/05/D/HS2/03565) as well as the Ministry of Higher Education within the Program
Narodowy Program Rozwoju Humanistyki (The National Program of Humanities Development) (grant
number: 0113/NPRH2/H11/81/2013).
References:
Barnes, J. 2006. Strength and weakness at the interface: positional neutralization in phonetics
Bell, A. 1984. “Language style as audience design”. Language in society 13. 145-201.
21
Bell, A. 2001. “Back in style: reworking audience design”. In: Eckert, P. and J.R. Rickford (eds.),
Boersma, P. and D. Weenink. 2014. Praat: doing phonetics by computer [Computer program].
Cheshire, J. 1992. “Register and Style”. In: Bright, W. (ed.), International. Encyclopedia of
Crosswhite, K. 2003. “Vowel reduction”. In: Hayes, B., R. Kirchner, and D. Steriade (eds.),
Durand, J. and M. Pukli. 2004. How to construct a phonological corpus: PRAAT and the PAC
project. http://w3.erss.univ-tlse2.fr/membres/jdurand/DurandPukliPACtranscription.pdf
22
Farnetani, E. and M. G. Busa. 1999. “Quantifying the range of vowel reduction”. Proceedings of
Fosler-Lussier, E. and N. Morgan. 1999. “Effects of speaking rate and word frequency on
Fowler, C. and J. Housum.1987. “Talkers signalling of new and old words in speech and
listeners perception and use of the distinction”. Journal of Memory and Language 26.
489-504.
Harris, J. 2005. “Vowel reduction as information loss”. In: Carr, P., J. Durand. and C. J. Ewen,
132.
2015).
Jacewicz, E., R. A. Fox, C. O’Neill and J. Salmons. 2009.: “Articulation rate across dialect, age,
Jassem, W. 2003. “Illustrations of the IPA: Polish”. Journal of the International Phonetic
23
Jaworski, S. 2007. Lenition processes in English and other languages: A hierarchy of
Koopmans-van Beinum, F. J. 1980. Vowel contrast reduction: an acoustic and perceptual study
Laan, G. P. M. 1997. “The contribution of intonation, segmental durations, and spectral features
to the perception of a spontaneous and a read speaking style”. Speech Communication 22.
43-65.
Lindblom, B. 1963. “Spectrographic study of vowel reduction”. Journal of the Acoustical Society
Moon, S. J. and B. Lindblom. 1994. “Interaction between duration, context, and speaking style in
English stressed vowels”. Journal of the Acoustical Society of America, 96(1). 40-55.
3224.
Berkeley.
Raymond, W. D., R. Dautricourt and E. Hume. 2006. “Word-medial /t,d/ deletion in spontaneous
24
Roach, P. J. 2004. “Illustrations of the IPA. British English: Received Pronunciation”. Journal of
Sobkowiak, W. 2008. English Phonetics for Poles. (3rd edition). Poznań: Wydawnictwo
Poznańskie.
Spilkova, H. 2014. Phonetic reduction in spontaneous speech: an investigation of native and non-
Trondheim.
Vayra, M., C. Avesani and C. Fowler. 1999. “On the phonetic bases of vowel-consonant
Comparison Between Dutch in Belgium and the Netherlands”. Language and Speech
47(3). 297-30.
Zwicky, A. M. 1972. “On casual speech”. Papers from the Eighth Regional Meeting of the
25