INTERSPEECH 2013
Recognizing words across regional accents:
The role of perceptual assimilation in lexical competition
Catherine T. Best1, Jason A. Shaw 1, Elizabeth Clancy1
1
MARCS Institute, University of Western Sydney, Australia
c.best@uws.edu.au, j.shaw@uws.edu.au, e.clancy01@gmail.com
spoken in L1 regional accents that are unfamiliar and
phonetically disparate from listeners’ native accent offer new
insights on abstract and episodic effects in lexical access.
Abstract
Unfamiliar regional accents disrupt spoken word recognition
by L2 and L1 learners and L1 adults, and confuse ASR and
smart systems. Little is known, however, about which aspects
of non-native accents hinder word recognition, or what
processes are involved. We assessed how Australian English
(AusE) listeners’ recognition of words in unfamiliar accents is
affected by two types of cross-accent perceptual assimilation:
1) other-accent phones that constitute ‘deviant’ versions of the
matching AusE phonemes (Category Goodness assimilation:
CG); 2) phones that cross a native phonological boundary, i.e.,
assimilate to mismatching AusE phonemes (Category Shift:
CS). Eyetracking (“visual world”) revealed the timecourse of
lexical competition during online identification of words
spoken in Jamaican (JaME: vowel differences from AusE) and
Cockney English (CknE: consonant differences), while
choosing among four printed choice words: target, onset and
offset competitors, unrelated distracter. Recognition was
slower, and both competitor types were considered more and
longer for JaME and CknE than AusE pronunciations; these
effects were stronger for CS than CG differences. We
conclude that: 1) perceptual assimilation plays a key role in
cross-accent word recognition; 2) lexical competition involves
not only onsets but also later aspects of words; 3) vowel and
consonant variations affect lexical competition similarly.
A likely factor in recognizing words in other L1 accents is
the ways in which the unfamiliar vowel and consonant
pronunciations are perceptually assimilated to phonemes in the
listener’s native accent. Therefore, we extended the principles
of the Perceptual Assimilation Model [PAM: 21, 22] to crossaccent word recognition, using critical vowels and consonants
that should be assimilated either to the same native-accent
phoneme but as a deviant token (Category Goodness type:
CG), or to a different, contrasting native phoneme (Category
Shifting type: CS, a novel extension of PAM’s Two Category
[TC] assimilation type). We designed two experiments to
compare how CG and CS accent differences in specific vowels
and consonants influence spoken word recognition.
2. Overview of experiments
We adapted the visual world paradigm, which provides
sensitive indices of the timecourse of lexical competition
during spoken word recognition [23, 24], as follows: 1) spoken
target words were presented in isolation rather than in carrier
phrases, because phonetic and phonological properties of
carrier sentences in either the native or unfamiliar accents
could confound target word recognition [25]; 2) the onscreen
choices participants used to indicate what they had heard were
printed words instead of pictures [26, 27]; 3) “not there”
appeared centrally to increase task sensitivity [26, 27]; 4) the
choice sets included the target word, a phonetically and
orthographically unrelated distractor, a target word onset
competitor, and an offset competitor, an innovation added to
probe how later portions of spoken words affect lexical access.
Some word recognition models posit left-to-right lexical
access that privileges onsets [13, 28], but others allow effects
from later in the word. For example, Shortlist [14, 15] assumes
bottom-up phoneme activation while exemplar models [16-18]
assume the lexicon is built of stored exemplars, yet in both
views any word position can contribute to lexical competition.
Index Terms: spoken word recognition, regional accent,
phonological categories, perceptual assimilation
1. Introduction
Regional accent variation is known to perturb spoken word
recognition, especially in second language (L2) learners [1, 2]
but does so even in native (L1) adults [3-6] and L1 learners [79]. Indeed, accent differences plague not only humans but also
automatic speech recognition systems (ASRs) and “smart”
devices [10, 11]. Little is known, however, about what
processes underlie cross-accent recognition, or about which
types of variations between the native/more familiar accent
and the unknown/less familiar accents cause those difficulties.
2.1. Participants
Models of native spoken word recognition provide crucial
guidance on the likely processes involved, i.e., those identified
by prior research that relied on stimuli in the listeners’ native
accent. The primary debate in that literature has been over
whether lexical access from spoken words is accomplished by
abstract processes involving identification of the words’
component phonemes [12-15], or by episodic memories or
stored traces of experienced exemplars [16-18]. More recently,
proponents of both views have acknowledged that lexical
access requires both processes, and have called for
development of hybrid models [17, 19, 20]; such models have
yet to be fleshed out, however. Investigations with words
Copyright © 2013 ISCA
Fourteen native Australian English listeners participated in
both experiments in a single session, all recruited from the
Intro Psychology pool at UWS. Two further participants were
tested but removed from the data as English was not their L1.
All reported having no exposure to the two accents of this
study: Jamaican Mesolect and Cockney English (SE London).
2.2. Key manipulations
2.2.1. Accents
2128
25- 29 August 2013, Lyon, France
When gaze duration to the crosshair reached 200ms, it and the
rectangle were replaced with the words “Not there” (Figure 1).
The target word played over the loudspeaker 100ms later. This
allowed participants to preview the choice words prior to
hearing the target, and ensured central fixation at its onset.
Participants completed 8 practice trials before the test phase.
Each experiment used words spoken in Australian English
(AusE) versus another regional accent rarely heard in
Australia. Because English accents differ mainly in their
vowels, Experiment 1 used a non-native regional accent with
many vowel differences from AusE: Jamaican Mesolect
English (JaME) [29-31]. Consonant differences among
English accents are less frequent and more restricted (e.g., to
specific words or positions). Yet the impact of consonant
differences on word recognition is of interest, given evidence
that consonants and vowels play different roles in word
structure and processing [32-35]. Thus, Experiment 2 used a
non-native accent differing from AusE primarily in certain
consonants (CknE: southeast London) [36].
3. Experiment 1: Jamaican English
The first experiment allowed us to test for differential effects
of CS and CG vowel differences on word recognition, by
using Jamaican Mesolect English (JaME) as our unfamiliar
accent. JaME differs from Australian (AusE) primarily in its
vowels. Consequently, most accent differences are localized in
syllable nuclei rather than in syllable margins.
2.2.2. Assimilation types
Target words were selected to have a single target vowel
(JaME: Exp. 1) or consonant (CknE: Exp. 2) that differed from
the AusE pronunciation. The critical phoneme in half of the
words for each unfamiliar accent displayed a CG difference
from AusE; those for the remainder showed a CS difference.
3.1. Materials
3.1.1. Audio target words
Target words were selected from an existing recorded corpus
of multiple tokens of isolated words produced by two female
native speakers of JaME (recorded in St Catherine’s parish,
Jamaica) and two of AusE (Sydney) chosen to match the
JaME speakers’ voice qualities and age. In each word used in
the present study the critical vowel in the JaME realization
differs from AusE such that our listeners should perceptually
assimilate it as either a CG (/ /) or CS (/
) difference from their native accent. All other
phonemes were pronounced similarly to AusE. We used 64
monosyllabic and 64 bisyllabic words (critical vowel in the
stressed initial syllable), evenly divided between high and low
frequency words (re: British [Celex] and/or AusE [SMH]). We
used one token per word per speaker, selected for best match
of voice quality and pitch contour among the speakers. We
added 35 dB of white noise to all experimental trial targets
(but not practice targets), to assure below-ceiling performance.
2.2.3. Printed choice word sets
The printed choice words for onset competitors were selected
to have the same onset [(C)(C)V] as the predicted AusE
assimilation of the target word when it was spoken in JaME
(Exp. 1) or CknE (Exp. 2). Offset competitors for
monosyllable targets shared the coda [V(C)(C)]; those for
bisyllable targets shared the final syllable [(C)(C)V(C)(C)],
relative to expected AusE assimilation of the target. Unrelated
distractors had no matching letters or phonemes, including
expected assimilations, in the same position as in the target.
2.3. Procedure
Participants were seated in a quiet room in front of a computer
monitor with an eye-tracker below it (Tobii x120). They
positioned their chin and forehead on a chin rest located 70cm
from the monitor. The audio target words played from a laptop
computer through a loudspeaker beside the monitor. Prior to
testing, the eye-tracker was calibrated to the participant’s gaze.
3.2. Results
Figure 2 shows the proportion of looks (fixation proportion) to
each printed word type (Onset competitor, Offset competitor,
and Unrelated distracter) as a function of time. The time
window shown in Figure 2 spans 500 ms to 1500 ms after the
start of a trial. This window was selected for analysis because:
1) at 500 ms the mean fixation proportion to the center (“Not
there”) in all conditions had fallen below 0.5 but was still
somewhat greater than that to any other choice, and 2) target
fixations had reached a plateau by 1500 ms in all conditions.
Figure 1: Schematic of an experimental trial. Participants click on the
crosshair after reading the words, left panel; fixate on the crosshair
until the eye-tracker detects their eyes (red square), middle panel;
then “not there” replaces the crosshair, the spoken target word is
played, and the participant clicks on the printed word corresponding
to the word they heard, final panel. The example word set assesses
perceptual assimilation of a Category Shifting (CS) vowel difference:
AusE DU(de) [dud] onset distractor and (t)OUR [tuə] offset distractor
compete with the target DOOR as spoken in JaME: [du].
The four panels in Figure 2 show the different stimulus
conditions. The top two panels show the control condition,
AusE-accented words. The bottom two panels show fixation
proportions to JaME-accented English. The panels on the left
show Category-Shifting (CS) words and those on the right
show Category-Goodness (CG) words. Comparison of the top
two panels with the bottom two panels reveals that, relative to
AusE controls (top panels), looks are more liberally distributed
across items for JaME-accented words. First, it is clear that the
trajectory of looks toward the target word is steeper for AusEaccented (top) than JaME-accented (bottom) words. Second,
the decline in looks to the distracters, particularly onset
competitors, is much more gradual for JaME-accented words
than for AusE-accented words. This indicates that the
competitor words were more distracting when target words
The trial procedure was designed to assure the participant
was fixating the center of the monitor as each audio target
word played. Trials began with a display of the four printed
choice words, one per quadrant, with a central crosshair.
Participants were asked to silently read the four words, then
click on the crosshair and gaze at it until a red rectange outline
appeared, triggered by the eye-tracker’s detection of fixation.
2129
were heard in the unfamiliar accent.
interactions between accent and assimilation type [F(1,13) =
25.81, p < .001] and between accent and distracter type
[F(2,12) = 9.18, p < .01] were also significant, as was the
three-way interaction [F(2,12) = 4.25, p < .05]. Separate posthoc ANOVAs were run for each distracter type, with accent
and assimilation type as factors. The effect of accent was
significant for onset distracters [F(1,13) = 47.83, p < .001] and
unrelated distracters [F(1,13) = 19.60, p < .001] and was
nearly significant for offset distracters [F(1,13) = 4.52, p =
.053]. The CS-CG assimilation type difference was not itself
significant for any of the distracter types. However, the accent
x assimilation type interaction was significant for both onset
[F(1,13) = 12.55, p < .01] and offset distracters [F(1,13) =
19.06, p < .001] but n.s. for unrelated distracters [F(1,13) < 1].
Figure 2 also shows differences between CS and CG
words in JaME-accented English. Fixation proportion for
onset competitors, but also for offset competitors, shows a
more gradual decline for CS words than for CG words. This
indicates accent differences that cross a category boundary
evoke greater competition between lexical items, even to some
degree for the later portion of the word (offset competitors).
ASSIMILATION TYPE
CG
0.6
AusE
0.4
3.3. Discussion
0.2
ACCENT
0.0
As expected, the unfamiliar accent slowed word recognition.
Jamaican-accented English evoked more looks to on-screen
competitors than did the same words produced in Australian
English, indicating that perceptual assimilation of other-accent
vowels to listeners’ native accent systematically affects lexical
competition and slows the process of spoken word recognition.
Importantly, the design of the study and its target words
revealed two additional novel findings: 1) vowels that display
category-shifting (CS) differences from the native accent
hinder word recognition more than those showing only withincategory goodness differences (CG); 2) nonetheless, CG
vowel variants also hinder recognition, and according to the
same patterns; 3) while lexical competition is strongest for
word onset phonetic similarities, it is non-negligibly affected
as well by word offset similarities (rime portion).
0.6
JaME
Mean proportion
Mean fixation
CS
Target
Onset
Offset
Unrelated
NotThere
0.4
0.2
0.0
500
700
900
1100
1300
1500 500
700
900
1100
1300
1500
Time (ms)
Figure 2: Mean fixation proportion to each printed choice item
(separate lines) by condition (accent*assimilation type). The top two
panels show AusE. The bottom two panels show looks to JaME.
Category Goodness [CG] (left) and Category Shifting [CS] (right)
words are shown for both accents.
Figure 3 shows mean fixation proportion across the 5001500ms window by condition and distracter type for
Experiment 1. For onset and offset distracters, assimilation
type modulates the effect of accent. Differences in fixation
proportion are greater for CS words than for CG words. To
evaluate statistical significance of this pattern, we conducted a
three-way repeated measures ANOVA on arcsine-transformed
fixation proportions (averages across the 500-1500ms
window). The factors were accent {AusE, JaME}, assimilation
type {CS, CG}, and distracter type {Onset competitor, Offset
competitor, Unrelated distracter}.
Next, we addressed whether consonant variations also
elicit lexical competition, specifically whether or not they
elicit the same patterns found with vowels. As noted earlier,
converging evidence suggests consonants and vowels play
substantively different roles in the phonological organization
of words and their recognition. Thus, it was uncertain whether
consonant differences would impact word processing similarly
to vowels, especially whether the different and more varied
positions of the consonants enhances or reduces their impact.
4. Exp 2: Cockney English (SE London)
4.1. Materials
4.1.1. Audio target words
Cockney was chosen for this experiment because most of its
vowels are very similar to AusE while certain consonants
differ, including both CG type assimilations (initial /t/ as [ts];
/r/ as [w]) and CS type assimilations (CknE /θ/ as /f/; initial /h/
as [ ]; medial/final /t/ as []; medial /ð/ as [v]; final /l/ as []).
Given the many phonotactic constraints on these consonant
realization differences, the critical consonant in CknE vs.
AusE target words varied among initial, medial and final
position. The JaME words did not display such positional
variation, as their critical vowel was always the nucleus of the
stressed syllable. We selected 1- and 2-syllable target words
following the same principles as Experiment 1. We again used
an existing corpus of words produced by two adult female
speakers of CknE (recorded in southeast London) and two new
female AusE speakers (Sydney), match for voice qualities and
age. Again, 35 dB of white noise was added to all tokens.
Figure 3: Mean fixation proportion to distracters by accent (JaME vs.
AusE) and assimilation type (CG vs. CS) between 500-1500ms. Error
bars display standard errors of the mean (s.e.m.).
The main effects were all significant: accent [F(1,13) =
73.86, p < .001], distracter type [F(2,12) = 65.70, p < .001],
and assimilation type [F(1,13) = 7.41, p < .05]. The
2130
distracter type interaction was significant [F(2,12) = 21.01, p <
.001], as was the three way interaction for accent, assimilation
and distracter type [F(2,12) = 43.57, p < .001].
4.2. Results
Figure 4 is structurally parallel to Figure 2. It shows
proportion of looks (fixation proportion) over time for each
distracter type (Onset competitor, Offset competitor, Unrelated
distracter) as a function of condition: Accent -- AusE (top),
CknE (bottom) x assimilation type -- CG (left), CS (right). The
same time window was used for analyses as in Experiment 1,
500ms to 1500ms, for the reasons given earlier. Comparison of
the top two panels with the bottom two panels reveals that,
relative to AusE controls (top panels), looks are more liberally
distributed across items for CknE-accented words. Again, as in
Experiment 1, the trajectory of looks toward the target word is
steeper for AusE-accented (top) than CknE-accented (bottom)
words, and the decline in looks to distracters (particularly
onset competitors) is much more gradual for CknE-accented
words than AusE-accented words. Thus, competitors were
again more distracting for target words in the unfamiliar
accent.
To pursue the source of the three-way interaction, separate
post-hoc ANOVAs were run for each distracter type with
accent and assimilation type as factors. These showed that
accent had a significant effect on onset competitors [F(1,13) =
45.78, p < .001] and offset competitors [F(1,13) = 24.06, p <
.001]. The interaction between accent and assimilation type
was also significant for both onset [F(1,13) = 7.90, p < .05]
and offset competitors [F(1,13) = 10.85, p < .01]. For
unrelated competitors, neither accent [F(1,13) = 3.33, p = .09]
nor the interaction between accent and assimilation type
[F(1,13) = 4.53, p = .08] reached significance.
ASSIMILATION TYPE
CG
0.6
AusE
0.4
0.2
ACCENT
0.0
0.6
CknE
Mean proportion
Mean fixation
CS
Target
Onset
Offset
Unrelated
NotThere
0.4
Figure 5: Mean fixation proportion to distracters by accent (CknE vs.
AusE) and assimilation type (CG vs. CS) between 500-1500ms. Errors
bars display s.e.m. values.
0.2
0.0
500
700
900
1100
1300
1500 500
700
900
1100
1300
1500
4.3. Discussion
Time (ms)
Figure 4: Mean fixation proportion to each item on the screen
(separate lines) by condition (accent*assimilation type). The top two
panels show AusE. The bottom two panels show looks to CknE.
Category Shifting [CS] (left) and Category Goodness, [CG] (right)
words are shown for both accents.
The key results of replicated Experiment 1 findings with a
very different unfamiliar accent, CknE, in which the target
words had been selected to exploit CG and CS type consonant
realization differences, rather than vowel differences, from
AusE. Remarkably, these similarities emerged despite the
different and more varied word positions of the critical
consonants (initial, medial and final word positions), as
compared to the constant nuclear position of the critical
vowels in the stressed syllable of Experiment 1 target words.
Figure 4 also shows differences between CG and CS
words produced in Cockney-accented English. Fixation
proportion for onset competitors shows a more gradual decline
for CS words than for CG words. This indicates that accent
differences that cross a category boundary evoke greater
competition between lexical items.
5. Conclusions
Figure 5 shows mean fixation proportion across the 5001500ms window (Figure 4) by condition and distracter type.
The patterns reveal variable effects of assimilation type across
accents and distracter types. There are more looks to both
onset and offset distracters for CknE-accented English than for
AusE-accented English. However, it appears that the effect of
accent is modulated by assimilation type in different ways for
onset and offset distracters. CS words draw more looks to
onset distracters than do CG words. Offset distracters show the
opposite pattern: CG words draw more looks than CS words.
Across both vowel and consonant differences, spoken word
recognition was slower for words spoken in the two unfamiliar
regional English accents than in listeners’ native AusE accent.
Moreover, while recognition of non-native-accented words
was disrupted more by onset than offset competitors, the latter
did systematically affect JaME and CknE word recognition.
As predicted, effects were larger for CS than CG type accent
differences. We conclude that: 1) perceptual assimilation plays
a key role in cross-accent recognition; 2) lexical competition
occurs not only in onsets but also later in words; 3) vowel and
consonant variations affect lexical competition similarly.
We again conducted a 3-way repeated measures ANOVA,
as in Experiment 2. The main effects of accent [F(1,13) =
45.62, p < .001] and distracter type [F(2,12) = 40.29, p < .001]
were significant, but that for assimilation type was not
[F(1,13) < 1]. Nor was the interaction between accent and
assimilation type significant. However, the assimilation type x
6. Acknowledgements
Australian Research Council research grants DP0772441 and
DP120104596 contributed support to this research.
2131
7. References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
Escudero, P. and Chladkova, K. “Spanish listeners’ perception of
American and Southern British English vowels”, J Acou Soc
Am, 128: EL254-260, 2010.
Mitterer & McQueen, J. “Processing reduced word-forms in
speech perception using probabilistic knowledge about speech
production”, J Exp Psy: Hum Perc Perf, 35: 244–263, 2009.
Adank P., Evans B., Stuart-Smith J., Scott S. “Comprehension of
familiar and unfamiliar native accents under adverse listening
conditions”, J Exp Psy: Hum Perc Perf: 35, 520–529, 2009.
Bradlow, A. and Bent, T. “Perceptual adaptation to non-native
speech”, Cognition, 106: 707-729, 2008.
Floccia et al., “Does a regional accent perturb speech
processing?”, J Exp Psy: Hum Perc Perf, 32: 1276–1293, 2006.
Floccia et al., “Regional and foreign accent processing in
English: Can listeners adapt?”, J PsychoLing Res, 38: 379–412,
2009.
Best, C.T., Tyler, M.D., Gooding, T., Orlando, C. and Quann, C.
“Development of phonological constancy: Toddlers’ perception
of native- and Jamaican-accented words. Psychological Science,
20: 539-542, 2009.
Nathan, L., Wells, B. and Donlan, C. “Children's comprehension
of unfamiliar regional accents: a preliminary investigation”, J
Child Lang, 25, 343-365, 1998
Schmale, R., Cristia, A., Seidl, A. and Johnson, E.K.
“Developmental changes in infants’ ability to cope with dialect
variation in word recognition”, Infancy, 15(6), 650–662, 2010.
Benzeghiba, M., et al “Automatic speech recognition and speech
variability”, Speech Comm, 49: 763-786, 2007.
Henton, C. “Bitter pills to swallow: ASR and TTS have drug
problems”, Int J Speech Tech, 8: 247-257, 2005.
Cutler, A., Eisner, M., McQueen, J. and Norris, D. “How
abstract phonemic categories are necessary for coping with
speaker-related variation”, in Fougeron C et al [Eds, Laboratory
Phonology 10, 91-111, de Gruyter, 2010
Marslen-Wilson, W.D. “Functional parallelism in spoken wordrecognition. Cognition, 25(1–2): 71-102, 1987.
Norris, D. “Shortlist: A connectionist model of continuous
speech recognition”, Cognition, 52(3): 189–234, 1994.
Norris, D. and McQueen, J. “Shortlist B: a Bayesian model of
continuous speech recognition”, Psych Rev, 115(2): 357–395,
2008.
Goldinger, S. “Words and voices: Episodic traces in spoken
word identification and recognition memory”, J Exp Psy: Learn
Mem Cog, 22: 1166-1183, 1996.
Goldinger, S. “A complementary-systems approach to abstract
and episodic speech perception”, Int Congr Phon Sci, 16: 49-54,
2007.
Johnson, K. “Speech perception without speaker normalization:
An exemplar model”, in K. Johnson and J. Mullennix ([Eds]
Talker variability in speech processing, 145-16, Academic, 1997.
Cutler, A. “Abstraction-based efficiency in the lexicon”,
Laboratory Phonology, 1: 301-318, 2010.
Pierrehumbert, J. “The next toolkit”, J Phon, 34: 516-530, 2006.
Best, C.T. "A direct realist perspective on cross-language speech
perception”, in W Strange [Ed], Speech perception and linguistic
experience, 167-200, York Press, 1995.
Best, C.T. and Tyler, M.D. Nonnative and second-language
speech perception”, in M. Munro and O. Bohn [Eds] Second
Language Speech Learning, 13-34, Benjamins , 2007.
Tanenhaus, M.K., Spivey-Knowlton, M.J., Eberhard, K.M., and
Sedivy, J.C., “Integration of visual and linguistic information
during spoken language comprehension”, Science, 268: 1632–
1634, 1995.
Weber, A. and Cutler, A., “Lexical competition in non-native
spoken-word recognition”, J. Memory & Lang., 50: 1-25, 2004.
McQueen, J.M.. and Viebahn, M.C. “Tracking recognition of
spoken words by tracking looks to printed words”, Q J Exp
Psych, 60(5): 661-671, 2007.
Brower, S. “Processing strongly reduced forms in casual speech
(PhD dissertation, Radboud University Nijmegen). http://
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
2132
View publication stats
pubman.mpdl.mpg.de/pubman/item/escidoc:576607:5/componen
t/escidoc:1480745/Brouwer(2010)_THESIS_authorVersion.pdf,
2010.
Mitterer, H. “The mental lexicon is fully specified: Evidence
from eye-tracking”, J Exp Psych: Hum Perc Perf 37(2): 496-513,
2011.
Marslen-Wilson, W., and Zwitserlood, P. “Accessing spoken
words: The importance of word onsets”, J Exp Psych: Hum Perc
Perf, 15(3): 576–585, 1989.
Devonish, H. and Harry, O.G. “Jamaican Creole or Jamaican
English: Phonology”, in E.W. Schneider, K. Burridge, B.
Kortmann, R. Mesthrie, and C. Upton [Eds], A Handbook of
Varieties of English: Phonology, 450-480. Mouton de Gruyter.
2004.
Patrick, P. L. “Jamaican Creole morphology and syntax”, in E.
W. Schneider, K. Burridge, B. Kortmann, R. Mesthrie, & C.
Upton [Eds], A Handbook of Varieties of English. Volume 2:
Morphology and Syntax, 407-439, Mouton de Gruyter, 1999.
Wassink, A.B. Theme and variation in Jamaican vowels.
Language Variation and Change, 13: 135-159, 2001.
Goldsmith, J. ”An overview of autosegmental phonology”,
Linguistic Analysis, 2: 23–68, 1976.
Cutler, A., Sebastián-Gallés, N., Soler-Vilageliu, O., and Van
Ooijen, B. “Constraints of vowels and consonants on lexical
selection: Cross-linguistic comparisons”, Mem & Cogn, 28(5):
746-755, 2000.
Van Ooijen, B. “Vowel mutability and lexical selection in
English: Evidence from a word reconstruction task. Memory &
Cognition, 24(5): 573-583, 1996.
Bonatti, L.L., Peña, M., Nespor, M. and Mehler, J. “Linguistic
constraints on statistical computations: The role of consonants
and vowels in continuous speech processing”, Psychological
Science, 16: 451-459. 2005.
Foulkes, P. and Docherty, G. “Urban Voices: Accent Studies in
the British Isles”, Arnold, 1999.