LANGUAGE AND SPEECH, Vol.

24, Part 2,1981 125

CROSS-LANGUAGE PHONETIC INTERFERENCE:

ARABIC TO ENGLISH*

i
i
JAMES EMIL FLEGE
Northwestern University
and
ROBERT PORT
Indiana University

This study compares phonetic implementation of the stop voicing contrast produced in
Arabic by Saudi Arabians and by both Americans and Saudis in English. The English stops
produced by Saudis manifested temporal acoustic correlates of stop voicing (VaT, stop
closure duration, and vowel duration) similar to those found in Arabic stops. Despite such
phonetic interference from Arabic to English, however, American listeners generally had
little difficulty identifying the English stops produced by the Saudis, with the exception of
/p/. This phoneme, which is absent in Arabic, was frequently produced with glottal pulsing
during the stop closure interval. The timing of /p/, however, suggests that the Saudis did
grasp the phonological nature of /p/ (i.e., that the contrast between /p-b/ is analogous to
that between /t-d/ and /k-g/) but were unable to control all the articulatory dimensions
by which this sound is produced.

Perhaps the most important and obvious aspect of foreign-accented speech is sound
substitutions, such as [s] for /e/ in French-accented "I sink so." But a large part of
what leads to the perception of accentedness probably cannot be adequately represented
by a segmental phonetic transcription. We began this study with the hypothesis that both
the phonological structure and phonetic characteristics of a speaker's native language will
influence his pronunciation of sounds in a foreign language learned in adulthood. Cross-
language interference may occur at several levels of organization. First, a speaker might
mispronounce a sound in a foreign language because no comparable sound exists in the
phonemic inventory of his native language (Lado, 1957). But if such a novel sound is
composed of features that specify sounds which do exist in the speaker's native language,
however, a contrastive analysis based on phonemic principles (see Flege, 1979) predicts
that it will be learned with relatively little difficulty (Weinreich, 1953"p. 22). If distinc-
tive features are indeed "commutable" (Jakobson, 1962, p. 420) and can thus be trans-
ferred from sound to sound, then a foreign language speech sound that represents a "hole
in the pattern" of the native language phonemic inventory should be easy to learn.
Second, interference might occur at the level of segmental phonetic features even if the

* This research is based on an Indiana University Ph.D. thesis by the first author which
was supervised by the second author. It was funded in part by NICHD grant HD 12511
to Indiana University, and by a Post-doctoral Fellowship (NIH grant NS 07107) to
the first author through the Institute for the Advanced Study of the Communication
Processes, University of Florida.
126 Arabic-English Phonetic Interference

more abstract phonological features that specify a sound have been correctly combined.
Support for the existence of this kind of interference would exist if language learners
were to mispronounce only certain allophones of a novel foreign language phoneme. And,
third, interference might result from cross-language differences in the phonetic imple-
mentation of a feature.
It has been claimed that a segmental phonetic transcription can, in principle, describe
all the linguistically controllable aspects of speech (Chomsky and Halle, 1968) but even
the best phonetic transcription can probably not capture perfectly an idiolect or accent. 1
Research on speech timing, for example, suggests that similar sounds found in different
languages may have quite different patterns of temporal implementation (Lehiste, 1970;
Kohler, 1979; Port, AI-ani and Maeda, 1980). Such cross-Iangu.age timing differences
may not be directly perceptible at a segmental level to most listeners, but they may well
contribute to the perception of accentedness and even, in some cases, result in diminished
intelligibility (10nasson and McAllister, 1972; Huggins, 1976). Although there is relatively
little cross-language research on coarticulation, it seems likely that this aspect of sub-
segmental phonetic implementation might sometimes also prove incommensurable across
languages. For example, the degree to which vowels preceding nasal consonants are
nasalized seems to vary from language to language (Clumeck, 1976).
In the present study we examined several acoustic dimensions that are phonetic
correlates of the phonological contrast between voiced and voiceless stops. Voice-onset
time (VOT) is a measure of the time between release of stop closure and the onset of
glottal pulsing (voicing). This acoustic dimension often distinguishes classes of stops
like /ptk/ and /bdg/, and may be sufficient to cue the perceptual distinction between
such stop categories (Lisker and Abramson, 1964, 1967, 1971). Duration of the closure
interval of a stop as well as the duration of vowels preceding a stop are two other
important temporal acoustic correlates of the voicing contrast in many languages (Lehiste,
1970; Klatt, 1976). And, finally, the presence or absence of glottal pulsing (voicing)
during the closure interval of a stop is very often an important spectral dimension by
which voiced and voiceless stops are distinguished (Lisker, 1978).
We recorded and measured phonetically similar material representing colloquial Saudi
Arabian Arabic, American English, and the accented English produced by Saudi Arabians
using the same instrumental techniques. Arabic was chosen as the counterpoint to English
in this study because the phonetic contrast between voiced and voiceless stops in Arabic
appears to differ from that of English (Yeni-Komshian, Caramazza and Preston, 1977;
Port, AI-ani and Maeda, 1980) and because Arabic lacks one of the stops found in English,
the voiceless labial stop /p/ (AI-ani, 1970). These cross-language differences offered the
opportunity to assess how a difference in phonological inventory as well as more subtle
differences in the phonetic implementation of a phonological contrast would affect
production of foreign language speech sounds by adult language learners.

1 See Monsen (1976) for an interesting discussion of the effect on intelligibility of

non-segmental phonetic differences between the speech of normal and deaf speakers
of English.
 J.E. Flege and R. Port 127

EXPERIMENT 1 : ARABIC

Since few previous studies provide data concerning the phonetic basis of the stop
voicing contrast in Arabic, it was first necessary to examine stops in the Saudi Arabian
dialect of Arabic in order to determine to what extent phonetic characteristics of Arabic-
accented English directly result from Arabic-specific patterns of phonetic implementation.

Methods

Six adult male native speakers of Arabic, all university graduates from central or
northeastern Saudi Arabia residing in Bloomington at the time of the study, served as
subjects. All the speakers reported having a [g] in their native dialect of Arabic. Speakers
read randomized lists of the Arabic words listed below from 3 x 5 in. cards, inserting each
test word into a constant carrier sentence [?agra wamfilelbeyt] 'I read __ and then
I go home':

'measured'
'difficult'
'Tass'
'led'
'kissed'
taas
baas
'encircle'
kaas
'kat' (tobacco)
'stepped'
'grew
'cup' old'faab
Initial faak
gaat
faag
gaad
stops Final stops

The test words were chosen so as to provide word-initial and word-final stop voicing
con trast in CV:C minimal pairs. Phonologically long vowels were chosen instead of
phonologically short vowels because a pilot study revealed that the duration of the long
vowel /aa/ is closer in duration to English /re/ (the vowel used in test words in the subse-
quent English experiment) than is short Arabic /a/. Flege (1979) found that in a pre-
dental stop environment the duration of Arabic /aa/ was 177 msec, short Arabic /a/, 98
msec. English /re/ averaged 187 msec when produced in a comparable phonetic context
by Americans. Each sentence was produced in colloquial Saudi Arabian (rather than
Classical
or Standard Arabic)2 while subjects were seated about 15 in. in front of a
microphone (Electrovoice Model 635A) in a sound-proof booth. The experimenter
monitored production of each sentence from outside the recording booth to ensure that
test words carried main sentence stress and that subjects did not introduce pauses or

2 Since reading colloquial Arabic represents an unusual task for speakers of Arabic, we
took precautions to insure that our subjects produced the test material in their native
dialect. Before the experiment each speaker listened to instructions recorded in
colloquial Saudi Arabian Arabic emphasizing the importance of producing the sentence
material in colloquial rather thall Stalldard or Classical Arabic. Since there is no /g/
in Classical Arabic, several words that are produced with [gj in Saudi Arabian dialects
(e.g., [gamac:.j v. [qamac;,j 'full moon') were presented as examples of words produced
ill colloquial A rabic. A n A rabie-speaking linguist later listened to the recordings and
confirmed that they had been produced ill colloquial Arabic.
128 Arabic-English Phonetic Interference

Fig. 1. Four acoustic intervals measured in the Arabic and English experiments: 1) initial
stop duration; 2) VOT; 3) vowel duration; 4) final stop duration. Words in the
Arabic experiment were preceded by [a] and followed by [w]; those in the
English experiment by [ey] and [A]. The final stop in the test word tap in this
utterance (arrow) was produced by a Saudi Arabian speaker of English with
glottal pulsing through the entire closure interval.

noticeable changes in speaking tempo.

The test material was recorded on a Revox (Model A700) tape recorder at 7 1/2
Lp.s. Wideband spectrograms were made of six tokens of each test word (Voice Identifi-
cation, Model 700) and four contiguous intervals measured by hand to the nearest 5 msec
according to the following segmentation criteria: 1) the closure interval of word-initial
consonants was measured from the offset of the preceding vowel (indicated by a sharp
decrease in energy in the region of F 1 and F2) to the beginning of the noise burst
signalling release of the stop closure interval or, in the case of fricatives, to onset of
energy in FI of the following vowel; 2) VOT was measured from the beginning of the
release burst to the first visible striation in the region of FI; 3) vowel duration was
measured from the onset of energy in F I to the offset of energy in F I and F2; and 4)
the closure interval of word-fmal consonants was measured from offset of the preceding
vowel to the beginning of the release burst (for stops) or to onset of energy in F 1 in the
following vowel (for unreleased stops and fricatives). Segmentation of these intervals
is illustrated in Fig. 1 for a sentence from the English experiment (to follow). In addition,
a binary judgment of the presence or absence of glottal pulsing during the stop closure
interval was made. Stops that showed visible periodic striations on the spectrogram
through at least half the closure interval were considered to have been produced with
glottal pulsing.
 177
117
37V:
176
101
124
52
116
119
80
72
85
VOT
134
131
75
82
77
90
135
(29)
(13)
(18)
(20)
(28)
(15)
(23)
(17)
(22)
(19)
C2 190
170
177
183
189
186
167
173
(23)
(21)
(17)
(22)
(26)
(28)
J.E. Flege and R. Port
(24) 129
)) VOT values
C] were not measured
generally for initial
Standard
showed /bdg/ glottal
deviations
continuous because these voiced
pulsing throughstops
in parentheses. the stop closure interval
Mean duration of phonetic intervals in Arabic, in msec.
Phonetic Interval

Results and discussion

Word-initial stops. In word-initial position /t/ and /k/ are aspirated stops in Saudi
Arabian Arabic, as shown in Table 1. The mean VOT of /t/ was 37 msec (range of all
tokens: 20-65 msec) and of /k/, 52 msec (range: 30-85 msec). These values, which are
130 Arabic-English Phonetic Interference

about 20 msec longer than the VaT values reported for utterance-initial stops in Lebanese
Arabic by Yeni-Komshian et al. (I977), are considerably longer than for the "short-lag"
stops found in languages like French or Spanish (Lisker and Abramson, 1964). On the
other hand, they are less aspirated (Le., have shorter VaT values) than. "long-lag" stops
found in languages such as English or Danish (Lisker and Abramson, 1967; Fischer-
Jrj>rgensen, 1968).
In addition to a VaT difference, the durations of the stop closure intervals of voiced
and voiceless stops in initial position were also different. Pre-stress It I was about 8 msec
longer than Idl, and Ikl was about 10 msec longer than Ig/. These duration differences
of about 12% were significant at the 0.01 level.
Place of articulation was found to exert an effect on the duration of stop closure
intervals similar to that found in English and many other languages (Lehiste, 1970). The
effect of place on the duration of stops (p < 0.01 for both voiced and voiceless stops)
was a decrease in duration of the closure interval as place of articulation moved further
back in the mouth (cL Fischer-Jrj>rgensen, 1964).
Vowel duration. The duration of the long vowel preceding voiced stops (ldl and Igl)
was not significantly longer than vowels preceding voiceless stops (It I and Ikl). The
difference in means amounts to only about 3% or 6-7 msec. This seems to violate the
claimed universality of the stop voicing effect on preceding vowel duration (Chen, 1970).
Our results here are not in agreement with Port et al. (I980) who reported a voicing
effect on preceding stressed vowels of about 8% or 13 msec in three-syllable words.
Word-final stops. In word-final position the closure interval of voiced and voiceless
stops did not show a significant contrast as did the word-initial stops. Our finding of no
duration difference as a function of voicing is in agreement with the finding of Port et
al. (I980) for speakers of several non-Saudi dialects of Arabic. Arabic thus seems to differ
from English and at least other Germanic languages in which voiceless stops are longer
than voiced stops (lbdgl) in post-stress position (Lisker, 1957; Elert, 1964; Kohler,
1979).
Glottal pulsing. Voiced and voiceless stops were distinguished by the presence or
absence of glottal pulsing. Table 2 indicates the percentage of stops in initial and final
position that exhibited visible glottal pulsing during at least half the closure interval.
Both voiced stops (ld,g/) were produced with glottal pulsing far more frequently than
were their voiceless cognates (It,kl) (p < 0.01) in both word-initial and word-final
position).

Conclusions

The stop voicing contrast of Saudi Arabian Arabic differs from that of American
English in several ways. Word-initial Arabic voiceless stops (It,kl) seem to be produced
with somewhat shorter VaT values than similar stops in English (Lisker and Abramson,
196 7). Voiceless stops in Saudi Arabian Arabic are produced with longer closure intervals
than homorganic voiced stops in word-initial, pre-stress position. This temporal contrast
does not exist in English (Stathopoulos and Weismer, 1979). There does not appear to
 J.E. Flege and R. Port 131

TABLE 2

Percentage of stops produced by Saudis with glottal pulsing

visible through at least half the closure interval;
"n" is number of tokens analyzed

11
100
94
92
69214
36
35
/k/
/d/
/g/
/g/
36
/t/6 Initial Position
/b/
/b/
36
36 Final Position
100
89
%

be a temporal contrast either between the closure intervals of voiced v. voiceless stops in
word-final (post-stress) position, or in the duration of stressed vowels preceding voiced
v. voiceless stops. English possesses both of these inversely related temporal correlates
of stop voicing.
Given that previous studies of a number of languages have reported a stop voicing
effect on preceding vowel duration (e.g., Chen, 1970), the present finding of no contrast
in vowel duration in Saudi Arabian Arabic is somewhat surprising. Studies of other
Arabic dialects have reported small or nonsignificant effects (Port et al., 1980; Port
and Mitleb, 1980) but Keating (1979) recently reported a similar negative finding for
both Czech and Polish. Thus, it appears that this phonetic context effect on vowel
duration may not be a phonetic universal as is often supposed.
Based on these findings we may conclude that Saudi Arabians learning English as a
foreign language will be faced with a number of clear cross-language phonetic differences.
To produce English stops without an Arabic accent a Saudi will need to modify Arabic
patterns of phonetic implementation or else acquire novel English-specific patterns beside
his existing Arabic patterns. If phonetic interference is direct and persistent, Saudis.may
be expected to maintain the stop voicing correlates of Saudi Arabian Arabic when
producing English stops. In addition, Saudis will also need to learn to produce English
/p/, since the phoneme does not exist in their native language.
In the next experiment we directly compared production of English stop voicing
by native speakers of English and Arabic in order to determine whether Saudis learn to
produce English stops according to English phonetic norms.
 132 tab
bat
mean:
80
160
72
92
204
dab
67
71
63
146
bad
97
87
76
90
163
199
135
55
59
30
51
139
153
94
(9) 43
ISO
gab
bag
(26) 114
77
116
138
134
133
94
78
82
138
151
71
83
back
bat
98
75
67
tab
cab
65
96
93
174
162
90
147
135
77
(19)
(16)
(24) -g/
/kArabic-English
/t-d/ Phonetic Interference
(11)
(12)
(16)
(24)
(10)
(19)
(33)
(15)
(13)
(27)
(9)
(38)
(35)
(22)
(12)
(8)
(17) (14)
(17)
(27)
(26)
(12)
(13)
(15)
(31)
(9)
(10)
(22)
(21)
(23)
(9)
(17)
(14)
ap tap groups, in msec. Standard deviations in parentheses
/b-p/
el
ration of
p Closure English stop closures and vowels produced
TABLE by three
3 Initial Stop speaker
Closure
 J.E. Flege and R. Port 133

EXPERIMENT 2: ENGLISH

Methods
Procedures for the English experiments were as similar as possible to those of the
Arabic experiment. As in the Arabic experiment, subjects read a randomized list of
minimal-pair test words that differed according to the voicing of word-initial or word-
final stops produced at all three places of articulation, as shown below:

Initial Final
pat bat tap tab
tab dab bat bad

cab gab back bag

The vowel /re/ was chosen because it most nearly resembles the Arabic vowel /aa/ found
in the test words of the Arabic experiment. The carrier sentence used in the English
experiment ("I say __ again to Bob"), was chosen to approximate the syllabic structure
of the carrier sentence used in the Arabic experiment.
Three groups of speakers (six in each) served as subjects. One group consisted of
Americans (Group Am), and two groups consisted of Saudi Arabian students at Indiana
University (Groups Ar} and Ar2). Three speakers in both Saudi groups had previously
served as subjects in the Arabic experiment. Speakers in the two Saudi groups were male
university graduates ranging in age from 24 to 32. Those in Arl had lived less than one
year (mean: 8 months) in the U.S. at the time of the study, while speakers in Ar2 had
lived in the U.S. for over two years (mean: 39 months). A preliminary questionnaire
indicated that speakers in both groups had received comparable English language training
in Saudi Arabia and had similar career objectives. Thus, any phonetic difference between
the two groups of Saudis should be due primarily to learning based on experience
speaking English.
The same acoustic correlates of stop voicing examined in the Arabic experiment -
segment duration, VOT, and glottal pulsing - were measured in this experiment according
to the same criteria. Measurement reliability was estimated by making a separate set of
duration measurements from 32 duplicate spectrograms (I98 acoustic intervals) produced
by one speaker.3 The average error was found to be 2.5 msec (range: 0-20 msec).
Computer-implemented data analysis was conducted as for the Arabic experiment.

Results and discussion

Results, presented in Table 3, indicate that phonetic differences between Arabic and
English lead to non-English phonetic characteristics in the English produced by Saudi
Arabians.

3 The intervals measured were closure of initial and final stops, pre- and post-stress
VOT, vowel duration, and utterance duration.
 134 Arabic-English Phonetic Interference

u
...
.2 --
0[50u:Jc0'+-E I
Q)...
L- 20
25
10
Arl
Ar2
Arl
Am
AR05 Am
AR
Ar2
15 /t-d/
/k-g/
Q)
Q)
Q)
e
1./V>
:J
.!;C
Ar2
~

Fig. 2. Mean closure duration differences between word-initial voiced and voiceless
stops produced by four speaker groups, in msec. The mean durations of voiced
stops are subtracted from those of homorganic voiceless stops. Results from
the Arabic experiment (Group AR) are juxtaposed to those of the English
experiment (Groups Am, Arl, Ar2).

Word-initial stops. The Saudi speakers (Arl, Ar2) produced a temporal correlate of
stop voicing for word-initial stops which was not produced by Americans. This temporal
contrast between /ptk/ and /bdg/ is displayed in Fig. 2, where the mean durations of
voiced stop closures are subtracted from the mean durations of homorganic voiceless
stops. Here we see that the Saudis made the closure intervals of voiceless stops longer
than those of voiced stops in word-initial position, a contrast which was significant
(p < 0.01) in all but one case (the /t-d/ contrast produced by Arl). The Americans, on
the other hand, either produced no temporal contrast or else made voiced stops slightly
(but non-significantly) longer than voiceless stops.
In order to display the influence of Arabic on the Saudis' English, results from the
Arabic experiment (marked at AR in Fig. 2) are juxtaposed to results from the English
experiment.
vaT values of the Saudis' English stops also closely resemble values found in Arabic.
As shown in Table 4, the VaT of /pkt/ produced by the Saudis (Arl, Ar2) averaged
about 25 msec less than VaT values produced by the Americans (Am). Both the effect
of place of articulation on VaT, and the difference in VaT between Americans and
both Saudi groups were significant (p < 0.01). In Fig. 3 we have cumulatively plotted
 J.E. Flege and R. Port 135

TABLE 4

Mean voice-onset times (VOT) produced by three speaker groups, in msec.

Standard deviations in parentheses

47
30
35
41
56
6721
14
cab
46
tap
/k/
/t/
(14)
(20)
(11)
(15)
(12)
Voice-Onset Time (VOT)
(14)
(18)
(10)
pat /p/
Ar2
Arl
Am

c:J 0Cii0aE:J>...0
U...
1/1
Q)
150
30
90
f-cu
::a. 0
u Q)
,/ V ,/
V ,/ V
/ / ,/
EI
? 0 Stop Closure
VOT

Am Al2 Arl Am Ar2 Arl AR Am Ar2 Arl AR

Ipl ItI Ikl

Fig. 3. Mean voice-onset time (VOT) and stop closure duration lined up at the onset of
the following vowel, in msec. Results from the Arabic experiment (Group AR)
are juxtaposed to those of the English experiment (Groups Am, Arl , Ar2).
 25
5
---
~
:;:::
Q)
Q) 00::JIf)~:---c136
UQ
E:JC
Q))
5 : 15 10
20 Arabic-English Phonetic Interference
.sc:
Q)

,--I
I 1

I I
I I
I I
I I
1 1

1 I
1 I

I I
I I

Ip-bl It- dl Ik - g 1

Fig. 4. Mean closure-duration difference between word-final voiced and voiceless stops
produced by four speaker groups, in msec. The mean durations of voiced stops
are subtracted from those of homorganic voiceless stops. Results from the
Arabic experiment (Group AR) are juxtaposed to those of the English experi-
ment (Groups Am, Arl, Ar2). The histogram for the Americans' (Group Am)
/t-d/ contrast represents 12 alveolar stops (of 72 tokens) that were not flapped.

VOT and the duration of the closure interval of /ptk/ produced by the three speaker
groups alongside similar results for Arabic from the Arabic experiment (AR). We see here
that the Americans (Am) produced longer VOT but shorter stop closure intervals in
English than the Saudis (Arl, Ar2) (p < 0.01). It is interesting to note that the sum of
the VOT and stop closure intervals for word-initial /t/ and /k/ remain fairly constant for
Saudi speakers in both the English and Arabic experiments (AR, Arl, Ar2). Since the
two experiments were designed to be as similar as possible,4 it is surprising to see that the
Saudis speaking English (Arl, Ar2) do not approximate the longer VOT of English /t/
and /k/, but instead tend to slightly shorten VOT (vis-ii-vis Arabic values (AR) from
Experiment I) and to lengthen the closure intervals of initial stops relative to Arabic
values. This suggests a compensatory relation between the closure intervals of voiceless

4 Experiments in two languages must be compared with great caution unless identical
phonetic material and procedures are used in both (see, e.g., Barry, 1974). Unfor-
tunately, it was impossible to find a full list of minimal pairs that are real words in
both Arabic and English. Use of nonsense CVCs seemed inadvisable because the focus
of study was the learning of English stop voicing rather than some hypothetical
phonetic ability.
 J.E. Flege and R. Port 137

- 0000c--00c
>0
Q) ~
=
Q)))'
II)
C
~
Q
E
Q
Ar2
Ar2
30
20
4010
Ar1
0 Am Am /t-d/
Q)
/p-b/

Am Ar2 Ar1 AR

/k-g/

Fig. 5. Mean vowel-duration contrast produced by four speaker groups, in msec. The
mean durations of vowels preceding voiceless stops are subtracted from the
durations of vowels preceding homorganic voiced stops. Results from the Arabic
experiment (Group AR) are juxtaposed to those from the English experiment
(Groups Am, Arl, Ar2).

stops and VOT. It is reminiscent of Weismer's claim (1980) that there may be a constant-
duration gesture of devoicing in English, and implies that the duration of VOT and
an adjacent closure interval may not be independently controlled.
Vowel duration. The effect of stop voicing on vowel duration is much smaller in the
Saudis' than in the Americans' English. As displayed in Fig. 4, the mean durations of
vowels preceding voiceless stops are subtracted from those of vowels preceding homorganic
voiced stops. We see that the Americans (Am) made vowels longer before voiced than
voiceless stops at all three places of articulation, a finding reported for English in many
previous studies (e.g., House and Fairbanks, 1953; Peterson and Lehiste, 1960). The
Saudis (Arl, Ar2), on the other hand, produced a much smaller vowel duration contrast
than the Americans; their differences reached significance in only three of six minimal
pairs (/t-d/ for Arl ; /k-g/ for Arl and Ar2). Note that the relatively small effect of stop
voicing on vowel duration in Arabic-accented English is closely comparable to the small
and nonsignificant effect found in Arabic in Experiment I (AR) and plotted in Fig. 4
for /t-d/ and /k-g/.
Word-final stops. The closure-duration contrast between word-final voiced and voice-
less stops produced by Saudis (Arl, Ar2) was much smaller than that produced by native
English speakers (Am). In Fig. 5 these contrasts are displayed by subtracting the mean
 138 Arabic-English Phonetic Interference

durations of voiced stops from those of homorganic voiceless stops. We see here that the
Americans (Am) made Iptkl longer than Ibdgl in final position, as expected from previous
studies of English (e.g., Lisker, 1957).5 The apparent duration contrast between the
Americans' It I and Idl is based on the few tokens (12 of 72) of It I and Idl that were not
flapped (where a flap was operationally defined as having a closure interval of 40 msec
or less). The flapped Itls and Idls were about equal in duration. The Saudis (Arl, Ar2),
on the other hand, produced much smaller duration contrasts between final voiced and
voiceless stops than the Americans. The newly-arrived Saudis (Arl) produced no signifi-
cant difference in any pair of word-final voiced-voiceless stops, but the relatively more
experienced Saudi speakers of English (Ar2) did make the closure intervals of voiceless
stops longer than those of voiced stops at all three places of articulation.
The relatively small magnitude of the Saudis' contrast between word-final stops
compared to the Americans' is clearly related to the absence of a duration contrast
between final voiced and voiceless stops in Arabic. In Experiment I (marked as AR in
Fig. 5) we found that the durations of the unflapped It I and Idl of Saudi Arabian Arabic
were about equal, while the closure interval of Igl was actually somewhat shorter than
that of Ikl in final position. Note that Saudis did not flap English It I or Id/. This is
somewhat surprising in view of the recent finding by Port and Mitleb (1980) that speakers
of Jordanian Arabic who had lived in the U.S. for about the same length of time as
speakers in our group Ar2 flapped word-final post-stress alveolar stops (in phrases like
"bat again") in a similar experimen tal con tex t.
Just as for stops in word-initial position, the effect of place of articulation on the
duration of final stops was significant for all three speaker groups (p < 0.01), the closure
interval shortening as the place of articulation moved further back in the mouth.
Glottal pulsing. Both Americans (Am) and Saudis (Arl, Ar2) produced the phono-
logically voiced stops Ibdgl with glottal pulsing, as seen in Table 5. (Note that the word-
initial stops being analyzed were intervocalic since they occurred sentence-medially after
the word say.) The native and non-native speakers of English differed, however, in their
production of voiceless stops. The Americans (Am) generally kept the closure intervals
of Iptkl free of glottal pulsing (except for the normally flapped It I which we did not
attempt to measure and have left out of the table). Both groups of Saudis (Arl, Ar2),
however, produced a larger percentage of Ipls with glottal pulsing than did the Americans
in both word-initial and word-final position (p < 0.0 I). The relatively less experienced
Saudi speakers of English (Arl) produced Ipl with glottal pulsing more frequently than
the Saudis (Ar2) who had lived for several years in the U.S. (p < 0.01).
The glottal pulsing observed during the closure interval of the Saudis' Ipls was stronger
than the "edge" vibrations noted by Lisker and Abramson (1967) as can be seen in Fig.
1. Moreover, it was generally audible when isolated by electronic gating and would there-
fore probably contribute to the perception of these stops as voiced. The glottal pulsing
we observed may have resulted from an insufficiently wide abduction of the vocal folds,

5 The unstressed syllable immediately after the keyword in the carrier sentence seems to
have made the "word-final" stops of this study comparable to the "in tervocalic"
stops of Lisker 's (1957) work.
 94-
-100
100
36
100
397
35
%991
940a
076
97
39
83
36
/d/
/k/
/t/036
a6
47
11
94
3 36
100
/g/97
/p/ J.E. Flege and R. Port139
/b/ through atthese
least
The alveolar halfproduced
stops
stops the closure
were by interval;
generally Am"n"were
groupproduced isasnumber
flaps
not of tokens
analyzed analyzed.
because
al
e Position
of stops produced by three speaker groups with Initial
glottal Position
pulsing visible

or else initiation of abduction which occurred too late to insure voicelessness during the
closure interval of /p/ (see Weismer, 1980). We cannot be entirely certain, of course,
that the Saudis' /t/ and /k/ were not also voiced because of the limited dynamic range of
a sound spectrograph. But since only the Saudis' /p/ frequently exceeded our criterion,
we can probably conclude that the Saudis' laryngeal control differed during their produc-
tion of /p/ as compared to /t/ and /k/. Future research using other instrumental techniques
should establish in greater detail whether glottal pulsing observed during a /p/ produced
by Saudis differs from that seen in /b/ (and other stops) since this question bears directly
on the issue of how speakers learn to control laryngeal timing during stop production.
We have seen that phonetic differences between Arabic and English seem to have a
direct influence on Saudis' production of English stops. The question remains, however,
whether the acoustic differences between stops produced by Americans and Saudis noted
here - as well as other acoustic differences we have not examined - will lead to
perceptual confusions for English-speaking listeners. The next experiment addresses this
issue.
 140 Arabic-English Phonetic Interference

EXPERIMENT 3: INTELLIGIBILITY OF STOPS PRODUCED BY SAUDIS

The English experiment showed that the stop voicing contrast produced by Americans
and Saudis differed along several phonetic dimensions. Some such differences might
only contribute to the perception of foreign accent, while others might result in mis-
perception. This experiment was designed to test our impression that many of the Saudis'
intended /p/s were perceivable as /b/, even though they had been produced under fairly
ideal conditions. This finding would not be surprising since English /p/ is widely considered
to pose a "problem" for Arabs learning English by those who teach them (e.g., Aziz,
1974).

Methods
The English sentences produced by both groups of Saudis (Arl, Ar2) were dubbed
from the master tape onto listening tapes using a matched pair of Revox Model A700
tape recorders. Extraneous sounds and repeated utterances were deleted, and pauses
inserted where necessary to yield 2.5 sec intervals between utterances. Care was taken
to insure that any variation in signal strength on the original recording was equalized on
the listening tapes.
Two randomizations of the test sentences were presented free-field to seven native
American graduate students in linguistics, none of whom had studied Arabic. They were
selected because of their experience in phonetics and in transcribing sounds of foreign
languages. These listeners heard the tapes at a comfortable level while seated equidistant
from an Advent loudspeaker in a quiet room. Although the listeners knew which English
words the Saudis had intended to produce, they were instructed to transcribe any real or
possible English word they heard.
A confusion matrix was prepared from listener responses to initial and final stops in
minimal pairs (pat/bat, tap/tab; tab/dab, bat/bad; cab/gab, back/bag) representing 478
responses to each English stop (6 tokens, 12 speakers, 6 listeners plus 3 tokens each for
a seventh listener who was interrupted after one randomization). Chi-square tests were
performed to determine the effect of place of articulation, phonological voicing, position
within the syllable, and speaker group on intelligibility.

Results
As seen in the confusion matrices in Table 6, the American listeners had difficulty
identifying some stops produced by the Saudis. In word-initial position there were seven
times as many confusions based on voicing than on place of articulation (p < 0.01).
About 2/3 of the voice confusions were between /p/ and /b/. In this regard it is important
to note that the expected confusion pattern for English speakers and listeners (Miller
and Nicely, 1955) is to find many place-of-articulation confusions but relatively few
confusions based on voicing.
We found twice as many confusions in word-final than word-initial position, a
difference which was significant at the 0.01 level. Here, too, errors due to voicing greatly
outnumbered those due to place (p < 0.01). And, finally, the relatively more experienced
 J.E. Flege and R. Port 141

TABLE 6

Percent identification of word-initial stops (above) and word-final stops (below)

produced by 12 Saudi speakers as /ptkbdg/. Each stop was presented for
identification 72 times to 7 listeners. Not included are 15 non-stop responses

-0s::
CI> I89I 95
993/t/
77%
/k/
/g/ 10II93
/d/51 /b/ 193
I4 Initial
6 Position
] /b/ 22
/p/ perceived
-0 /k/

-0s:: II190
I78 93
86 t/21 22
8/d/
10
/50%
/k/
/g/ /b/II 2 866 4 I3 4 Final Position
/p/ /b/ 49
] perceived

Saudi speakers of English (Ar2) produced fewer /p/s that were heard as [b) (16%) than
did the less experienced Saudis (36%) (p < 0.01).
 142 Arabic-English Phonetic Interference

Discussion
The confusion of /p-b/ is readily interpretable in terms of acoustic measurements
made in the English experiment. The /p/ in pat produced by the Saudis may have often
been heard as [b] both because its VaT was very short and because glottal pulsing
frequently occurred during the closure interval. Word-initial /t/ and /k/ were probably
seldom identified as voiced stops because they were not produced with glottal pulsing
and because the VaT of /t/ and /k/ was proportionally closer to English values than that
of /p/. It is surprising that the Saudis' initial /b/ was sometimes heard as [p] since it was
nearly always produced with glottal pulsing through the entire closure interval and
without aspiration at stop release, both of which should support perception of a voiced
stop by American listeners. Perhaps the American listeners, hearing too many IbIs,
randomly identified some as /p/.
In final position several acoustic dimensions seem to have led to confusions of /p/
and /b/. The relative shortness of the vowel in tab produced by the Saudis probably led
American listeners to hear some of th~ final stops in that word as [p] (see, e.g., Raphael,
1972). Both the frequent presence of glottal pulsing during the closure interval and the
lack of a temporal contrast between final /p/ and /b/ probably led the American listeners
to hear some of the final stops of tap as [b] .
We cannot be certain, of course, that the acoustic dimensions we examined in the
English experiment are alone responsible for these perceptual confusions, nor adequately
assess the effect on intelligibility of the acoustic differences we discovered between the
Saudis' and Americans' stops. Still, it seems likely that both glottal pulsing and the
articulatory timing variables we noted did contribute to a deficit in intelligibility.
Moreover, this experiment verifies the existence of a serious intelligibility problem for
/p/ and /b/ produced by Arab learners of English, as would be predicted by a contrastive
analysis (Lado, 1957).

GENERAL DISCUSSION

Phonetic level interference

Our results show that phonetic implementation of stop voicing in Saudi Arabian
Arabic directly influenced Saudis' production of English stop voicing. Having found a
small and non-significant effect of stop voicing on preceding vowel duration in Arabic,
we observed a correspondingly small effect in Arabic-accented English. The Saudis
produced a duration contrast between the closure intervals of word-initial voiced and
voiceless stops in English similar to that observed in Arabic, but the Saudis tended not to
produce a similar duration contrast between word-final English stops. This seemed to
reflect the lack of such a contrast in Arabic. Finally, the Saudis' VaT for English
stops was shorter than the VaT values produced by Americans, corresponding fairly
closely to Arabic VaT values. Thus the Saudis were clearly generalizing temporal
correlates of stop voicing found in Arabic to their production of English stops.
This means we cannot account for foreign accent strictly in terms of segmental
 J.E. Flege and R. Port 143

phonemic or phonetic feature differences between languages (see Kenstowicz and Kisse-
berth, 1979, p. 154), for even non-segmental differences in temporal implementation
carryover from one language to another. Since the temporal specification of speech
sounds can apparently vary in unpredictable ways from language to language (Lehiste,
1970; Port et al., 1980; Keating, 1979; Kohler, 1979), they must be learned and, in this
sense, be considered part of the linguistic knowledge of speakers. Yet it is often assumed
that the linguistic control of speech is restricted to segmental units of phonetic transcrip-
tion (Chomsky and Halle, 1968; cf. Lisker and Abramson, 1971; Lisker, 1974). Our
results tend to undermine this notion of a linguistic phonetic space restricted to a fixed
universal set of segmental elements.

The special case of Ipl

The Saudis' inability to produce a good English /p/ provides further evidence of inter-
ference at the level of phonetic implementation. Both groups of Saudis made the closure
interval of /p/ in pat much longer than that of /b/ in bat, even though no such duration
contrast was produced by Americans and no such /p-b/ contrast exists in Arabic. This
contrast in Arabic-accented English seems to represent an extrapolation by the Saudis of
the duration contrast between word-initial /t-k/ and /k-g/ which exists in Arabic to the
/p-b/ contrast of English. For word-final stops we found that the more experienced Saudi
speakers of English (Ar2) produced a small duration contrast between /bdg/ and /ptk/,
while the less experienced Saudis (Arl) did not. Thus the Saudis produced qualitatively
similar (non-English) duration contrasts between all three stop voicing pairs (fp-b/,
/t-d/, /k-gf) even though the first doesn't exist in Arabic and the second is ordinarily
neutralized by flapping in American English. The fact that the Saudis generalized a
pattern of temporal correlates of stop voicing evenly across all three places of articulation
suggests, first, that they recognized the functional similarity of the voicing contrast in
English and Arabic and, second, were aware that /p/ is equivalent to a "voiceless /b/"
or a "labial /t/." That is, they seemed to treat voicing and place as commutable at a
featurallevel.
Although the Saudis generalized the Arabic pattern of timing of articulatory closure
to a new segment (fp/), the same cannot be said of their control of glottal pulsing. The
difference in glottal pulsing that characterized the Saudis' contrast between /t-d/ and
/k-g/ (both in English and Arabic) was not clearly present in the Saudis' English /p-b/
contrast (many of their /p/s were produced with glottal pulsing during the closure
interval). Several studies of first and second language also suggest that it may be more
difficult to learn to control a new pattern of glottal-supraglottal timing than one
involving purely supraglottal timing. Smith (I979) reports that American children as old
as 4)2 produced a clearer contrast in duration than glottal pulsing between post-stress
voiced and voiceless stops (cf. Kewley-Port and Preston, 1974). Fischer-J</Jrgensen (I 968)
found that an adult French-Danish bilingual contrasted voiced and voiceless stops in
French and Danish by means of language-specific temporal correlates of stop voicing that
are appropriate to those two languages. However, this speaker did not maintain a
phonetic distinction between her French and Danish stops with respect to glottal pulsing.
144 Arabic-English Phonetic Interference

She produced /bdgJ in French without glottal pulsing during the closure interval as if
they were Danish voiced stops. Finally, data reported by Suomi (1976) suggest that
Finnish learners of English succeeded better in learning temporal correlates of the English
stop voicing contrast that do not appear in Finnish than in learning to contrast English
stops by means of glottal pulsing. It may be, then, that coordination of laryngeal control
with particular supraglottal articulatory gestures is an especially difficult articulatory
skill for both first and second language learners to acquire.

CONCLUSIONS

Although the universality of a voicing effect on preceding vowels almost has the status
of a truism of phonetic timing, it cannot be maintained as a strong universal. There may
be a tendency for vowels to be relatively longer before voiced consonants than before
voiceless consonants, but such a contrast is not uncontrollable by phonological factors.
lt is possible that Arabic and English differ in the internal structure of their syllables in
ways that result in language-specific timing differences, such as the voicing effect on the
duration of closure intervals in initial v. final stops.
Results of these experiments have important implications for a theory of interference
in second language acquisition. A difference in the phonemic inventories of Arabic and
English did not seem to be the principal cause of the Saudis' difficulty in producing a
perceptually effective English /p/. The timing of the labial articulation of /p/ was just
what one would expect if the Saudis were producing a voiceless /b/, a finding which
demonstrates their awareness of the phonological and phonetic features of /p/. The
Saudis' primary difficulty was in adjusting the glottis in such a way as to prevent glottal
pulsing from occurring during the closure interval of /p/. Although this instance of non-
commutability suggests an interdependence between features (those defining voicelessness
and labiality), it would be viewed by many as part of the implementation rules applied to
a matrix of phonetic features, and thus peripheral to the phonetic segments themselves.
Temporal correlates of the stop voicing contrast produced by Saudi Arabians exhibited
-even after several years in an English-speaking environment - only a modest amount of
modification in the direction of the English pattern of phonetic implementation (cL
Flege, 1980). Such timing effects are also currently considered to be a part of a sub
segmental level of phonetic implementation. Thus, our conclusion must be that the most
important interference from a first to a second language during the process of foreign
language acquisition occurs at the level of phonetic implementation rather than at an
abstract level of organization based on features.

REFERENCES

AL-ANI. S. (1970). Arabic Phonology (The Hague).

AZIZ. Y. (1974). Some problems of English consonant sounds for the Iraqi learner. English Language
Teaching Journal, 28, 166-168.
 J.E. Flege and R. Port 145

BARRY, W. (1974). Language background and the perception of foreign accent. Journal of Phonetics,
2,65-89.
CHEN, M. (1970). Vowel length variation as a function of the voicing of the following consonant
environment. Phonetica, 22, 129-159.
CHOMSKY, N. and HALLE, M. (1968). The Sound Pattern of English (New York).
CLUMECK, H. (1976). Patterns of soft palate movements in six languages. Journal of Phonetics, 4,
337-351.
ELERT, C.-<::. (1964). Phonologic Studies of Quantity in Swedish (Stockholm).
FISCHER-JRGENSEN, E. (1964). Sound duration and place of articulation. Zeitschrift fur Phonetik
Sprachwissenschaft und Kommunikationsforschung, 17, 175-207.
FISCHER-J9IRGENSEN, E. (1968). Les occlusives franyaises et danoises d'un sujet bilingue. Word, 24,
112-152.
FLEGE, J. (1979). Phonetic interference in second language acquisition. Ph.D. dissertation, Indiana
University.
FLEGE, 1. (1980). Phonetic approximation in second language acquisition. Language Learning, 30,
117-134.
HOUSE, A. and FAIRBANKS, G. (1953). The influence of consonant environment upon the secon-
dary characteristics of vowels. Journal of the Acoustical Society of America, 25, 105-113.
HUGGINS, A. (1976). Speech timing and intelligibility. In 1. Requin (ed.), A ttention and Perfor-
mance, VII (Hillsdale, N.J .).
JAKOBSON, R. (1962). On the identification of phonemic entities. In Roman Jakobson, Selected
Writings, I (The Hague), p. 418426.
JONASSON, J. and McALLISTER, R. (1972). Foreign accent and timing: An instrumental phonetic
study. Papers from the Institute of Linguistics, University of Stockholm, 14,1140.
KEATING, P. (1979). A Phonetic Study of a Voicing Contrast in Polish. Ph.D. dissertation, Brown
University.
KENSTOWICZ, M. and KISSEBERTH, C. (1979). Generative Phonology, Description and Theory
(New York).
KEWLEY-PORT, D. and PRESTON, M. (1974). Early apical stop production: A voice onset time
analysis. Journal of Phonetics, 2, 195-210.
KLA TT, D. (1976). Linguistic uses of segmental duration in English: Acoustic and perceptual evidence.
Journal of the Acoustical Society of America, 59, 1208-1221.
KOHLER, K. (1979). Dimensions in the perception of fortis and lenis plosives. Phonetica, 36, 332-343.
LADO, R. (1957). Linguistics Across Cultures: Applied Linguistics for Language Teachers (Ann
Arbor).
LEHISTE, I. (1970). Suprasegmentals (Cambridge, Mass.).
LISKER, L. (1957). Closure duration and the intervocalic voiced-voiceless distinction in English.
Language, 33,4249.
LISKER, L. (1974). On time and timing in speech. In T.A. Sebeok, A.S. Abramson, D. Hymes, H.
Rubenstein, E. Stankiewicz and B. Spolsky (eds.), Current Trends in Linguistics, 12 (The
Hague), pp. 2387-2418.
LISKER, L. (1978). Rabid vs. rapid: A catalogue of acoustic features that may cue the distinction.
Haskins Laboratories Status Report on Speech Research, SR-54.
LISKER, L. and ABRAMSON, A.S. (1964). A cross-language study of voicing in initial stops: Acous-
tical measurement. Word, 20,384422.
LlSKER, L. and ABRAMSON, A.S. (1967). Some effects of context on voice onset time in English
stops. Language and Speech, 10, 1-28.
LISKER, L. and ABRAMSON, A.S. (1971). Distinctive features and laryngeal control. Language, 47,
767-785.
MILLER, G. and NICELY, P. (1955). An analysis of perceptual confusions among some English
consonants. Journal of the Acoustical Society of America, 27, 338-352.
MONSEN, R. (1976). The production of English stop consonants in the speech of deaf children.
Journal of Phonetics, 4, 2941.
 146 Arabic-English Phonetic Interference

PETERSON. G. and LEHISTE. I. (1960). Duration of syllable nuclei in English. Journal of the Acous-
tical Society of America, 32,693-703.
POR T. R., AL-ANI. S. and MAEDA, S. (1980). Temporal compensation and universal phonetics.
Phonetica, 37, 235-252.
POR T. R. and MITLEB. F. (1980). Phonetic and phonological manifestations of the voicing contrast
in Arabic-accented English. Research in Phonetics, 1 (Dept. of Linguistics, Indiana Univer-
sity),137-165.
RAPHAEL. L. (1972). Preceding vowel duration as a cue to the perception of the voicing character-
istic of word-final consonants in American English. Journal of the Acoustical Society of
America. 51, 1296-1303.
SMITH. B. (1979). A phonetic analysis of consonant devoicing in children's speech. Journal of Child
Language, 6, 19-28.
STATHOPOULOS. E. and WEISMER. G. (1979). The duration of stop consonants. In J. Wolf and D.
Klatt (eds.), Speech Communication Papers: 97th Meeting of the Acoustical Society of
America (New York), pp. 197-201.
SUOMI. K. (1976). English Voiced and Voiceless Stops as produced by Finnish and Native Speakers.
Jyvaskkyla Contrastive Studies, 2, (University of Jyvaskkyla, Finland).
WEINREICH. U. (1953). Languages in Contact: Findings and Problems (The Hague).
WEISMER, G. (1980). Control of the voicing distinction for intervocalic stops and fricatives: Some
data and theoretical considerations. Journal of Phonetics, 8,427438.
YENI-KoMSHIAN, G., CARAMZZA, A. and PRESTON. M. (1977). A study of voicing in Lebanese
Arabic. Journal of Phonetics, 5, 3549.