下載/瀏覽

Download Report

Transcript 下載/瀏覽

Production and perception of a novel, secondlanguage phonetic contrast
James Emil Flege 1988, J. Acoust. S oc. Am.
93 (3), March 1993:1589-1608
Adapted into a class lecture note
by Raung-fu Chung
1
Outline
I.
II.
III.
IV.
V.
Literature Review & Hypothesis
Experimental design
Experiment 1-production
Experiment 2-identification of /t/
Experiment 3-choose the preferred /t/ and
/d/
VI. Experiment 4-imitation
VII. Discussion
VIII. General discussion
2
I. Literature Review &
Hypothesis
1. L1 would influence L2 learning.
Ex. French also use vowel duration as the segmental
cues to identify voiced and voiceless sounds. The
vowel duration of /p/,/t/,/k/</b/,/d/,/g/. Therefore, when
they learn English, it’s easy for them to produce those
words.
If the late learners treated word-final English /d/’s and /t/’s
as new then, by hypothesis, they should master the
word-final English /t/-/d/ contrast. (Hiramatsu,1990)
However, the recent studies didn’t support this
prediction (Flege, 1988; Crowther and Mann, 1992)
Hypothesis 1. Mandarin is a language without final
stops, and Taiwanese has final stops /k/,/t/, /p/. In the
present study, we want to know whether Mandarin
speakers and Taiwanese speakers would perform 3
differently.
I. Literature Review &
Hypothesis
2. The relation between production and perception.
Ex. In Dutch, perception precedes production. They
didn’t produce with longer vowel duration before /d/
than /t/, but when they hear it, they know their vowel
duration is different. Dutch also has vowel duration
difference. (Elsendoor)
Ex. In Japanese, some Japanese subjects may produce
English /r/ and /l/ better than they identify these nonnative sounds. (Goto, 1971)
2. We want to know whether perception precedes
production or not.
3. We also want to the role “age” plays in L2 acquisition.
4
II. Experimental design -subjects
They are all from Birmingham, Alabama.
5
II. Experimental design
Four reasons to choose vowel duration as
measurement.
1. Vowel duration is a unique status in English.
2. Vowel duration can be easily measured.
3. An incorrect specification of vowel-duration
contributes to foreign accent (Elsendoorn,
1984.)
4. Production and perception of vowel duration
has been compared readily.
6
III. Experiment 1Production
1.
2.
3.
4.
The subjects were asked to read out 14 words with
CVC structures in a carrying phrase “I will say
.”
The 14 words are the vowels /i/,
/a/,
,/e/,
and/u/, which were inserted into /b t/ and /b d/
frames.
If the subjects speak too fast, they will be asked to
say it again.
To avoid the possible effects of exposure to the
perceptual stimuli on production, experiment 1 was
carried out on the first day.
The measurement is PM device.
7
III. Experiment 1-Result
1. All five groups produce longer vowel duration in /d/ than in /t/.
2. From one-way ANOVA, post-hoc test Tukey’s HSD revealed
that NE and TC produced significantly larger differences than
TB, TA, and MA. No other differences were significant.
8
III. Experiment 1Discussion
1. The lack of a a word-final /t/ and /d/ in an
adult learner’s L1 doesn’t guarantee the
mastery of the contrast in English.
2. The lack of significant differences between
TA and TB suggest that an additional 4 years
of English learning is insufficient to increase
vowel duration difference measurably.
3. Late learners are also aware that /t/ differs
phonologically from /d/ in English.
9
III. Experiment 1Discussion
4. Look at the between groups in TC: the age
expose to English.
A. 3 subjects started to learn English
between 3-6 years old (125ms)
B. 9-13 years old (92ms)
C. Most subjects started to learn English
after 7 years old.
 The earlier they learn English, the closer
they are to native speakers.
10
IV. Experiment 2Identification of /t/
Although TB didn’t produce significantly
larger vowel difference than TA and MA.
If they can perceive better than TA and
MA, it can prove the hypothesis that
“perception precedes the production.”
11
IV. Experiment 2- Method
1.
2.
3.
4.
5.
Only beat-bead continua and bat-bad continua will be
tested.
The range of vowel duration in beat-bead is 140-461
ms. The range of vowel duration in bat-bad is 133451 ms. The interval is 20 ms.
It was played through the headphone. The subjects
were instructed to push a button marked “t” and “d”.
The 2 CVC continua were presented in
counterbalanced order.
The natural-edited stimulus will be used instead of
synthetic stimulus.
The dependent variable was the percentage of times
each stimulus was identified as ending in /t/.
12
IV. Experiment 2—Result
(/t/ percentage)
13
IV. Experiment 2-Result
(/t/ percentage)
1.
2.
3.
4.
According to the subjects’ performance, NE and TC
were considered as a group, and TB, TA, and MA
were considered another.
It can be sure that the long vowel duration will be
considered as before /d/, and the short vowel
duration will be considered as before /t/.
The slopes for beat-bead were significantly shallower
than those for bad-bad.
The experienced L2 learners have been found to
have shallower slopes than native speakers. It
suggested that segmental phonetic contrasts are less
“sharply defined” for them than native speakers.
14
IV. Experiment 2-Result
(boundary analysis)
The group doesn’t have effect, but the age has
effect. The performance of TC is closer to NE
because of the greater vowel duration.
15
IV. Experiment 2-Result
(consistency)
16
IV. Experiment 2-Result
(consistency)
1. The consistency scores were higher for
short-duration than for the long-duration
for all groups except the MA.
2. The lowest point of the consistency is
because when the vowel duration is at
the boundary of long- and short-vowel
duration, subjects were not sure
whether it is before /t/ or /d/.
17
IV. Experiment 2Discussion
Fact: Different from Experiment 1(the production test),
NE and TC are significantly different from TB, TA,
and MA. But in Experiment 2, shortening the
duration of vowels in natural edited bat-bad stimuli
decreased the percentage of /t/ judgment given by
the subjects in all groups. There was not a
significant difference between native speakers and
TB. It suggested that the experienced late learners
have acquired a native like sensitivity to vowel
duration.
Conclusion: Compare with Experiment 1, it supported the
hypothesis “the perception of vowel duration cues
precedes its production.”
18
IV. Experiment 2Discussion
Questions for this conclusion:
1. The lack of between-group differences and
the lack of statistical power.
2. Several findings in Experiment 2 suggested
that the late learners didn’t use vowel
duration in a truly native like manner.
3. For identification functions that could be
subject to probit analysis, the late learners’
phoneme boundaries occurred at shorter
vowel duration than NE and TC.
19
V. Experiment 3-Choose
the preferred /t/ and /d/
In order to solve the questions raised by
Experiment 2, Experiment 3 was taken.
Hypothesis: Flege (1988) said as in L1
acquisition, alignment of production and
perception may occur gradually in adult
learning.
 The possibility existed that the method of
adjustment would reveal significant perceptual
difference between NE and TB. In this way,
“perception precedes production” won’t be
correct.
20
V. Experiment 3- Method

1.
2.
3.
4.
Elsendoor used the method of adjustment to
examine the range of acceptable vowel duration
preceding /t/ and /d/ in English words for NE and
Dutch.
17 beat, 17 bead, 17 bat, and 17 bad are
recorded in order according to the vowel duration.
The orders of the target words and continua
were counterbalanced across subjects.
The stimuli were played out in order from the
shortest to the longest before each session “I will
say
.”
The first 2 of 17 trails in each session were not 21
analyzed.
V. Experiment 3- Result
22
V. Experiment 3- Result
1. NE and TC showed 2 distinct distributions for
words judged to have the best sounding /t/s
and /d/s.
2. MA and TA’s preferred /t/s and /d/s
overlapped considerably. TB overlapped MA
and TA somewhat less.
3. TC and TB resembled the native speakers in
showing a somewhat larger effect of vowel
duration for beat-bead than bat-bad. On the
other hand, TA and MA showed the opposite
pattern, and maintained the pattern in
Experiment 2.
23
V. Experiment 3Discussion
A post-hoc test revealed that average
difference scores were significantly
greater for NE and TC than TA and MA.
The NE and TC didn’t differ significantly,
however from TB.
 Therefore, the goal of the Experiment 4
is to resolve whether the experienced
learners did or did not make the same
perceptual use of vowel duration as NE.
24
VI. Experiment 4Imitation
1. A speech imitation task affords a unique
opportunity to asses the relation between
sensory and motor process.
2. The relation between categorization and
imitating.
 It appeared that discontinuities in imitated
vowel durations resulted from the subject’s
having covertly categorized the final stop as
/t/ or /d/ based on variation in preceding
vowel duration.
25
VI. Experiment 4Imitation
The present experiment assumed that nonnative speakers would show significantly
discontinuity in imitated vowel durations only
if they covertly categorized word-final stops
in the CVC stimuli as /t/ or /d/.
Hypothesis

In beat-bead and bat-bad continua, one might expect
the native speakers to be less accurate in imitating
vowel duration than non-natives because
categorization induced inaccuracy.

In isolated vowel imitation, native speakers might be
more accurate than the non-natives because of their
experience using vowel duration as a cue to
26
segmental phonetic distinction.
IV. Experiment 4-Method
1. The carrying phrase “I will say
.” will be
played to the subject, and the subject
imitated and say the whole phrase again.
17 beat-bead continua; 17 bat-bad continua;
17 isolated vowels, which were derived from
the beat-bead continua.
2. On the first day, the isolated vowels are tested
to avoid the interference caused by the other
experiments, and on the second day, CVC
continua were tested.
27
IV. Experiment 4- Result
(isolated vowels)
28
IV. Experiment 4- Result
(isolated vowels)
1. Both NE and TC produced vowels they
are roughly 100 ms larger than the
stimulus vowels.
2. TB and MA tended to overshot stimulus
durations to a greater extent for short
than long duration stimuli.
29
IV. Experiment 4- Result
(beat-bead continua)
30
IV. Experiment 4- Result
(beat-bead continua)
1.
2.
3.
The discontinuity of NE and TC take place in the
similar locations. (NE is around the 7-8 stimuli, and
TC is around 8-9 stimuli.)
The differing locations of these discontinuities
corresponded to differences in identifications of the
beat-bead stimuli by the NE and TC in Experiment 2.
It suggested that the discontinuities aroused from the
covert categorization of final stops in the beat-bead
stimuli.
There was no evidence of a discontinuity for 2
inexperienced late learner group.
31
IV. Experiment 4- Result
(bat-bad continua)
NE and TC had discontinuity at the same location
(around 8-9 stimuli.)
32
VI. Experiment 4-Result
(mean discontinuity scores)
33
VI. Experiment 4-Result
(mean discontinuity scores)
1. In isolated vowel duration, NE has the
least discontinuity because they use
vowel duration as a cue to identify /t/
and /d/.
2. The discontinuity scores of bat-bad
continua is larger than beat-bead
continua. The result is consistent with
the previous experiment 2.
34
VI. Experiment 4Discussion
1. The aim of the imitation experiment was to
2.
3.
determine if the three groups of late L2 learners
could show the same evidence of sensory-motor
linkage as NE.
The subjects in all five groups showed high
stimulus-response correlations. However, only the
native speakers and the childhood L2 learners
showed clear discontinuity in their beat-bead and
bat-bad continua.
TB is more similar to TA and MA, but not NE and
TC.  It suggested that the experienced late
learners probably didn’t establish native like
phonetic representations for the contrast between
word-final English /t/ and /d/.
35
VII. Discussion
1.
2.
3.
Although experienced and inexperienced Taiwanese late
learners never differed statistically, the consistent
difference in the four experiments showed the additional
L2 experience did enable the relatively experienced late
learners to approximate English phonetic norms to a
greater extent than the relatively inexperienced late
learners.
From Experiment 3, the foreign accented-scores were
also correlated with perceptual data obtained using the
method of adjustment.
From Experiment 4, we know the larger the discontinuity
scores for isolated vowels, the more accented the nonnatives are.
36
VIII. General Discussion
(production-perception
relation)
The pattern of significant and non-significant between
group differences obtained here is consistent with the
hypothesis that non-natives will resemble native speakers
more closely in perceiving than producing vowel duration
differences.
Data for individual subjects, on the other hand, were not
consistent with the “perception before production”
hypothesis.
For this, Leather concluded that there is “no simple
correspondences” between production and perception,
although the two may manifest an “independent association”
were in very early stages of L2 learning (1991, p.140).
Perhaps the study would have supported the “perception
before production” hypothesis had more experienced late
learners been examined.
37
VIII. General discussion
(Between-group differences)
1. Age Those who learned English after the
age of 9 years produced somewhat smaller
overall voicing effects than whose who began
learning English before the age of 9 years.
2. The experience of L2 learning The effect of
additional L2 experience for late learners
seems to have been a modest movement in
the direction of L2 phonetic norms.
3. The effect of Mandarin or Taiwanese as the
mother language Non-significant
differences between these two groups were
38
observed.