modelling the perceptual development of first- and second-language phonological contrasts with the...
Post on 19-Dec-2015
215 views
TRANSCRIPT
Modelling the perceptual developmentof first- and second-language
phonological contrastswith the Gradual Learning Algorithm
Paola Escudero, University of [email protected]
Paul Boersma, University of AmsterdamAmsterdam, February 22, 2001
Sound Contrasts
Sound contrasts and acoustic information An example of a sound contrast in English
What are the acoustic differences between the two?
Acoustic cues to “ship” vs. “sheep”
60 70 80 90 100 110 120
300
330
360
390
420
450
480
Duration (ms)
Elspeth’s production environment(Scottish English)
40 50 60 70 80 90 100110120130140150160170180
270
300
330
360
390
420
450
480
510
Duration (ms)
ship
sheep
lid
lead
filling
feeling
Snicker
sneaker
Liz’s production environment(Southern English)
40 50 60 70 80 90 100110120130140150160170180
270
300
330
360
390
420
450
480
510
Duration (ms)
ship
sheep
lid
lead
filling
feeling
Snickersneaker
Elspeth and Liz’s production environments, averaged
60 70 80 90 100 110 120
300
330
360
390
420
450
480
Duration (ms)
L /I/
L /i/
E /I/
E /i/
Do Elspeth and Liz perceive [350 Hz, 80 ms] as “ship” or as “sheep”?
60 70 80 90 100 110 120
300
330
360
390
420
450
480
Duration (ms)
?L /I/
L /i/
E /I/
E /i/
Why using the nearest production prototype in perception?
Why not perceiving it as an /e/ or an /o/?
Answer: “likelihood maximisation”, the most likely category
Functional principle: minimise the probability of perceptual confusion
How do Elspeth and Liz perceive the segments reliably?
[350 Hz,
80 ms]
350 Hz
not /I/
80 ms
not /i/
80 ms
not /I/
350 Hz
not /i/
/I/ *! *
/i/ * *
[350 Hz,
80 ms]
350 Hz
not /i/
80 ms
not /i/
80 ms
not /I/
350 Hz
not /I/
/I/ * *
/i/ *! *
E
L
How about baby Elspeth and baby Liz?
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Baby Elspeth or baby LizSpectral reliance 0.5%, duration reliance -0.3%
How about baby Elspeth and baby Liz?
Little Elspeth makes a mistake when categorising [350 Hz, 80 ms]
[350 Hz,
80 ms]
350 Hz
not /i/
80 ms
not /i/
80 ms
not /I/
350 Hz
not /I/
/I/ * *
/i/ *! *
Little Elspeth and little Liz learn to perceive “ship” and “sheep” reliably
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Adult ElspethSpectral reliance 97.7%, duration reliance 4.9%
Little Elspeth and little Liz learn to perceive “ship” and “sheep” reliably
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Adult LizSpectral reliance 60.7%, duration reliance 48.2%
The perception of real adult Elspeth and Liz
4.49 4.09
ScottishAverage
reliance 11%, weight 10%
[i]
[I]
[i:]
[I:]
5.2
3.71
SouthernAverage
reliance 32%, weight 30%
[i]
[I]
[i:]
[I:]
How about adult Isabel and adult Carmen that move to Scotland
and Southern England, respectively?
Do they learn to perceive English “ship” and “sheep”? If so, how do they do it?
Isabel and Carmen’s production environment (Spanish)
60 70 80 90 100 110 120
270
300
330
360
390
420
450
Duration (ms)
/i/
/e/
Isabel and Carmen’s perception (Spanish)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Adult Isabel or adult CarmenSpectral reliance 99.8%, duration reliance 1.3%
New production environment (English)
Scotland
60 70 80 90 100 110 120
300
330
360
390
420
450
480
Duration (ms)
E /I/
E /i/
How Isabel learns to perceive “ship” and “sheep” (1)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Isabel after entering Scotland (after copying the grammar)Spectral reliance 99.8%, duration reliance 1.3%
Scotland
How Isabel learns to perceive “ship” and “sheep” (2)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Isabel after 1 month in ScotlandSpectral reliance 98.8%, duration reliance -0.1%
Scotland
How Isabel learns to perceive “ship” and “sheep” (3)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Isabel after 2 months in ScotlandSpectral reliance 98.9%, duration reliance -2.1%
Scotland
How Isabel learns to perceive “ship” and “sheep” (4)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Isabel after 4 months in ScotlandSpectral reliance 99.5%, duration reliance 4.3%
Scotland
How Isabel learns to perceive “ship” and “sheep” (5)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Isabel after 8 months in ScotlandSpectral reliance 99.2%, duration reliance 3.0%
Scotland
How Isabel learns to perceive “ship” and “sheep” (6)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Isabel after 100 months in ScotlandSpectral reliance 97.7%, duration reliance 5.4%
Scotland
The perception of real Isabel
4.4 4.8
mt
reliance -10%
[i]
[I]
[i:]
[I:]
4.38 4.67
mf
reliance -6%
[i]
[I]
[i:]
[I:]
South England
New production environment (English)
60 70 80 90 100 110 120
300
330
360
390
420
450
480
Duration (ms)
L /I/
L /i/
How Carmen learns to perceive “ship” and “sheep” (1)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Carmen after entering Southern England (after duplicating a category)Spectral reliance -0.5%, duration reliance -0.2%
South England
How Carmen learns to perceive “ship” and “sheep” (2)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Carmen after 1 month in Southern EnglandSpectral reliance 21.6%, duration reliance 34.1%
South England
How Carmen learns to perceive “ship” and “sheep” (3)
South England60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Carmen after 2 months in Southern EnglandSpectral reliance 40.5%, duration reliance 32.2%
South England
How Carmen learns to perceive “ship” and “sheep” (4)
South England60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Carmen after 4 months in Southern EnglandSpectral reliance 50.8%, duration reliance 35.9%
How Carmen learns to perceive “ship” and “sheep” (5)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Carmen after 8 months in Southern EnglandSpectral reliance 56.7%, duration reliance 29.6%
South England
How Carmen learns to perceive “ship” and “sheep” (6)
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Carmen after 100 months in Southern EnglandSpectral reliance 65.1%, duration reliance 35.0%
South England
The perception of real Carmen (1)
5.29
3.5abg
reliance 40%, weight 46%
[i]
[I]
[i:]
[I:]
6.5
2.5
1.92jg
reliance 37%, weight 33%
[i]
[I]
[i:]
[I:]
BUT...! The perception of real Carmen (2)
4.25
4.75ef
reliance 86%
[i]
[I]
[i:]
[I:] 3.4
3.29af
reliance 87%, weight 88%
[i]
[I]
[i:]
[I:]
L1 and L2 categorisation: Elspeth vs. Isabel
4.49 4.09
ScottishAverage
reliance 11%, weight 10%
[i]
[I]
[i:]
[I:]
4.4 4.8
mt
reliance -10%
[i]
[I]
[i:]
[I:]
4.38 4.67
mf
reliance -6%
[i]
[I]
[i:]
[I:]
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Adult ElspethSpectral reliance 97.7%, duration reliance 4.9%
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Isabel after 1 month in ScotlandSpectral reliance 98.8%, duration reliance -0.1%
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Isabel after 8 months in ScotlandSpectral reliance 99.2%, duration reliance 3.0%
L1 and L2 categorisation: Liz vs. Carmen
5.2
3.71
SouthernAverage
reliance 32%, weight 30%
[i]
[I]
[i:]
[I:]
5.29
3.5abg
reliance 40%, weight 46%
[i]
[I]
[i:]
[I:] 3.4
3.29af
reliance 87%, weight 88%
[i]
[I]
[i:]
[I:]
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Adult LizSpectral reliance 60.7%, duration reliance 48.2%
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Carmen after 2 months in Southern EnglandSpectral reliance 40.5%, duration reliance 32.2%
60 72 84 96 108 120
280
320
360
400
440
480
Duration (ms)
Carmen after 8 months in Southern EnglandSpectral reliance 56.7%, duration reliance 29.6%
Summary
For our perception model, we assumed:1) OT perception grammar with GLA2) Grammar copying (full transfer for L2)3) Category duplication
With these assumptions, we can handle:1) L1 Scottish and Southern English2) L2 Scottish and Southern English, partially
BUT!!
We could not completely handle L2 English!!
Solution?: ‘Cues categorised separately’, the L2 grammar consists of the following constraints instead
1) “350 Hz not /mid/”, “350 Hz not /high/”
2) “80 ms not /short/”, “80 ms not /long/”
Conclusion
Cue reliance depends on cue reliability, or:Differences in production account for differences in perception.
The functional principle underlying this production-
perception dependence is “minimisation of perceptual confusion”.
This functional principle follows from our formal modelling of the perception grammar.