title slide
DESCRIPTION
Title Slide. Stop-Consonant Perception in 7.5-month-olds: Evidence for gradient categories. Bob McMurray & Richard N. Aslin Department of Brain and Cognitive Sciences University of Rochester. With thanks to Julie Markant & Robbie Jacobs. Learning Language. Meaning. Lexicon. S. VP. - PowerPoint PPT PresentationTRANSCRIPT
![Page 1: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/1.jpg)
Stop-Consonant Perception in 7.5-month-olds:
Evidence for gradient categories
Bob McMurray & Richard N. Aslin
Department of Brain and Cognitive SciencesUniversity of Rochester
Title Slide
With thanks to Julie Markant & Robbie Jacobs
![Page 2: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/2.jpg)
Understanding spoken language requires that children learn a complex mapping…
Learning Language
What is the form of this mapping?
How do the demands of learning affect this representation?
Lexicon
All labs
Bob’s lab
NP
the lab
S
VP
produced
MeaningAcousticAcoustic LexiconLexicon
Language Understanding
![Page 3: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/3.jpg)
Speech perception and word recognition require mapping…
Learning Speech
What representations mediate acoustics and lexical or sublexical units?
How does learning affect this representation?
AcousticAcoustic Sublexical Units
/b/
/la//a//l/ /p/
/ip/
Sublexical Units
/b/
/la//a//l/ /p/
/ip/
LexiconLexicon
Syntax, semantics,
pragmatics…
Speech Recognition
…continuous, variable perceptual input toa something discrete, categorical.
![Page 4: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/4.jpg)
1) Acoustic mappings: Categorical and gradient perception in adults and infants.
2) Infant speech categories are graded representations of continuous detail.
3) Statistical learning models and sparse representations.
4) Conclusions and future directions.
Overview
Overview
![Page 5: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/5.jpg)
What is the nature of the mapping between continuous perception and discrete categories?
How are these representations sensitive (or not) to within-category detail?
Categorization & Categorical Perception
Representation of Speech Detail
Empirical approach:• Use continuously variable stimuli.• Explore response using
Discrimination Identification (adults)Habituation (infants)
![Page 6: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/6.jpg)
Categorical Perception 1
B
P
Subphonemic within-category variation in VOT is discarded in favor of a discrete symbol (phoneme).
• Sharp labeling of tokens on a continuum.
VOT
0
100
PB
% /p
/
ID (%/pa/) 0
100
Discrim
ination
Discrimination
• Discrimination poor within a phonetic category.
Categorical Perception
![Page 7: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/7.jpg)
Categorical Perception 2
Many tasks have demonstrated within-category sensitivity in adults...
Discrimination Task Variations Pisoni and Tash (1974) Pisoni & Lazarus (1974)Carney, Widin & Viemeister (1977)
Training Samuel (1977)Pisoni, Aslin, Perey & Hennessy (1982)
Goodness Ratings Miller (1997)Massaro & Cohen (1983)
BUT…
And lexical activation shows systematic sensitivity to subphonemic detail (McMurray, Tanenhaus & Aslin, 2002).
![Page 8: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/8.jpg)
Infant Categorical Perception 1
Infants have shown a different pattern.
For 30 years, virtually all attempts to address this question have yielded categorical discrimination.
Categorical Perception in Infants
Exception: Miller & Eimas (1996).• Only at extreme VOTs.• Only when habituated to non- prototypical token.
GWB
![Page 9: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/9.jpg)
Infant Categorical Perception 3
Nonetheless, infants possess abilities that would require within-category sensitivity. Su
cking
Rate
(inter
est)
B BB B B B P P P P
Suck
ing R
ate (in
terest
)
B BB B B B P P P P
• Infants can use allophonic differences at word boundaries for segmentation (Jusczyk, Hohne & Bauman, 1999; Hohne, & Jusczyk, 1994)
• Infants can learn phonetic categories from distributional statistics (Maye, Werker & Gerken, 2002).
![Page 10: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/10.jpg)
Distributional Learning 2
Speech production causes clustering along contrastive phonetic dimensions.
Distributional Learning
E.g. Voicing / Voice Onset TimeB: VOT ~ 0P: VOT ~ 40
Result: Bimodal distribution
Within a categories, VOT is distributed Gaussian.
VOT0ms 40ms
![Page 11: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/11.jpg)
• track frequencies of tokens at each value along a stimulus dimension.
VOT
freq
uenc
y
0ms 50ms
Distributional Learning 1
Distributional Learning
To statistically learn speech categories, infants must:
• This requires ability to track specific VOTs.
• Extract categories from the distribution.
+voice -voice
![Page 12: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/12.jpg)
?Question 1
Prior examinations of speech-categories used:
• HabituationDiscrimination not ID.Possible selective adaptation.Possible attenuation of sensitivity.
• Synthetic speechNot ideal for infants.
• Single exemplar/continuumNot necessarily a category representation
Experiment 1: Reassess this issue with improved methods.
![Page 13: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/13.jpg)
HTPP 1Misperception 3
Head-Turn Preference Procedure (Jusczyk & Aslin, 1995)
Infants exposed to a chunk of language:
• Words in running speech.
• Stream of continuous speech (ala statistical learning paradigm).
• Word list.
Head-Turn Preference Procedure
After exposure, memory for exposed items (or abstractions) is assessed by comparing listening time to consistent items with inconsistent items.
![Page 14: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/14.jpg)
HTPP 2Misperception 3
Test trials start with all lights off.
![Page 15: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/15.jpg)
HTPP 2Misperception 3
Center Light blinks.
![Page 16: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/16.jpg)
HTPP 3Misperception 3
Brings infant’s attention to center.
![Page 17: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/17.jpg)
HTPP 3Misperception 3
One of the side-lights blinks.
![Page 18: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/18.jpg)
When infant looks at side-light……he hears a word
Beach… Beach… Beach…
HTPP 4Misperception 3
![Page 19: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/19.jpg)
…as long as he keeps looking.
HTPP 5Misperception 3
![Page 20: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/20.jpg)
Experiment 1 MethodsMisperception 3
Experiment 1
7.5 month old infants exposed to either 4 b-, or 4 p-words.
80 repetitions total.
Form a category of the exposed class of words.
PeachBeachPailBailPearBearPalmBomb
Measure listening time on…
VOT closer to boundaryCompetitors
Original words
Pear*Bear*BearPearPearBear
![Page 21: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/21.jpg)
Experiment 1 StimuliMisperception 3
B* and P* were judged /b/ or /p/ at least 90% consistently by adult listeners.
B*: 97%P*: 96%
Stimuli constructed by cross-splicing naturally produced tokens of each end point.
B: M= 3.6 ms VOTP: M= 40.7 ms VOT
B*: M=11.9 ms VOTP*: M=30.2 ms VOT
![Page 22: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/22.jpg)
Experiment 1 Familiarity vs.
NoveltyMisperception 3
Novelty/Familiarity preference varies across infants and experiments.
1221P
1636B
FamiliarityNoveltyWithin each group will we see evidence for gradiency?
Familiarity vs. Novelty
We’re only interested in the middle stimuli (b*, p*).
Infants were classified as novelty or familiarity preferring by performance on the endpoints.
![Page 23: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/23.jpg)
Categorical
Experiment 1 Fam. vs. Nov. 2Misperception 3
Gradiency
What about in between?
After being exposed to bear… beach… bail… bomb…
Infants who show a novelty effect……will look longer for pear than bear.
Gradient
Bear*Bear Pear
List
enin
g Ti
me
![Page 24: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/24.jpg)
4000
5000
6000
7000
8000
9000
10000
Target Target* Competitor
Lis
teni
ng T
ime
(ms)
Experiment 1 Results
Experiment 1 Results Nov
BP
Exposed to:
Novelty infants (B: 36 P: 21)
Target vs. Target*:Competitor vs. Target*:
p<.001p=.017
![Page 25: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/25.jpg)
Experiment 1 Results Fam
Familiarity infants (B: 16 P: 12)
Target vs. Target*:Competitor vs. Target*:
P=.003p=.012
4000
5000
6000
7000
8000
9000
10000
Target Target* Competitor
Lis
teni
ng T
ime
(ms) B
P
Exposed to:
![Page 26: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/26.jpg)
Experiment 1 Results Planned PMisperception 3
Planned Comparisons
Infants exposed to /p/
NoveltyN=21
P P* B
.024*
.009**
P P* B
.024*
.009**
4000
5000
6000
7000
8000
9000
10000
Lis
teni
ng T
ime
(ms)
P* B4000
5000
6000
7000
8000
9000
.018*
.028*
.018*
P
Lis
teni
ng T
ime
(ms)
.028*
FamiliarityN=12
![Page 27: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/27.jpg)
NoveltyN=36
<.001**>.1
<.001**>.2
4000
5000
6000
7000
8000
9000
10000
B B* P
Lis
teni
ng T
ime
(ms)
Experiment 1 Results Planned BMisperception 3
Infants exposed to /b/
FamiliarityN=16
4000
5000
6000
7000
8000
9000
10000
B B* P
Lis
teni
ng T
ime
(ms)
.06.15
![Page 28: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/28.jpg)
Experiment 1 ConclusionsMisperception 3
7.5 month old infants show gradient sensitivity to subphonemic detail.
• Clear effect for /p/• Effect attenuated for /b/.
Experiment 1 Conclusions
Contrary to all previous work:
![Page 29: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/29.jpg)
Experiment 1 Conclusions 2Misperception 3
Reduced effect for /b/… But:
Bear Pear
List
enin
g Ti
me
Bear*
Null Effect?
Bear Pear
List
enin
g Ti
me
Bear*
Expected Result?
![Page 30: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/30.jpg)
Experiment 1 Conclusions 3Misperception 3
• Bear* Pear
Bear Pear
List
enin
g Ti
me
Bear*
Actual result.
• Category boundary lies between Bear & Bear*• Between (3ms and 11 ms).
• Will we see evidence for within-category sensitivity with a different range?
![Page 31: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/31.jpg)
Experiment 2Misperception 3
Same design as experiment 1.
VOTs shifted away from hypothesized boundary (7 ms).
Train
40.7 ms.Palm Pear Peach Pail
3.6 ms.Bomb* Bear* Beach* Bale*
-9.7 ms.Bomb Bear Beach Bale
Test:
Bomb Bear Beach Bale -9.7 ms.
![Page 32: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/32.jpg)
Experiment 2 Results FamMisperception 3
Experiment 2 Results
Familiarity infants (34 Infants)
4000
5000
6000
7000
8000
9000
B- B P
Lis
teni
ng T
ime
(ms) =.05*
=.01**
![Page 33: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/33.jpg)
Experiment 2 Results NovMisperception 3
Experiment 2 Results
Novelty infants (25 Infants)
=.02*
=.002**
4000
5000
6000
7000
8000
9000
B- B P
Lis
teni
ng T
ime
(ms)
![Page 34: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/34.jpg)
Experiment 2 ConclusionsMisperception 3
Experiment 2 Conclusions
• Within-category sensitivity in /b/ as well as /p/.
VOT
Adult boundary
/b/ /p/
Cat
egor
y M
appi
ngSt
reng
th Adult Categories
• Shifted category boundary in /b/: not consistent with adult boundary (or prior infant work). Why?
![Page 35: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/35.jpg)
Experiment 2 Conclusions 2Misperception 3
/b/ results consistent with (at least) two mappings.
VOT
Adult boundary
/b/ /p/
Cat
egor
y M
appi
ngSt
reng
th 1) Shifted boundary
• Inconsistent with prior literature.
• Why would infants have this boundary?
![Page 36: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/36.jpg)
Experiment 2 Conclusions 3Misperception 3
2) Sparse Categories/b/
VOT
Adult boundary
/p/
Cat
egor
y M
appi
ngSt
reng
th
unmappedspace
HTPP is a one-alternative task. Asks: B or not-B not: B or P
Sparse categories may in fact by a by-product of efficient statistical learning.
![Page 37: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/37.jpg)
Model IntroMisperception 3
Distributional learning model/b/
VOT
Adult boundary
/p/
Cat
egor
y M
appi
ngSt
reng
th
unmappedspace/b/
VOT
Adult boundary
/p/
Cat
egor
y M
appi
ngSt
reng
th
unmappedspace
Computational Model
1) Model distribution of tokens asa mixture of gaussian distributions over phonetic dimension (e.g. VOT) .
2) After receiving an input, the Gaussian with the highest posterior probability is the “category”.
VOT
3) Each Gaussian has threeparameters:
![Page 38: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/38.jpg)
Model Intro 2Misperception 3
Statistical Category Learning
1) Start with a set of randomly selected Gaussians.
2) After each input, adjust each parameter to find best description of the input.
3) Start with more Gaussians than necessarymodel doesn’t innately know how many
categories. -> for unneeded categories.
VOT VOT
![Page 39: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/39.jpg)
Model Intro 3Misperception 3
![Page 40: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/40.jpg)
Model Overgen Misperception 3
Overgeneralization • large • costly: lose phonetic distinctions…
![Page 41: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/41.jpg)
Model UndergenMisperception 3
Undergeneralization• small • not as costly: maintain distinctiveness.
![Page 42: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/42.jpg)
Model err on side of caution
To increase likelihood of successful learning:• err on the side of caution.• start with small
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 10 20 30 40 50 60
Starting
P(Su
cces
s)
2 Category Model3 Category Model
![Page 43: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/43.jpg)
Model Sparseness
Sparseness coefficient: % of space not mapped to any category.
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0 2000 4000 6000 8000 10000 12000
Training Epochs
Avg
Spa
rsity
Coe
ffic
ient
Starting
VOT
.5-1
Unmapped space
Small
![Page 44: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/44.jpg)
Model Sparseness
2
Sparseness coefficient: % of space not mapped to any category.
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0 2000 4000 6000 8000 10000 12000
Training Epochs
Avg
Spa
rsity
Coe
ffic
ient
20-40
Starting
VOT
.5-1
![Page 45: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/45.jpg)
Model Sparseness
3
Sparseness coefficient: % of space not mapped to any category.
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0 2000 4000 6000 8000 10000 12000
Training Epochs
Avg
Spa
rsity
Coe
ffic
ient
12-173-11
Starting
VOT
.5-1
20-40
![Page 46: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/46.jpg)
Model Conclusions
Small starting ’s lead to sparse category structure during infancy—much of phonetic space is unmapped.
Occasionally model leaves sparse regions at the end of learning.
1) Competition/Choice framework:• Additional competition or selection mechanisms
during processing allows categorization on the basis of incomplete information.
Model Conclusions
To avoid overgeneralization……better to start with small estimates for
![Page 47: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/47.jpg)
Model Conclusions 2
• Similar properties in terms of starting and the resulting sparseness.
2) Non-parametric models
VOT
Categories• Competitive Hebbian Learning
(Rumelhart & Zipser, 1986).• Not constrained by a particular
equation—can fill space better.
![Page 48: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/48.jpg)
Conclusions 3
Final Conclusions
Infants show graded response to within-category detail.
/b/-results suggest regions of unmapped phonetic space.
Statistical approach provides support for sparseness.• Given current learning theories, sparseness results
from optimal starting parameters.
Empirical test will require a two-alternative task.• AEM: train infants to make eye-movements in
response to stimulus identity.
![Page 49: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/49.jpg)
Future Work
Future Work
• Infants make anticipatory eye-movements along predicted trajectory, in response to stimulus identity.
• Two alternatives allows us to distinguish between category boundary and unmapped space.
![Page 50: Title Slide](https://reader035.vdocuments.us/reader035/viewer/2022062811/56815f6d550346895dce7648/html5/thumbnails/50.jpg)
Last Word
Early speech categories emerge from an interplay of
• Exquisite sensitivity to graded detail in the signal.
• Long-term sensitivity to statistics of the signal.
• Early biases to optimize the learning problem.
-60 -40 -20 0 20 40 60 80VOT
The last word