Language and Language and ThoughtThought
Lecture 2Lecture 2
WhorfWhorf
Categorical PerceptionCategorical Perception
StatisticsStatistics
1897 1907 1917 1927 1937 1947
Benjamin WhorfBenjamin Whorf
1956
1941 Died at 44.
Born Winthrop, MA
MITChem. E.
FirePreventionEngineer
Interest inLinguistics
First PaperNahuatlNahuatl Aztec
YaleE. Sapir
FieldworkArizona –Arizona –Modern Modern NahuatlNahuatl
Yale: Research FellowshipsLecturer
Introduction to Linguistic Relativity
Whorfian HypothesisWhorfian Hypothesis(Sapir-Whorf Hypothesis)(Sapir-Whorf Hypothesis)
Whorf (1956, p. 213): The categories and types
that we isolate from the world of phenomena we do not find there because they stare every observer in the face; on the contrary, the world is presented as a kaleidoscopic flux of impressions which has to be organized by our minds – and this means largely by the linguistic systems in our minds.
TerminologiesTerminologies
Linguistic Determinism (strong)Linguistic Determinism (strong) Linguistic Relativity (weak)Linguistic Relativity (weak)
Whorf as a fire Whorf as a fire prevention engineerprevention engineer
Observation: Many fires are started because of Observation: Many fires are started because of terrible language.terrible language.– E.g. carelessness around “Empty” gasoline E.g. carelessness around “Empty” gasoline
drums. drums.
“Empty” ok to flick cigarette butt
Empty-looking “Empty” ok to flick cigarette butt
Thought Language/Behavior.
Language Thought/Behavior.
Whorf’s argumentWhorf’s argument ArgumentArgument
– Languages vary in their conceptual Languages vary in their conceptual repertoire. repertoire.
– Thought is dependent on language.Thought is dependent on language.– Thus, speakers of different Thus, speakers of different
languages think differently.languages think differently. Evidence?Evidence?
– Languages vary!Languages vary!
FIX: Need separate measure of thought!
Another problem?Another problem? Suppose it is true that Suppose it is true that
1.1. Eskimos make fine discriminations of snow, and Eskimos make fine discriminations of snow, and Americans do not.Americans do not.
2.2. Eskimos have more words for snow than Eskimos have more words for snow than AmericansAmericans
– Now what’s the cause & effect?Now what’s the cause & effect?
Eskimos make fine snow discriminations BECAUSE Eskimos make fine snow discriminations BECAUSE they have lots of snow words. they have lots of snow words.
OROR
Eskimos learn to make fine snow discriminations Eskimos learn to make fine snow discriminations AND SO they have lots of snow words.AND SO they have lots of snow words.
Now how do you tease things apart???
Dissociating language Dissociating language and circumstanceand circumstance Move Americans to Vail or AspenMove Americans to Vail or Aspen
‘‘sugar’sugar’ ‘granule’‘granule’ ‘powder’‘powder’
Move Eskimos to BermudaMove Eskimos to Bermuda
Experiments in various Experiments in various domainsdomains
Some examples:Some examples: ColorColor ObjectObject Space Space TimeTime Number Number Theory of MindTheory of Mind
People who grew up without learning language.
Infants who have not learned language.
Animals who do not speak a language.
Subject PopulationSubject PopulationSpeakers of another language.
Aphasics: patients who suffered brain damage leading to language problems.
ColorColor
Ring ON Pole
Cup ON Saucer
Telephone ON Wall
Lady ON TV
Moustache ON Face
Spatial PrepositionsSpatial Prepositions
IM? UM?
AUF?
AN?
IM?
• “Where is the girl?”
– The girl is south of the umbrella.
– The girl is at the tilted side of the umbrella.
– The girl is to the left of the umbrella.
FigureReference Object
Spatial Frames of Spatial Frames of ReferenceReference
ReorientationReorientation
“Left of the blue wall”
NumberNumberPiraha:“one-two-many” counting system.
Theory of MindTheory of Mind
What is categorical What is categorical Perception?Perception?
Fre
quen
cy (
Hz)
Time (msec)
ba da ga
What is Categorical What is Categorical Perception?Perception?
Example of Categorical Perception
SpeakingSpeaking
Vocal Tract: Vocal Fold Lips(Modeled as a tube)
Vocal Fold Lips
Average Man - Length = 17.4cm
Vocal Tract ModelVocal Tract Model
F1
F2
F3
Vocal Tract = 17.4 cmSpeed of sound = 34800 cm/sec
Speed = Distance/Time = Wavelength x Frequency
Freq = Speed/Wavelength
L = 17.4cm
500Hz
1500Hz
2500Hz
λ = 4L
λ = 4L/3
λ = 4L/5
SpeakingSpeaking
c d
a
b
c da bon top of his deck
Voca
l fo
lds
Lip
s
Vocal TractVocal Tract
http://www.exploratorium.edu/exhibits/vocal_vowels/vocal_vowels.html
ee
oo
oh
ah
eh
SpectrogramSpectrogram
Time --------------
frequ
ency
http://www.haskins.yale.edu/featured/patplay.html
Pattern Playback Pattern Playback MachineMachine
SpectrogramSpectrogram
Steady StateSteady State Transitional StateTransitional State
Methods for Testing Methods for Testing Categorical PerceptionCategorical Perception
IdentificationIdentification– Randomly play the audio clips and Randomly play the audio clips and
asked to identify the phonemeasked to identify the phoneme DiscriminationDiscrimination
– Randomly play pairs and asked to make Randomly play pairs and asked to make Same-different JudgmentSame-different Judgment Same pairsSame pairs Different pairsDifferent pairs
IdentificationIdentification
IdentificationIdentification– Randomly play the audio clips and asked to Randomly play the audio clips and asked to
identify the phonemeidentify the phoneme If there is CP, what should the graph If there is CP, what should the graph
look like? look like? – X-axis stimuli arranged in a continuum X-axis stimuli arranged in a continuum with with
very small incremental differencevery small incremental difference between between the stimulithe stimuli
– Y-axis % Identification as the tested Y-axis % Identification as the tested categorycategory
IdentificationIdentification(idealized results)(idealized results)
% Iden
tifica
tion a
s C
ate
gory
X
0
20
40
60
80
100
2 3 4 5 6
Stimulus #
1 7
Fre
quen
cy (
Hz)
Time (msec)
ba da ga
What is Categorical What is Categorical Perception?Perception?
0
10
20
30
40
50
60
70
80
90
100
-9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9
Stimulus ID
% I
den
tifi
cati
on
ba
da
ga
Categorical PerceptionCategorical Perception(Idealized Data)(Idealized Data)
Methods for Testing Methods for Testing Categorical PerceptionCategorical Perception
IdentificationIdentification– Randomly play the audio clips and Randomly play the audio clips and
asked to identify the phonemeasked to identify the phoneme DiscriminationDiscrimination
– Randomly play pairs and asked to make Randomly play pairs and asked to make Same-different JudgmentSame-different Judgment Same pairsSame pairs Different pairsDifferent pairs
Discrimination StudyDiscrimination Study
Last example of ba/da/ga varied Last example of ba/da/ga varied transitional state (up, down of F2).transitional state (up, down of F2).
In this example, Varying Voice Onset In this example, Varying Voice Onset Time.Time.
Voice Onset Time Voice Onset Time (VOT)(VOT)
VOT: VOT: time between consonant release and vocal cord vibration
[p][b]
So what is the difference in VOT between So what is the difference in VOT between VOICELESS [b] and VOICED [p]?VOICELESS [b] and VOICED [p]?– SHORT VOT SHORT VOT voiced voiced– LONG VOT LONG VOT voiced voiced
Voice Onset Time Voice Onset Time (VOT)(VOT)
Short VOT = ?Short VOT = ? Long VOT = ?Long VOT = ? Which one is /di/ Which one is /di/
and which one and which one is /ti/?is /ti/?
di ti
Discrimination StudyDiscrimination Study
Same/Different?0ms 60ms
Same/Different?0ms 10ms
Same/Different?40ms 40ms
Why is this pair difficult?
(i) Acoustically similar?
(ii) Same Category?
DiscriminationDiscrimination
Same/Different0ms 60ms
Same/Different0ms 10ms
Same/Different40ms 40ms
A More Systematic Test
0ms
20ms
40ms
20ms
40ms
60ms
D T
D
T T
D
Within-Category Discrimination is Hard
Categorical PerceptionCategorical Perception(Idealized Discrimination Data)(Idealized Discrimination Data)
% C
orr
ect
Dis
crim
inati
on
0
20
40
60
80
100
1-2 2-3 3-4 4-5 5-6
Pairs by VOT
Question 1Question 1
Is speech perception innate?Is speech perception innate?– Do newborns have categorical Do newborns have categorical
perception?perception? If CP requires exposure to language If CP requires exposure to language
(e.g., knowledge of minimal pairs in (e.g., knowledge of minimal pairs in one’s language), then NO.one’s language), then NO.
If CP is innate, then YES.If CP is innate, then YES.
– How do we test newborns?How do we test newborns?
High Amplitude High Amplitude Sucking ProcedureSucking Procedure
Infant given a pacifier that Infant given a pacifier that measures sucking ratemeasures sucking rate
HabituationHabituation – Infant sucks – Infant sucks to hear sound (e.g. ba) until to hear sound (e.g. ba) until bored.bored.
TestTest – Play sound (e.g., ba – Play sound (e.g., ba or pa). Is there or pa). Is there dishabituationdishabituation??– Infants will suck to hear sound Infants will suck to hear sound
if the sound is no longer if the sound is no longer boring.boring.http://psych.rice.edu/mmtbn/language/sPerception/video/sucking_h.movhttp://psych.rice.edu/mmtbn/language/sPerception/video/sucking_h.mov
http://www.learner.org/vod/vod_window.html?pid=1620 (2:50 min. into videoclip) http://www.learner.org/vod/vod_window.html?pid=1620 (2:50 min. into videoclip)
Stimuli for Eimas et. Stimuli for Eimas et. al’s Studyal’s Study BA vs. PABA vs. PA Vary Voice Onset Time (VOT): time btw Vary Voice Onset Time (VOT): time btw
consonant release and vocal cord consonant release and vocal cord vibrationvibration
VOT in milliseconds
0 20 40 60 80
PABA
PredictionsPredictions
Between Between CategoryCategory
BABA11-PA-PA
Within Within CategoryCategory
BABA11-BA-BA22
Within Within CategoryCategory
ControlControl
BABA11-BA-BA11
Innate Innate Categorical Categorical PerceptionPerception
dishabituatdishabituatee
remain remain habituatedhabituated
remain remain habituatedhabituated
Untuned Untuned SensitivitySensitivity
dishabituatdishabituatee
dishabituatdishabituatee
remain remain habituatedhabituated
InsensitiveInsensitive remain remain habituatedhabituated
remain remain habituatedhabituated
remain remain habituatedhabituated
BA1 = VOT 20ms; BA2 = VOT 0ms; PA = VOT 40ms
Results for Eimas et. al’s Results for Eimas et. al’s StudyStudy
ME
AN
NU
MB
ER
OF
SU
CK
ING
RE
SP
ON
SE
dishab
no
no
Question 1 AnswerQuestion 1 Answer
Q1: Is Speech Perception Innate?Q1: Is Speech Perception Innate? Many other studies since tested:Many other studies since tested:
– Infants (Neonates) on other contrasts.Infants (Neonates) on other contrasts. Consensus: Yes to Innate Q. Consensus: Yes to Innate Q.
– Infants do not discriminate all physically Infants do not discriminate all physically equal acoustic difference; they show equal acoustic difference; they show heightened sensitivity to those that are heightened sensitivity to those that are important for language.important for language.
– BUT… there is language-specific fine-tuning…BUT… there is language-specific fine-tuning…
Provisional ConclusionsProvisional Conclusions
Speech Perception makes use of Speech Perception makes use of some auditory mechanisms which some auditory mechanisms which evolved prior to languageevolved prior to language– These abilities are innateThese abilities are innate
Becoming a Native Becoming a Native ListenerListener Languages differ in their Languages differ in their
inventories of phonemes.inventories of phonemes.
What develops or changes in our What develops or changes in our speech perception abilities?speech perception abilities?
Language Specific Fine Tuning
Japanese vs. EnglishJapanese vs. English(Miyawaki et al. 1975)(Miyawaki et al. 1975)
RA
LA
AMERICANS
Dental Stop – tip of tongue touching back of front teeth
Retroflex Stop – tongue curled so tip is behind alveolar ridge
Hindi (spoken in India)
unvoiced unaspirated retroflex vs. dental stop
(English /t/ is typically somewhere between the two)
Can you hear the Can you hear the difference?difference?
Hindi
dental
retroflex
Uvular – tongue is raised against the velum
Velar – tongue is raised behind the velum
Salish (Native North American language):
glotalized voiceless stops
(they are actually ejectives - ejective is produced by obstructing the airflow by raising the back of the tongue against or behind the velum)
When does changes in When does changes in sensitivity occur?sensitivity occur?
Infancy
Adulthood
… And testing method?
Conditioned Head-TurnConditioned Head-Turn ConditioningConditioning
– Child hears a Child hears a string of sounds.string of sounds.
– Conditioned to Conditioned to turn head when turn head when detects a change detects a change (e.g., bell (e.g., bell whistle) with whistle) with rewardreward
TestTest– Speech sounds Speech sounds
(e.g., da, da, da, (e.g., da, da, da, da, ta,…)da, ta,…)
– Does the child Does the child turn his or her turn his or her head with head with changed from da changed from da to ta?to ta?
Werker: http://www.learner.org/vod/vod_window.html?pid=1630Kuhl: http://www.learner.org/vod/vod_window.html?pid=1631
When does Change When does Change Occur?Occur?
6-8m 8-10m 10-12m 11-12m
6-8m 8-10m 10-12m 11-12m
What is changing? What is changing? Two contrasting views: 1 or 2?Two contrasting views: 1 or 2?
Maintenance or LossMaintenance or Loss– If you don’t use it, you lose it.If you don’t use it, you lose it.– Parallel aspects of early visual Parallel aspects of early visual
development.development. Functional ReorganizationFunctional Reorganization
– Existing architecture reorganized for Existing architecture reorganized for higher level of processing.higher level of processing.
What is changing? What is changing? Two contrasting viewsTwo contrasting views
What is changing? What is changing? 1. Maintenance or Loss View1. Maintenance or Loss View
Structure-changingNon-native boundariesdisappear.Resulting in native language phonetics
Phonetics
Acoustics
Phonology
What is changing? What is changing? 2. Functional Reorganization2. Functional Reorganization
Structure-buildingNative languagephonemesbuilt fromuniversal phones
Phonetics
Acoustics
Phonology
Which view?Which view? Werker (1997) noted some problems for Werker (1997) noted some problems for
the the maintenance or loss viewmaintenance or loss view..1. Many of the uncategorized sounds do appear in the 1. Many of the uncategorized sounds do appear in the
native language but just are not meaningful (e.g., as native language but just are not meaningful (e.g., as allophones), and speakers can be made aware of the allophones), and speakers can be made aware of the difference.difference.
Example:Example:– /p/ is only /p/ is only aspiratedaspirated in “pin” and not “spin’ in “pin” and not “spin’– /p/ in “pin” and “spin” are /p/ in “pin” and “spin” are allophonesallophones in English in English– But could be But could be minimal pairsminimal pairs in some other in some other
languages.languages.
Which view?Which view? Werker (1997) noted some problems for the Werker (1997) noted some problems for the
maintenance or loss viewmaintenance or loss view..1. Many of the uncategorized sounds do appear in the native 1. Many of the uncategorized sounds do appear in the native
language but just are not meaningful (e.g., as allophones), and language but just are not meaningful (e.g., as allophones), and speakers can be made aware of the difference.speakers can be made aware of the difference.
2. Children who fail to show categorical perception for non-native 2. Children who fail to show categorical perception for non-native phonemes can acquire a new language without an accent.phonemes can acquire a new language without an accent.
3. Adults can be trained to make non-native distinctions.3. Adults can be trained to make non-native distinctions.
4. Perceptual distinction is readily available for non-linguistic 4. Perceptual distinction is readily available for non-linguistic tasks.tasks.
Language Specific Fine Tuning
Which model?Which model?
Werker (1997): The evidence that poses Werker (1997): The evidence that poses problems for maintenance or loss view problems for maintenance or loss view supports the supports the functional reorganization functional reorganization viewview..
I.e., the view that:I.e., the view that:– Those perceptual categories which are Those perceptual categories which are
meaningful in the native language become meaningful in the native language become speech categories.speech categories.
– The remainder are perceived but not The remainder are perceived but not recruited in speech perception.recruited in speech perception.
Language Specific Fine Tuning
StatisticsStatistics
OutlineOutline Stats Terms SimplifiedStats Terms Simplified
– t-testst-tests– ANOVAs, Main effects and ANOVAs, Main effects and
InteractionsInteractions– Regressions, CorrelationsRegressions, Correlations
T-tests and ANOVAsT-tests and ANOVAs
T-tests: Compare 2 means.T-tests: Compare 2 means. ANOVA (Analysis of Variance): ANOVA (Analysis of Variance):
Compare multiple meansCompare multiple means– Yields significance of main or Yields significance of main or
interaction effectsinteraction effects
Hypothetical ExperimentHypothetical Experiment(Example of Main & Interactions (Example of Main & Interactions Effects)Effects) Dependent Measure: Number of Dependent Measure: Number of
GirlfriendsGirlfriends Independent Measure: Independent Measure:
– Wealth of bachelors according to Income Wealth of bachelors according to Income (Rich, Poor)(Rich, Poor)
– Looks of same bachelors according to Looks of same bachelors according to Oprah Oprah
(Handsome, Ugly)(Handsome, Ugly)
Design 2 x 2Design 2 x 2
RichRich PoorPoor
Hand-someHand-some
UglyUgly
# of GF
# of GF
# of GF
# of GF
Hypothetical Hypothetical ExperimentExperiment(Example of ANOVAs F1 vs. (Example of ANOVAs F1 vs. F2)F2) Is a female model more attractive Is a female model more attractive
in short or long skirt?in short or long skirt?– Model pictured in 10 different short Model pictured in 10 different short
skirts and 10 different long skirtsskirts and 10 different long skirts– 30 Males rated the model’s 30 Males rated the model’s
attractiveness in each skirt (1 = not attractiveness in each skirt (1 = not attractive to 7 = extremely attractive to 7 = extremely attractive)attractive)
Hypothetical Hypothetical ExperimentExperiment(Example of ANOVAs F1 vs. (Example of ANOVAs F1 vs. F2)F2) F1: Subject AnalysisF1: Subject Analysis
– Comparing subjectsComparing subjects– Averaging across items for each subjectAveraging across items for each subject
F2: Items AnalysisF2: Items Analysis– Comparing itemsComparing items– Averaging across subjects for each itemAveraging across subjects for each item
Hypothetical Hypothetical ExperimentExperiment(Example of ANOVAs F1 vs. (Example of ANOVAs F1 vs. F2)F2) F1: Subject AnalysisF1: Subject Analysis
F2: Items AnalysisF2: Items Analysis
Short LongFrederick H. 5.2 3.1Hef 7.0 7.0Rudy G. 5.2 4.9Bill C. 6.9 1.3
RatingShort Skirt 1 4.5Short Skirt 2 5.3…Long Skirt 1 6.7Long Skirt 2 3.5…