Faces - Life Sciences Outreach at Harvard University

advertisement
Prosopagnosia and
Face-Specific Mechanisms
Brad Duchaine
Vision Sciences Laboratory
Harvard University
http://www.faceblind.org
The nature of cognitive specializations
Domain-specific—mechanisms specialized for particular
types of content. e.g.-speech, faces.
Domain-general—mechanisms specialized for particular
processing tasks. e.g.-recognition, reasoning.
Prosopagnosia: Acquired & Developmental
Long considered an extremely rare condition
www.faceblind.org contacted by 400 prosopagnosics
Majority are developmental
Living with Prosopagnosia
“While traveling, I had a stopover at O'Hare and I was approached
by a stranger in the lounge area. It took 10-15 seconds of casual
conversation before realizing who it was. It was my brother.”
“I think prosopagnosia has worsened my current depression, if it’s not
the root cause of it. This condition always affects my ability to form
normal social links to others. I prefer to be a recluse because I can’t
confidently function any other way. My avoidance of people to
interact with socially is nearly phobic.”
Explanation in prosopagnosia
Face-Specific Mechanism
Within-Class Mechanism
Configural Processing Mechanism
Non-Decomposable Mechanism
Curvature Mechanism
Rapid Expertise Mechanism
Extended Expertise Mechanism
Case History: Developmental Prosopagnosic
Edward
•53-year-old right-handed man.
•Ph.D.s in physics and theology.
•Aware of problems as a child.
•Knows of no head trauma.
•MRI showed no abnormalities.
•General face processing impairment.
•Reports no difficulties with object recognition.
•No navigational difficulties.
Case History: Acquired Prosopagnosic
LJ
•16-year-old high school student.
•Incident at school dance.
•Knows of no head trauma.
•Incidents over last few years.
•Feels lonely in world devoid of facial information.
•Impairment beginning with face detection.
Case History: Acquired Prosopagnosic
LJ
•16-year-old high school student.
•Incident at school dance.
•Knows of no head trauma.
•Incidents over last few years.
•Feels lonely in world devoid of facial information.
•Impairment beginning with face detection.
•Reports normal object recognition.
•Navigational skills are deteriorating.
•CAT, MRI, and EEG are normal.
Edward’s Face Recognition
Famous Face Recognition
25 faces
Controls
21.6 (2.5)
Edward
3
Duchaine & Nakayama (2004) Neuron
LJ’s Face Recognition
Famous Face Recognition
32 faces
Controls
28.8 (3.2)
LJ
1
fMRI procedure
Localizer: Block-design with 5 stimulus classes.
Faces
Scenes
Bodies
Objects
Scrambled
FFA: Faces - Objects
Controls
Edward
LJ
PPA: Places - Objects
Controls
Edward
LJ
EBA: Bodies - Objects
Control
Edward
LJ
1.4
Repetition decrease in FFA
% Signal Change to Face 2
1.2
1
0.8
0.6
0.4
0.2
Different Face
0
Same Face
% Signal Change Same / % Signal Change Diff
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Explanations for prosopagnosia
A mechanism isn’t working, but what is its domain?
Proposed Domains
Predicted Impairments
•Upright faces (Farah, 1996)
•Configural Information
(Levine & Calvanio, 1989)
•Within-class recognition
(Damasio et al., 1982)
•Non-decomposable objects
(Farah, 1991)
•Curved surfaces
(Kosslyn et al., 1995; Laeng & Caviness, 2001)
•Rapid Expert Classes
(Gauthier et al., 1999)
•Extended Expert Classes
(Carey & Diamond, 1986; Carey, 1992)
? ?
Within-Class Mechanism
Mechanism for recognizing individual items.
(Damasio et al., 1982)
Within-Class Mechanism
Faces
Sunglasses
Horses
Guns
Cars
Houses
Tools
Landscapes
2.0
0.0
-2.0
-4.0
Edward A' z score
LJ A' z score
-6.0
-8.0
-10.0
-12.0
-14.0
-16.0
2.0
0.0
-2.0
-4.0
Edward A' z score
LJ A' z score
-6.0
-8.0
-10.0
-12.0
-14.0
-16.0
Faces: Individual Scores
1.00
0.90
0.80
A’
0.70
0.60
0.50
1.0
0.0
-1.0
-2.0
-3.0
-4.0
-5.0
-6.0
Response time
z scores
Non-Decomposable Mechanism
Mechanism for representing objects
difficult to decompose into parts (Farah, 1991)
May require holistic strategy.
Hypothesis not explicit about what objects qualify.
Curvature Mechanism
Mechanism for representing curved surfaces
(Kosslyn et al., 1995; Laeng & Caviness, 2001).
Laeng & Caviness (2001): Dogs, glasses, and cars.
Configural Processing Mechanism
Upright faces activate configural processing.
Face-specific?
or
General purpose?
Domain-general mechanism for configural processing
(Levine & Calvanio, 1989)
Parts
Spacing
Parts
Spacing
100
100
Spacing Changes
% Correct
Spacing Changes
% Correct
90
80
70
90
80
70
60
60
50
50
50
60
70
80
% Correct
Part Changes
90
100
50
60
70
80
% Correct
Part Changes
90
100
Configural Processing Mechanism
Demonstrates face-specific impairment.
Normal House spacing inconsistent with:
Configural processing hypothesis
Non-decomposable hypothesis
Upright vs Inverted
Non-decomposable hypothesis
Curvature hypothesis
Face Matching: Upright versus Inverted
Face Matching: Upright versus Inverted
100
% Correct
90
80
70
60
50
Controls
Edward
LJ
Face Matching: Upright versus Inverted
Edward processes upright and inverted faces similarly.
No special processing for upright faces.
LJ performs worse with upright faces than inverted faces.
Upright representations sent to “black hole”.
Normal inverted performance inconsistent with:
Non-decomposable hypothesis
Curvature hypothesis
Rapid Expertise Mechanism
Mechanism for recognition of items from expert categories
(Gauthier et al., 1997, 1999)
Rapid Expertise Mechanism
Edward not a face expert after 53 years.
LJ has lost his expertise with faces.
Rapid Expert Mechanism
Eight sessions of training (Gauthier & Tarr, 2002).
Sessions 1-4: Between 495-680 Test Trials
Sessions 5-8: 180 Test Trials
Naming
Verification
Yes
Triz
No
T (for Triz)
100
90
Naming
Naming
80
Scaled % Correct
100%
70
60
50
100%
40
30
100%
20
10
5 Greebles 10 Greebles 15 Greebles 20 Greebles
0
1
2
3
4
5
Session
6
7
8
100
Naming
90
80
Scaled % Correct
100%
70
3000
60
2500
50
100%
2000
40
1500
1000
30
100%
500
20
0
1
2
3
4
5
6
7
8
10
5 Greebles 10 Greebles 15 Greebles 20 Greebles
0
1
2
3
4
5
Session
6
7
8
100
90
Individual
Verification
80
Scaled % Correct
100%
70
60
50
100%
40
30
100%
20
10
5 Greebles 10 Greebles 15 Greebles 20 Greebles
0
1
2
3
4
5
Session
6
7
8
100
90
Individual
Verification
80
Scaled % Correct
100%
70
3000
60
2500
50
100%
2000
40
1500
30
1000
100%
500
20
0
1
10
2
3
4
5
6
7
8
5 Greebles 10 Greebles 15 Greebles 20 Greebles
0
1
2
3
4
5
Session
6
7
8
100
90
Family
Verification
% Correct
80
70
60
50
40
1
2
3
4
5
Session
6
7
8
100
90
2500
Family
Verification
% Correct
80
2000
70
1500
1000
60
500
0
50
1
3
2
5
4
7
6
8
40
1
2
3
4
5
Session
6
7
8
Rapid Expertise Mechanism
Results are inconsistent with hypothesis
Greeble results are inconsistent with:
•Within-class hypothesis
(Damasio et al., 1982)
•Curvature hypothesis
(Kosslyn et al., 1995; Laeng & Caviness, 2001)
Extended Expertise Mechanism
Mechanism for recognition of items from expert categories
(Diamond & Carey, 1986)
Extended Expertise Mechanism
Extended Expertise Mechanism
Extended Expertise Mechanism
100
Controls
Edward
LJ
% Correct
90
80
70
60
50
Faces
Bodies
Alternative
Explanation
Within-Class
Old-New
Tests
PartSpacing
X
X
Configural Processing
X
Curved Surfaces
X
Extended Expertise
X
Body
Greeble
Training Matching
X
X
X
X
X
X
X
Non-Decomposable
Rapid Expertise
Inverted
Matching
X
X
X
X
Explanation in prosopagnosia
Face-Specific Mechanism
Configural Processing Mechanism
Within-Class Mechanism
Non-Decomposable Mechanism
Curvature Mechanism
Rapid Expertise Mechanism
Extended Expertise Mechanism
General-purpose
Face-specific
“Richard Nixon?”
“Richard Nixon”
Mr. CK:
Agnosia without Prosopagnosia
CK can
recognize
faces
CK cannot recognize objects
Inverted faces
(Moscovitch et al., 1997)
Faces, Domains, and Natural Categories
Results strongly support existence of what have
been called domain-specific mechanisms
Domain-specificity and natural categories
Specialization for a natural category
Developmental Inferences
Edward never developed face-specific mechanisms.
His behavioral and fMRI results show that he
developed normal object recognition mechanisms.
Functionally dissociable and developmentally dissociable.
Inferences from Edward’s case
Mature
Mechanisms
Core
Mechanisms
Faces
General Objects
Specific
Developmental Mechanisms
Places
Poodle face palinopsia
Poodle face palinopsia
Expertise Criterion:
Comparable Verification RTs
3000
Edward individual
Edward family
LJ individual
LJ family
Response Time (msec)
2500
2000
1500
1000
500
0
1
2
3
4
5
Session
6
7
8
Expertise Criterion:
Comparable Verification RTs
3000
Edward individual
Edward family
LJ individual
LJ family
Response Time (msec)
2500
2000
1500
1000
500
0
1
2
3
4
5
Session
6
7
8
FFA: Faces – Objects
Right
Control
Right
FFA
Edward
PPA: Scenes - Objects
Control
Edward
Right
Right
PPA
EBA: Bodies - Objects
Control
EBA
Edward
3.0
Face
Object
% signal change
2.5
2.0
1.5
1.0
Right
0.5
0.0
1
2
3
4
5
6
7
8
6
7
8
6
7
8
-0.5
TR (2sec)
3.0
Face
Object
% signal change
2.0
1.0
0.0
1
2
3
4
5
-1.0
-2.0
-3.0
-4.0
TR (2 sec)
3.0
Face
Object
2.5
% signal change
2.0
1.5
1.0
0.5
0.0
1
2
3
4
5
-0.5
-1.0
TR (2 sec)
Edward
Edward
3.0
Control
Scene
Object
3.0
Scene
Object
2.5
% signal change
PPA
% signal change
2.5
2.0
1.5
1.0
2.0
1.5
1.0
0.5
0.5
0.0
1
2
3
4
5
6
7
8
0.0
-0.5
1
2
3
TR (2 sec)
3.0
Body
Object
3.0
6
7
8
6
7
8
Body
Object
2.5
2.0
% signal change
% signal change
5
TR (2sec)
2.5
EBA
4
1.5
1.0
0.5
2.0
1.5
1.0
0.5
0.0
0.0
1
2
3
4
5
6
7
1
8
2
3
4
5
-0.5
-0.5
TR (2 sec)
TR (2sec)
Imaging Results
Structural MRI showed no obvious abnormalities.
Cambridge Test of Face Memory
Examples
Test item with
identical images
Test item with
novel images
Test item with
novel images
with noise
Duchaine & Nakayama (under review) Neuropsychologia
72
66
60
Controls
Edward
54
Cumulative Score
LJ
48
42
36
30
24
18
12
6
0
0
6
12
Introduction
18
24
30
36
42
Novel images
Item Number
48
54
60
66
72
Novel images with noise
Future Directions
Developmental Prosopagnosia
Neural basis
Dissecting face processing
Etiology
Genetic basis of face perception
Autism & prosopagnosia
Plasticity/Therapies
Developmental Topographagnosia?
Psychophysics
Face recognition test
Training with inverted faces
Activation of face recognition
Rapid Expertise Hypothesis
100
100
Ed
Tina
90
Gayle
Frank
Maureen
Dana
90
80
Joe
% Correct (n = 25)
% Correct
70
80
70
60
50
Ed
Tina
Gayle
Frank
Maureen
Dana
Joe
40
30
60
20
10
50
0
Upright
Inverted
Sequential Face Matching
Famous Faces
Mr. CK:
Agnosia without Prosopagnosia
CK can
recognize
faces
CK cannot recognize objects
Inverted faces
(Moscovitch et al., 1997)
Z values for
NM’s scores
-10 -8 -6 -4 -2 0
2 4
6
8 10
Face OIT
Faces #1
Faces #2
Warrington
Famous Faces
Profiles
Accuracy
Response Time
Z values for
NM’s scores
-10 -8 -6 -4 -2 0
2 4
6
8 10
Face OIT
Faces #1
Faces #2
Warrington
Accuracy
Famous Faces
Response Time
Profiles
Emotion Hexagon
Eyes Test
Emotion Matching
Emotional Intensity
-10 -8 -6 -4 -2 0
2 4
6
8 10
1.00
Controls
DPs
0.95
0.90
0.85
Mean A'
0.80
0.75
0.70
0.65
0.60
0.55
0.50
Faces 1
Faces 2
Cars
Guns
Horses
Houses
Scenes
Tools
Configural Processing Hypothesis
Predicts that Edward will be impaired.
30
# Correct
Controls
Edward
20
10
0
Gestalt
Completion
Concealed
Words
Snowy
Pictures
Non-selective response to faces vs. objects
Object - Fixation
Face - Fixation
+
-
Problems with RT criterion
RT criterion is dependent
on proportions of different
trial types.
3
1
2
It says nothing about
proficiency.
3
Past results show that RT
criterion does not work.
(Gauthier et al., 1998)
Greeble Transfer or Task Learning?
(Gauthier et al., 1998)
Putative holistic/configural effects are not face-like
1. No evidence of a large inversion effect.
2. Part-whole difference between experts & novices.
3. Part-in-original vs. part-in-whole effects:
--Gauthier et al. (1998)—No effects.
--Gauthier et al. (2002)—Two effects in opposite directions.
4. Composite effect:
--Gauthier et al. (1997)—No effect.
--Gauthier et al. (1998)—No effect.
--Gauthier et al. (2002)—No effect.
Low- and Mid-Level Vision
Are Edward’s face processing impairments due to problems
with low-level or mid-level vision?
Visual acuity
Near vision
Far vision
Normal
Corrected-to-normal
Pelli-Robson Contrast Sensitivity Test
Normal
Birmingham Object Recognition Battery
Length match
Size match
Orientation match
Position of gap match
Normal
Normal
Normal
Normal
Paradigmatic Examples:
Language and Face Recognition
Chomsky—Rules and Representations
Fodor—Modularity of Mind
Pinker—Language Instinct
Cowie—What’s within?
Bates et al—Rethinking Innateness
Language is a difficult test case.
Face recognition more tractable ability.
Remaining Hypotheses
Face-specific hypothesis
(Farah, 1996; Moscovitch et al., 1997)
Extended expertise hypothesis
(Diamond & Carey, 1986; Carey, 1992)
? ?
Other than faces, no examples of classes
for which everyone has expertise.
Little evidence that expertise leads
to face-like processing.
Unclear how to test either hypothesis with Edward or LJ.
Edward:
Normal inversion effect for face detection
Low Density
Upright
Inverted
High Density
100
100
Low Density
90
90
80
80
Controls Upright
Edward
ControlsUpright
Inverted
Controls
Inverted
Edward Inverted
Edward Inverted
70
70
60
60
50
50
50
50
60
60
70
70
80
80
High
High Density
Density
90
90
100
100
Remaining Hypotheses: Double Dissociation
RM: Car expert
(Sergent & Signoret, 1992)
Mr. CK: Airplane & toy soldier expert
(Moscovitch et al., 1997)
Remaining Hypotheses: Critical Period
Face
Configural
Face configural processing does not develop
without input during the first months of life.
(Le Grand et al., 2001)
No critical period for non-face expertise.
Remaining Hypotheses
Face-specific Hypothesis
(Farah, 1996; Moscovitch et al., 1997)
Extended Expertise Hypothesis
(Carey & Diamond, 1986; Carey, 1992)
? ?
Download