Transcript Document

What made you respond face (or
word)?
• Something in your brain made you decide face or
word.
• Can we determine where this decision is made?
• Related domain: Motion Direction Discrimination
and area MT
What determines the percept and
the response?
• Observe a correlation
between motion direction
judgements and activity of
cells in area MT. (Britten et
al. 1996, Gold & Shadlen,
2000)
• If neurons that responded to
leftward motion were highly
active, the monkey chose
'left' as the decision.
From Schall, 1999
Evoked Response Potential (ERP)
and Face Stimuli
• N170: negative-going
potential at 170 ms
• Largest over the right
parietal lobe, also on
the left parietal lobe.
From Tanaka and Curran (2001)
N170 Properties:
• Faces produce the largest amplitude.
• Strong evidence of expertise: Bird experts have larger
N170's to pictures of birds than pictures of dogs. Dog
experts show the reverse. (Tanaka & Curran 2001).
• Mainly perceptually based: prior exposure of a face does
not produce large changes in the N170 for subsequent
presentations (Rik Henson, AIC 2003).
• Scalp distribution and latency suggest that the N170
component reflects the perceptual processing of complex
visual stimuli.
A Thought Question:
• What was going on in your perceptual regions
when you thought you saw a face or a word?
• Could we capture the current state (at least
indirectly) with the N170 component?
• Would the N170 be larger when you thought you
saw a face?
Central Question:
• Can we relate the size of the N170 to the
response in the noise-alone condition?
• Will it be larger when subjects think they see a
face?
An Experiment
• Show Faces and Words Embedded in
Noise:
High
Contrast
Faces
Low
Contrast
Faces
Noise
Alone
Low
Contrast
Words
High
Contrast
Words
Methods
• Face, word and noise-alone trials were presented in random
order.
• Ten naïve participants
– 120 trials per condition per subject
• On each trial: Did you see a face or a word?
– Subjects were told that a stimulus appeared on each trial, and that they
should guess if they were unsure.
EEG Recording Sites
Methods continued
• Analyze the data according to the subject’s responses on the
noise-alone trials.
• Incredibly important point: The noise was the same across
ALL trials and stimuli. Physically the same. Not just
identically distributed, but identical. There was only one
noise field for the entire experiment.
Together, these procedures hold the physical stimulus constant
on noise-alone trials.
Right Temporal Lobe (T6)
Amplitude (V)
4
2
0
-2
-4
High Contrast Word
High Contrast Face
-6
0
100
200
300
time (ms)
400
500
600
Central Question:
Will we see a larger N170 to the noisealone stimulus when subjects think
they see a face as opposed to a word?
Right Temporal Lobe (T6)
4
* t(9) = 2.74, p = .023,
two-tailed
Amplitude (V)
*
2
0
-2
-4
Responded 'Word' To Noise-Alone Trial
Responded 'Face' To Noise-Alone Trial
High Contrast Face
-6
0
100
200
300
time (ms)
400
500
600
Main Result:
• On noise-alone trials: Larger N170 when observers report
seeing a face than when report seeing a word.
• Occurs in 9 of the 10 subjects.
• No other differences in any other channel at the P100, N170
or P300 components.
• Unlikely to just reflect activity for an already-made decision.
• Relates activity in the perceptual processing areas to the
behavioral response. Greater activity in the N170 neurons is
associated with ‘face’ responses to the noise-alone stimulus.
One interpretation: Greater activity in the face processing region
biases the response towards a ‘face’ response.
Alternative Explanations for this
Greater Activity
• Attention to different spatial frequencies or face-like
features in the noise
– Unlikely to see rapid changes in spatial frequency tuning in a
mixed design.
– No P100 differences that might be associated with changes in
activity in different spatial frequency channels.
• One obvious exception: Prior trial priming
– Seeing a face on the previous trial may leave residual activity in
the face neurons or make subjects look for face-like features in the
noise.
Prior Trial Priming
Noise-Alone Trials Conditioned on
Prior Trial and Response
4
Amplitude
(V)
2
0
-2
Responded
'Face'
Responded
'Word'
Face on
Prior Trial
-4
Word on
Prior Trial
0
100
200
time (ms)
300
Third Possibility:
Stochastic Activity
• In the domain of binocular rivalry, Blake and Logothetis
introduced the idea of a process that involved stochastic
activity in perceptual regions, which could bias the response
toward one percept or the other at different points in time.
(Blake & Logothetis 2002).
• A similar process could be at work in the face processing
neurons:
When activity is high in the N170 neurons due to random
stochastic fluctuations, the observer may be biased to
respond ‘face’ on that trial.
Implications for Internal Noise
• Internal noise does not just limit performance or decrease
calculation efficiency, but also operates in feature space to
bias the response toward one alternative or another.
• Manipulations such as varying the power of the noise or
the stimulus pairs being compared may help constrain
models of internal noise.
• Supports trial-by-trial variability in parameters:
– Starting point and drift-rate variability in Ratcliff’s Diffusion
model.
– These sources of variability take on a perceptually-based
interpretation. Link with perceptual brain areas.
The Important Stuff:
RightNoise-Alone
Temporal LobeTrials:
(T6)
For
* t(9) = 2.74,
p = .023
Amplitude (V)
4
2
0
*
-2
High
Contrast
Faces
Low
Contrast
Faces
Noise
Alone
Low
Contrast
Words
High
Contrast
Words
-4
Responded 'Word' T o Noise-Alone T rial
Responded 'Face' T o Noise-Alone T rial
0
100
200
300
time (ms)
400
500
600
ERP and Faces
• Intracranial recordings reveal N200 at sites
in IT/fusiform gyrus
From Allison, Puce, Spencer, and
McCarthy (1999)
Which Stimuli Evoke an
N170/N200?
Any face or face-like
visual stimulus
N170 Properties:
Top-down Influences
• Could be just a face-detection system (Bentin, et. al. 1996)
• No effect of task demands such as selective attention
to faces vs. objects. (Caquil, Edmonds, & Taylor, 2000)
• No effect of familiarity of the face (Bentin & Deouell, 2000)
• fMRI - IT active anytime there’s a face. (Gauthier et al.
1999)
• However, IT also active while imagining a face.
(O’Craven & Kanwisher 2000)
Contextual Influences
On N170
Bentin et al (2002)
• Block 1:
– Stimulus Set A
• Block 2:
– Set B (experimental)
– Set C (control)
– Set D (all; targets)
• Block 3:
– Set A again (all)
Contextual Influences
Bentin et al (2002)
On N170• Face context elicits N170 to
schematic eyes
• Once the stimulus has been
interpreted as containing facelike features, a stronger N170
is produced.
N170 Summary
• Represents activity of face-selective neurons
most likely in area IT.
• Magnitude varies with the degree to which
physiognomic information is perceived in the
image.
• Is the N170 related to the response in the
noise-alone condition?
• If so, tie the activity in face-selective cells to
the percept and the response.
Would we get an N170 to the
noise-alone trials at all?
• Previous work showed only a weak N170 to random noise
stimuli.
• Pilot study: Blocked presentations of faces and words.
• Result: when looking for faces, get a larger N170 to noisealone trials than when looking for words.
• Replicates Bentin et al., extends to images that contain no
facial features. We see top-down effects on the N170 even
with no face-like features.
Face Condition :
Male or female?
Word condition:
Honesty or Trust?
Experiment 1 Results - T5
Grand Average for
T5
8
6
4
High-Contrast Face
Low-Contrast Face
Zero-Contrast Face
High-Contrast Word
Low-Contrast Word
Zero-Contrast Word
2
Voltage
(microvolts)
0
-2
-4
-6
-8
-10
-12
-100
0
100
200
300
Time (ms)
400
500
600
Experiment 1 Results - T5
Grand Average for T5
Zero-Contrast Word
Zero-Contrast Face
6
5
4
*t(8) = 2.62, p = .031
Voltage
(microvolts)
3
2
1
0
-1
-2
-3
-100
0
100
*
200
300
Time (ms)
400
500
600
Neural Processing of Face
Stimuli
• Cells in inferotemporal cortex (IT) are
known to respond selectively to faces
– (single cell recording, fMRI)
• Slight right hemisphere dominance
Experiment 1 Results - T6
Grand Average for T6
Zero-Contrast Word
Zero-Contrast Face
5
4
3
Voltage
(microvolts)
* t(8) = 2.35, p = .047
2
1
0
-1
-2
-3
-100
*
0
100
200
300
Time (ms)
400
500
600
Experiment 1 Conclusion
When an observer expects a face (versus
a word)
there is a greater N170