Transcript IA_CogCore

Interactive Activation
Cognitive Core Class
May 2, 2007
What is Interactive Activation?
• It is the idea that our perception, cognition,
attention, and action arise out of a mutual
constraint satisfaction processes involving bidirectional influences among active
representations via bi-directional connections.
• Interactive is used in the sense of mutual
engagement and influence:
– activity, interaction, sentiment (G. Homans)
• NOT in the sense of statistical interaction
The interactive activation model
Rumelhart & McClelland (1981)
• Units stand for
hypotheses at different
levels of representation.
• Activation is the system’s
only currency.
• Mutual excitation
between mutually
consistent hypotheses.
• Mutual inhibition
between mutually
inconsistent hypotheses.
• System settles to a
stable state across all
parts given inputs of any
type.
The Interactive Activation Model as
implemented for perception of letters
•
•
Feature, letter and word
units.
Between-layer connections
were + or -; only inhibitory
connections within.
–
•
•
•
Later the between layer
inhibition was deleted
Activation follows the ‘iac’
function.
Response selected from the
letter units in the cued
location :
Strength s is a monotonic
function of each unit’s
activation.
Findings Addressed by the IA Model
• The word superiority effect
(Reicher, 1969).
– Subjects identify letters in
words better than letters in
scrambled strings or single
letters.
• The pseudoword advantage.
– The advantage over single
letters and scrambled strings
extends to pronounceable nonwords.
• The contextual
enhancement effect.
– Increasing the duration of the
context or of the target letter
facilitates correct identification.
• Reicher’s experiment:
– Words are constructed in pairs
differing by one letter.
– E.g. READ vs ROAD.
– The ‘critical letter’ is the letter
that differs between members
of the pair.
– Critical letters occur in all four
positions.
– Display is followed by a mask
made of lines or letter
fragments.
– The critical letter and the
alternative then appear as
choice alternatives, with dashes
indicating the tested position.
– There are trials with scrambled
letter strings and single letters,
with the same critical letters
used.
How the Model
Works:
Words vs. Single
Letters
Word and Letter Level Activations for
Words and Pseudowords
Idea of ‘conspiracy effect’ rather than consistency with rules as
a basis of performance on ‘regular’ items.
The Contextual Enhancement Efffect
The TRACE model of
Speech Perception
• Extends the IA model directly
to speech.
• Units representing hypotheses
are reduplicated across spatial
positions.
• Competition is proportional to
degree of overlap.
– Results in segmentation as
well as selection.
• Top-down effects from word to
phoneme level explain:
– lexical effects on phoneme
identification
• Top-down effects from
phoneme to feature level
explain
– The “perceptual magnet
effect”
Lexical Effect on Word Segmentation
Discrimination and
identification of
stimuli varying on a
single cue
TRACE vs. MERGE
Compensation for Coarticulation
• Value of the dimension
‘acute’ that signals ‘g’ (or
other phoneme) depends
on what comes after it.
• In Elman & McClelland
(1986) we proposed that
phoneme units in one
position can modulate
connections from feature
to phoneme units in
other positions.
• This led to the idea:
Maybe top-down effects
can trigger compensation
for co-articulation.
Lexically-mediated compensation for
co-articulation (LMCC)
• ‘sh’ lengthens oral ‘tube’, ‘s’
shortens it.
• This colors a following /t/or /k/.
• Perceivers compensate for this
as illustrated in the bottom left
panel.
• Q: Can lexical information that
disambiguates a sound halfway between ‘s’ and ‘sh’ also
produce this effect?
– ChrismaX -> ‘s’
– fooliX -> ‘sh’
• Yes, the effect does occur,
though it is weaker than the
acoustically mediated effect, as
predicted by the simulation.
Other ‘knock-on’ effects
• LM Selective adaptation
– Hearing an ambiguous segment in a context causes
adaptation consistent with the contextually determined
identity of the sound: E.g. Right after hearing X in ‘fooliX’
you’re less likely to identify an ambiguous sound as ‘sh’.
• LM Retuning of speech perception
– Hearing an ambiguous segment in a context causes
retuning consistent with the contextually determined
identify of the sound: E.g. After hearing X in contexts like
‘fooliX’ interleaved with clear ‘s’ in contexts like
‘Christmas’ you will tend to identify the ambiguous sound
as ‘sh’.
Interactivity in the Brain
• Bidirectional Connectivity
• Interactions between V5 (MT) and V1/V2:
Bullier
• Subjective Contours in V1:
Lee and Nguyen
• Distributed Constraint Satisfaction in Binocular
Rivalry: Logothetis
Hupe, James, Payne, Lomber, Girard & Bullier (Nature,
1998, 394, 784-787)
• Investigated effects of
cooling V5 (MT) on neuronal
responses in V1, V2, and V3
to a bar on a background
grid of lower contrast.
• Cooling typically produces a
reversible reduction in firing
rate to the cell’s optimal
stimulus.
• Top down effect is greatest
for stimuli of low contrast.
If the stimulus is easy to
see when it is not moving,
top-down influences from
MT have little effect.
• Concept of ‘inverse
effectiveness’ arises here
and in many other related
cases.
*
Lee & Nguyen (PNAS, 2001,
98, 1907-1911)
• They asked the question:
Do V1 neurons participate in
the formation of a
representation of the illusory
contour seen in the upper panel
(but not in the lower panel)?
• They recorded from neurons in
V1 tuned to the illusory line
segment, and varied the
position of the illusory segment
with respect to the most
responsive position of the
neuron.
Response to the illusory contour is found at
precisely the expected location.
Temporal Response to Real and Illusory
Contours
Neuron’s receptive field falls right
over the middle of the real or illusory
line defining the bottom edge of the square
Figure shows a V1/V2 neuron
that showed strong modulation
in firing around epochs in
which the monkey perceives
the cell’s preferred stimulus.
From Leopold and Logothetis,
1996.
Top: psth’s show strong
orientation preference.
Bottom: When both stimuli are
presented simultaneously,
neuron is silent just before a
response indicating perception
of the null direction, but quite
active just before a response
(t < 0) indicating perception of
the preferred direction.
Leopold and Logothetis
(Nature, 1996, 379, 549553) found that some
neurons in V1/V2 as well
as V4 modulate their
responses in concert with
Monkey’s percept, as if
participating in a
massively distributed
constraint-satisfaction
process. However, some
neurons in all areas do not
modulate their responses.
Thus the conscious percept
appears to be correlated
with the activity of only a
subset of neurons. The
fraction of neurons that
covary with perception is
greater in higher areas.
• Black faces prime
perception of crime objects.
• Crime or basketball words
increase attention to black
faces and the locations in
which they occur.
• There is a stereotypicality
effect in face memory that
is accentuated by crime
primes.
Interactive Activation and
‘Seeing Black’
Eberhardt et al (2005)*
Guilt by association in an interactive activation model
??