the stroop task
TRANSCRIPT
Perceptron Example:
Computational model for the Stroop Task
Jaeseung Jeong, Ph.D
Department of Bio and Brain Engineering,
KAIST
GREEN YELLOW
BLUE BLUE
YELLOW GREEN
BLUE RED
The Stroop Task Can
Teach Us About:
• Attention
• Automaticity
• Learning
• Response Selection
• Word Reading
• Color cognition
• Experimental Methodology
GGGGG A A OOO
G A A O O
GGGG AAAAA O O
G A A O O
GGGGG A A OOO
rabbit
below above
* * *
above
“LEFT”
Variations
• Insufficient evidence to decide whether
similar processes are involved in all of the
Stroop-like variations.
• Dalrymple-Alford & Budayr (1966): First to
encourage presentation & timing of stimuli
individually. This method now dominates
300
400
500
600
Congruent Incongruent
Tim
e (
ms)
BLUE BLUE
Stroop
effect
300
400
500
600
Congruent Incongruent Control
tim
e (
ms
)
BLUE BLUE
interference
facilitation
Control Condition
####### rsmtlae fast
####### rsmtlae fast
####### rsmtlae fast
####### rsmtlae fast
Control Condition
fast red
fast yellow
fast red
fast green
Control Condition
fast red fast
fast yellow ready
fast red mouse
fast green fruit
Control Condition
fast red fast fat
fast yellow ready double
fast red mouse fat
fast green plan fruit
Semantic Effects
• Semantically-related distractor words (e.g.
blood, sky) have been used instead of
neutral words (e.g. double, fruit):
– some interference; more so as the semantic
association between word & color increases
– facilitation: small & dependent upon the
control condition used
• Using color words not in the response set
(e.g. purple) reduces the Stroop effect
Semantic Effects (cont) rabbit
• Congruent words facilitate responses
(compared to unrelated word or nonword)
• Same category words interfere most
• Associative word (e.g. “cheese” on picture
of a mouse) is same as an unrelated word.
• Interference in naming the color of an
incongruously colored object (e.g. a blue
banana), compared to a neutral object (e.g.
a blue book).
Acoustic Manipulations
• Using a manual response rather than vocal reduces interference
• Tying up the articulatory system (e.g. by saying “blah, blah …”) and using a manual response to the color reduced interference
• Interference increases with increasing pronouncability of nonwords, and with increasing similarity between nonwords and incongruent color words
Strategy
• Composition of entire set of trials
influences the participants’ strategy
– blocked vs. unblocked
• Cheesman & Merikle (1984):
– P’s could use info regarding proportion of
congruent primes if primes were perceived
cons’ly, but not if perceived uncons’ly (but
uncons’ primes still affected responses)
Stimulus Onset Asynchrony (SOA)
time
complete
stimulus
(e.g. GREEN)
ignored
stimulus (e.g.
GREEN)
SOA
SOA
• Dyer (1971):
– Color naming - interference decreases with
increasing SOA (0 – 500 ms SOA’s used)
• Glaser & Glaser (1982):
– Color naming - interference maximal at +/-
100 ms SOA
– Word reading – no effect of SOA
Hemispheric Differences
• Larger Stroop effect when the words are
presented to the left hemisphere than the
right
Age differences
Age
Str
oop E
ffect
0 6 20 60
Language - Bilinguals
• Stimulus: vert or green
• Response: “red” or “rouge”?
• Between-language interference ~ 75% of within-language interference
• Magiste (1984, 1985) studied relative proficiency of the person’s languages
– Whichever language someone was more proficient in caused more interference
Explanations
• Speed of Processing
• Automaticity
• Perceptual Encoding
• Parallel models
• Parrallel Distributed Processing
Speed of Processing
• Words read faster than colors are named
• Response from word reaches response
stage before response from color
• Results in interference / facilitation
• However, if SOA causes color to reach
response stage before word, does NOT
lead to reversal of Stroop effect.
• Therefore, theory is inadequate.
Automaticity
• Word reading is automatic & obligatory, color naming is a more controlled process.
• Automatic processes can interfere with controlled processes, but not vice versa
• Strategies should not affect automatic processes
• However, strategies caused by the % of congruent / incongruent trials do affect results
• Automaticity may be continuous rather than dichotomous: – This allows attention to assert some influence, but the
theory then looses some specificity and ability to test predictions decreases.
Parallel Models
• Response stage is active from start of trial.
• Each response option gains support as trial
goes on.
• Once a response reaches a threshold, that
response is chosen.
• Problems:
– Predicts symmetrical facilitation & interference. But a
solution is possible
– Could have same problems as speed of processing
account, but these are also fixable
Parallel Models
• “With fine tuning, Logan’s model can
encompass the existing data. However,
parallel models expressed only at the
conceptual level tend to have more ‘free
parameters’ than do sequential models,
which may be part of why they appear to
be more successful.” (MacLeod, 1991,p. 192).
Negative Priming
Table 1. Comparison of RTs for each NP condition for human behavioral data
Mean RTs ± SD (msec)
NP NP1 830 ± 274.9
NP2 708 ± 210.1
NP3 730 ± 234.1
non-NP Incongruent (NP1) 787 ± 288.6
Neutral (NP2) 702 ± 213.6
Congruent (NP3) 659 ± 202.4
Chung et al., Journal of Computational Neuroscience (In press)
Table 3. Comparison of RTs for each NP condition between conventional PDP and
TDNN models
Conventional PDP TDNN
(mean RTs ± SD) (mean RTs± SD)
NP NP1 16.3 ± 6.0 123.8 ± 138.2
NP2 13.4 ± 4.3 107.1 ± 31.5
NP3 12.2 ± 3.7 103.9 ± 34.3
non-NP Incongruent (NP1) 16.2 ± 6.1 113.2 ± 60.3
Neutral (NP2) 13.5 ± 4.3 104.8 ± 39.1
Congruent (NP3) 12.2 ± 3.7 100.0 ± 32.2
Chung et al., Journal of Computational Neuroscience (In press)
Chung et al., Journal of Computational Neuroscience (In press)
Chung et al., Journal of Computational Neuroscience (In press)
• Jensen (1965): w/ multiple administrations, the Stroop test is probably more reliable than any other psychometric test.
• Modifications only affect the magnitude (its quantitative form), not the pattern of the effect (its qualitative form
• 1935 – 1989: 700+ articles (~300 applied & ~400 theoretical)
• MacLeod’s review has been cited 365 times (1991-2004).
• Still not fully understood
Summary
2004: Time for another review? “I look forward to the progress that will be examined in the
subsequent review of the Stroop literature some time early in the next millennium” (MacLeod, 1991, p. 193)
1935: Stroop’s original article
1960’s: Research interest in Stroop paradigm blossoms
1973: Dyer’s review of research on Stroop effect
1991: MacLeod’s review of research on Stroop effect