1. The Effect of Episodic Retrieval on
Inhibition in Task Switching
Jim Grange, Agnieszka Kowalczyk, &
Rory O’Loughlin
2. A Problem of Control
• Humans live in a rich, multi-task environment
• Goal-directed behaviour requires selecting the
most relevant stimulus to act upon
3. A Problem of Control
• Stimulus selection is only half the battle:
– Stimuli are often multivalent
4. A Problem of Control
• When stimuli are multivalent, we must be
able to select the relevant task to perform
• We must also be able to maintain that
operation once selected so task-irrelevant
operations do not intrude
5. A Problem of Control
• We must also be able to maintain that task
once selected so task-irrelevant intrusions do
not occur
6. A Problem of Control
• We must also be able to switch away from
this task when our goals change
8. How is Task Switching Achieved?
• A possible solution:
– Activate task-relevant representations when they
are required
– Inhibit task-irrelevant representations when they
are no longer required
14. Inhibition in Task Switching
A B A
C B A
Backward Inhibition (BI) = RT(ABA) – RT(CBA)
“N–2 repetition cost”
15. Inhibition in Task Switching
• Why is this effect important?
– Can be used to investigate inhibition using
different approaches:
• Clinical
• Neuropsychological
• Neuroscience
• Individual Differences
16. Inhibition in Task Switching
• Why is this effect important?
– Many “inhibition” effects can be explained
without appeal to inhibitory mechanisms
• e.g., negative priming, Stroop performance
– N-2 repetition cost is—to date—robust against
these alternative explanations
17. Episodic Retrieval Account
• A key non-inhibitory account that can explain
a lot of “inhibitory-type” effects
• Automatic cue-based retrieval of episodic
traces of previous task experience
– Retrieval facilitates performance if it matches
current task demands
– Retrieval interferes with performance if it mis-
matches current task demands
30. Mayr (2002)
• Episodic retrieval cannot explain n-2
repetition cost in task switching
– Remains a strong marker of inhibition
• It is not clear, though, whether episodic
retrieval has any modulatory effect
31. Mayr (2002)
• Numerical trend for
smaller costs for
episodic matches
• F(1, 38) = 1.3, p=.26
• Can’t accept a null!
Error bars denote +/- 1 SE
32. Mayr (2002)
• Bayesian analysis of this
interaction (BF01 =
0.315) suggests null ~ 3
times more likely
• This only provides
“anecdotal” support for
null (Schoenbrodt et al.,
2016)
Error bars denote +/- 1 SE
34. The Present Study
• Replicate key aspects of Mayr’s (2002) design
• Used sequential Bayesian analysis to collect
compelling data
– We only stopped data collection once we had
“substantial” support for one hypothesis over the
other
– (i.e., whether episodic retrieval does or does not
modulate the n-2 repetition cost)
35. Sequential Bayesian Analysis
• Conduct Bayesian t-test after every participant
– N-2 repetition cost (resp. rep.) Vs.
– N-2 repetition cost (resp. switch)
• Bayes Factor
– Degree of support for one model (i.e., hypothesis)
compared to another model, given the data observed
– BF10 of 10 means alternative is 10 times more likely
than null, given the data
– BF10 of 0.1 means null is 10 times more likely than
alternative, given the data
36. Sequential Bayesian Analysis
• Stop data collection when the Bayes factor is
either:
– Greater than 6 (strong support for alternative)
– Less than 1/6 (strong support for null)
37. Method
• N = 76
• Replication of Mayr’s
design
• 4 blocks of 120 trials
• Task chosen randomly
(no repetitions)
• Stimulus location
chosen randomly
38. Results
• Sequence:
F(1, 75) = 94.14,
p < .001, η2
G = .018
• Response Rep.:
F(1, 75) = 18.21,
p < .001, η2
G = .004
• Interaction:
F(1, 75) = 9.60,
p < .01, η2
G = .001
Error bars denote +/- 1 SE
39. Results
• Bayes Factor:
• BF10 = 9.97
• Model of different n-2
repetition costs for
response repetition and
switch is 10 times more
likely than a null model
Error bars denote +/- 1 SE
40. Discussion
• N-2 repetition cost is modulated by episodic
retrieval
– When retrieval parameters match current task
demands, the n-2 repetition cost is reduced
– Almost halves the cost (!)
– Important if we wish to use this cost as an
individual difference marker of inhibition
42. Method
• Manipulated cue–task
complexity
• Arrow cues provide
bottom-up support for
response selection
• Shape cues have no
pre-experimental
association with tasks
Arrow Cues Shape Cues
43. Method
Arrow Cues
• Less reliance on
working memory
representations
• Less benefit /
interference from
episodic retrieval
• Reduced episodic
retrieval effect
Arrow Cues Shape Cues
44. Method
Shape Cues
• Greater reliance on
working memory
representations
• More benefit /
interference from
episodic retrieval
• Increased episodic
retrieval effect
Arrow Cues Shape Cues
45. Method
• Stopping rule same as before, but test is on 3-
way interaction
• Currently have 17 subjects
– Not ready to stop, so…
• …data are thus preliminary…
48. Error bars denote +/- 1 SE
Bayes Factor (interaction vs.
2 main effects model) =
10.93
49. Discussion
• Episodic retrieval effects larger with more
abstract cues
– Greater reliance on WM representations
– More interaction with retrieved episodic traces
50. Discussion
• No evidence for inhibition when episodic
retrieval matches current task demands
– 3ms for Arrow cues
– NEGATIVE 52ms for Shape cues (i.e., positive
priming)
• This data set currently matches prediction of a
pure episodic retrieval account of the n-2
repetition cost
52. Conclusions
• We have provided evidence for (at least) a
modulatory role of episodic retrieval during
task switching
• When retrieval matches current task
demands:
– Reduces the n-2 repetition cost (Exp. 1)
– Introduces an n-2 repetition benefit (Exp. 2)
53. Conclusions
• The n–2 repetition cost in task switching is (at
least) a contaminated measure
– Task-specific inhibition plus
– Episodic interference / facilitation
• Researchers needs to be cognisant of this
issue when using this effect as a “pure”
measure of inhibition
54. Thank You!
A copy of these slides will be available
on our lab’s website:
www.jimgrange.wordpress.com