The Fabric of Task Episodes
Ausaf A Farooqui1, Tom Manly1
Abstract
Behavior
is
executed
through
extended
tasks
(‘preparing
breakfast’). What makes the multitude of actions occurring within a
task episode parts of one entity and not a series of independent
actions? Participants executed trials on which they could either
choose the smaller of the two numbers or the smaller font
depending on the stimulus margin color. They were biased into
construing a recurring instance of three or five trials as a higher
task episode. Results suggested that the otherwise independent
trials constituting the construed task episode got executed through
a subsuming episodic routine that was assembled at the beginning
of the episode: RT on trial 1 of the task was the highest and was
higher before longer tasks, and the rule switch cost was absent
specifically
between
consecutive
trials
that
occurred
across
presumed task episode boundaries. Plausibly, instead of generating
the cognitive state related to the construed task episode de novo at
every step, cognition assembled a routine that would automatically
unfold into the sequence of these states across time in much the
same way it generates sequence of motor actions through motor
programs. Notably, because these states were related to the
construed task episode and not to its constituent trials the presence
of the subsuming routine was not affected by the lack of
foreknowledge about the sequence, identity and even the number
1 MRC-Cognition & Brain Sciences Unit, 15 Chaucer Road, Cambridge, UKThis work was funded by Medical
Research Council, UK: MC-A060-5PQ20
1
of constituent trials. Instead, it occurred every time a group of trials
was construed as forming a higher task.
Word count: 11076
1. Introduction
The studies presented here examined the cognitive
underpinnings of task episodes. By task episode we
mean an extended and defined period of purposive
behavior construed as a single task entity. Here while
participants
engaged
in
continuous
trials
of
an
experimental session, task-irrelevant cues were used to
bias them towards viewing each recurring instance of
three or five trials as a task2. This method allowed us to
study factors related to starting task episodes without
confounding them with factors pertaining to task rules,
working memory, attention, action selection etc.
2 While we use the phrase ‘task episode’ in this paper, we do not insist that in the behavioral hierarchy these are tasks as opposed to sub­tasks of a larger task. Our concern is that irrespective
of the hierarchical level, the sequence of trials be construed as a single task entity.
2
Most goal-directed behavior is executed through
task episodes – we prepare tea, take a bath, write
emails, play games, and not individually execute the
very
many
component
action
units.
While
task
identities through which we execute our behavior are
frequently episodic and extended in time, it is not clear
in
what
sense
is
the
segment
of
behavior
corresponding to an extended task like preparing
breakfast one entity, and not a series of independent
actions. In fact the very notion of task may appear
arbitrary (Vallacher & Wegner, 1989). Breakfast, after
all, can be prepared as one (‘prepare breakfast’), two
(‘prepare tea’ and ‘prepare toast’) or four tasks (‘boil
water’, ‘brew tea’, ‘toast bread’, ‘spread butter), or
even as a subtask of a larger task (‘morning routine’).
Many existing accounts of goal directed behavior
have suggested constructs like plans, frames, schema
3
and scripts that correspond to hierarchically organized
knowledge about stereotypical task situations (Miller,
Galanter & Pribram, 1960; Minsky, 1975; Schank &
Abelson, 1977; Broadbent, 1977; Norman & Shallice,
1986; Shallice & Cooper, 2011). For example, a script
about restaurant could be ‘find a seat’, ‘read the
menu’, ‘order food’; each of which may have their own
subscripts. Hierarchical nature of the recall and
instantiation of such hierarchical mnemonic entities
may create the hierarchical organization of behavior
into task and subtasks (e.g. Bower, 1970; Newell &
Simon, 1972).
This is conceptually supported by the presence of
signs of hierarchical cognition in task situations where
the knowledge of the behavioral sequence to follow is
represented
hierarchically
e.g.
motor
sequence
(Rosenbaum, Kenny & Derr, 1983), task sequence
4
(Schneider & Logan, 2006), events (Zacks & Tversky,
2001), chunks (Gobet & Simon, 1998). During the
execution of behavior this knowledge gets retrieved
hierarchically to construct routines that specify the
identity and sequence of steps across time. In this
paper we use the term routine for situations where
extended behavior is selected and instantiated as a
single
entity
preassembled
and
gets
structure
executed
that
through
translates
into
a
the
relevant sequence across time (e.g. Henry & Rogers,
1960; Rosenbaum et al. 1983; Sternberg, Knoll &
Turock,
1990;
Schneider
&
Logan,
2006).
The
structure is hierarchical in that it consists of a
sequence
of
component
acts,
and
the
resulting
cognition is hierarchical and episodic in that it consists
of
a
defined
episode
subsumed
by
the
routine
structure.
5
Hence, if a memorized chunk of motor or task
items is to be executed (akin for example to playing a
sequence of notes on piano or executing a memorized
list of task items) or recalled the entire sequence is
instantiated as a single routine that will specify the
identity and sequence of motor or task acts across time
(Rosenbaum, Kenny, & Derr, 1983; De Jong, 1995;
Anderson & Matessa, 1997; Anderson et al. 1998;
Kahana & Jacobs, 2000; Luria & Meiran, 2003; Lien &
Ruthruff, 2004; Schneider & Logan, 2006). This creates
specific behavioral signatures. It takes longer to
execute item marking the beginning of the routine than
subsequent items. Hence, when participants execute
out a memorized sequence of task items (e.g. AABB,
where A and B were color and shape judgements), the
reaction time is the longest for the first item (Lien &
Ruthruff, 2004; Schneider & Logan, 2006). The time to
initiate the routine frequently correlates with the
6
length and complexity of the sequence (preparation
effect) because the routine corresponding to the entire
sequence is preassembled prior to its execution, and
longer/more complex routines may take longer to
assemble (Henry & Rogers, 1960; Sternberg, Monsell,
Knoll, & Wright, 1978; Luria & Meiran, 2003); e.g. it
takes
longer
to
begin
speaking
bisyllabic
than
monosyllabic words (Klapp et al. 1973), task sequences
with greater number of item level task switches (e.g.
ABBA: two switches i.e. ABBA) take longer to begin
than those with lesser switches (e.g. AABB, single
switch; Schneider & Logan, 2006; Desrochers et al.
2015).
Thirdly, the long step 1 RT is seen even when the
same sequence is repeatedly iterated because the
routine structure has to be instantiated for every
execution of the sequence, suggesting that the routine
7
is not a detached plan that exists separate from the
executing machinery. Instead, the routine structure
may be better seen as the set of commands that will go
on to create the relevant sequence and dismantle in
the process of execution, hence, has to be reassembled
for the next iteration. Lastly, the benefit of repeating a
constituent act consecutively is seen only within a
routine
and
not
across
routines,
because
the
dismantling and/or assembly of routines at sequence
boundaries removes the cognitive trace pertaining to
individual acts subsumed by the routine (Lien &
Ruthruff, 2004; Schneider & Logan, 2006, 2007, 2015).
These
characterized
and
other
hierarchical
empirical
execution
studies
of
have
tasks
in
situations where the identity and sequence of task
items to be executed across time is known and the
knowledge is hierarchically organized. Using this
8
knowledge the entire sequence is prepared for in one
go and executed as a single routine that specifies the
identity and sequence of component steps, with the
purported hierarchical organization of the sequence
knowledge determining what all gets executed as parts
of the same routine (e.g. Schneider & Logan, 2015).
Thus, a memorized chunk of motor acts is executed
through a motor routine that creates the sequence of
motor components across time (e.g. Rosenbaum et al.
1983), and a memorized sequence of task items is
executed through a routine that creates the sequence
of item specific cognitive configurations that will allow
selection of relevant motor actions (e.g. Schneider &
Logan, 2006).
Here we suggest that the principle of routinized
execution may be far more ubiquitous in cognition than
suggested
by
past
experiments,
and
may
occur
whenever ensuing behavioral episode gets construed
9
as a task entity, regardless of the presence of
hierarchical
representations
in
memory
or
the
predictability of its component steps.
1.1 Task Episodes as Abstract Cognitive Routines
When my task is to edit an unfamiliar page of
writing, the decision corresponding to the higher level
task episode (‘edit paper’) is made only at the
beginning but the series of cognitive states generated
by that decision and the resulting task context
continues across time to inform, constrain and control
various aspects of my subsequent cognition and
behavior and keeps them in line with the ongoing task;
e.g. the context created by these states keeps my eye
movements and thoughts from wandering too much in
a task irrelevant way, it maintains the knowledge
required for editing on my mental foreground, and
ensures that of the numerous tasks possible in my
10
physical environment I continue to follow the one
determined by the task episode I am already in. I
therefore continue reading with the intent of editing
the text.
Likewise when I enter an unfamiliar kitchen to
prepare breakfast from whatever is available, I may not
know what all task items I will be executing, but I do
know what task episode I will be executing. This task
episodic context will determine where I look, how I
look, what kind of decisions I make, how I act, the
sequence of my decisions and actions, the kind of
knowledge I keep in my mental foreground and change
it across time etc. Again the chain of cognitive states
following from my decision to prepare breakfast
continue
across
time
effortlessly
forming
the
subsuming template for all other cognitive activities
during the episode till its completion.
11
In both these examples the sequence of task items,
steps or even subgoals was not known beforehand nor
was it predetermined by a common representation, and
depending on the task scenario and creativity, could
have
taken
many
forms,
including
those
never
previously encountered by the doer. Nonetheless, every
step of the sequence was executed through the higher
level cognitive states related to the main task, and in
this sense the multitude of actions during the episode
were parts of the same task. My boiling of water and
spreading butter on toast were subsumed by my
intention to prepare breakfast.
Key features of task episodes suggest that the
series of cognitive states related to the main task and
subsuming
the
execution
of
behavior
may
be
instantiated as a single routine, even when the
component items and actions executed during the
period are not predictable. First, as illustrated in above
12
examples this series of cognitive states seems to be
generated by a single act - the decision to execute any
extended task (editing the paper or preparing the
breakfast) is made once. Once initiated, subsequent
execution
occurs
through
the
chain
of
events
generated by that first act. The doer may have to
deliberate about the execution of a particular lower
level step, but she does not deliberate upon what task
episode she is executing, unless she gets distracted.
And the distraction proves the point. Now the doer has
to rethink what task episode she was in and where she
was
at
within
the
episode,
suggesting
that
the
distraction broke some chain of thought, i.e. the series
of cognitive states that had ensued from the act that
began the episode got interrupted and another task
episode level decision had to be made.
Second, the cognitive context that decides upon
and controls any individual step would have to be
13
simultaneously related to the goal of the task episode,
the current behavioral/cognitive state and how to
achieve
the
goal
from
that
state.
Consequently,
generating the relevant higher level cognitive state de
novo at each step would be slow and effortful.
Cognition
may
solve
this
problem
of
sequential
organization by instantiating the entire sequence of
states as a single routine (e.g. Dezfouli et al. 2014).
1.2 Current Study
We speculated that to the extent any extended
purpose is conceived as a task episode it will be
executed
through
an
episodic
routine
that
will
14
instantiate the series of episode related cognitive
states3 subsuming the period of execution, irrespective
of the predictability of the task and motor content of
the episode. Because these cognitive states are related
to the task identity the doer construes she is executing
(e.g.
‘preparing
breakfast’),
the
episodic
routine
instantiating these states will follow the contours of the
task episode as construed by the doer. Signs of
routines and consequent hierarchical organization of
cognition seen previously during the execution of
memorized motor acts and task lists will therefore be
seen whenever an arbitrary segment of ongoing
behavior is construed as a task episode.
Participants executed trials on which one of two
task items could be executed depending on the outer
margin color of the stimulus (blue: choose the lower
3 It may be argued that the higher level is not a series of cognitive states but a single state that spreads across time. This does not affect the crux of the argument ­ the series of states or a single state spreading across time seems to get generated by a single, higher level act. We prefer the construct of ‘series of states’ because the neural activities and pattern of neuronal configurations do not remain constant across task periods but change continuously across time (Sigala et al. 2009; Stokes et al. 2015)
15
value between the two numbers; green: choose the
smaller font). The task item to be executed on any trial
was not known beforehand, and the probability of it
being repeated or switched across successive trials
was the same. Across different experiments a variety of
means was used to bias subjects into conceiving a
sequence of consecutive trials (or a period of time) as a
defined task episodic unit. These included having to
additionally count trials in threes (1-2-3) or fives (1-2-34-5; experiments 1 and 6), temporal grouping of trials
along with an irrelevant countdown (experiments 2, 3
and 4), and the presence of an irrelevant outer margin
that stayed on for extended epochs of time and framed
the execution of extended series of trials (experiment
5).
If trials of such construed tasks were executed
through a single subsuming episodic routine assembled
16
at the beginning of the episode then RT will be highest
at trial 1 (because of the additional time needed to
assemble the routine), and longer episode will have
even higher trial 1 RT (because of the need to
assemble
longer
routines).
Further,
switch
cost
between consecutive trials will only be present within
an episode and not across episode boundaries because
trial related configurations will be ensconced within
the
episodic
configurations
and
hence
will
get
dismantled at episode boundaries leaving no advantage
for
repeating
the
consecutive
trials
Importantly,
these
same
task
across
signs
will
rule
operation
episode
appear
on
boundaries.
every
time
subjects conceive of their behavior corresponding to
the series of trials as a task episode.
17
2.1 Experiment 1
2.1.1 Methods
18
Figure 1. Trial blocks began with the instruction screen stating whether trials are to be executed in steps of 3 or 5. Subsequent
trials were to be executed while keeping a covert count (e.g. steps of 3: 1-2-3-1-2-3…). Trial rule was cued by the color of the
outer margins – blue: choose the smaller value, green: choose the smaller font. The block would end with a ‘?’, to which
participants keyed in the number of the step just executed.
Figure 1 shows the scheme of this experiment.
Participants were asked to keep a covert count of trials
in 3s (1-2-3-1-2-3) or 5s (1-2-3-4-5-1-2-3-4-5) depending
on an onscreen ‘steps of 3’ or ‘steps of 5’ instruction at
the beginning of each block of 6-35 trials. This screen
19
remained until participants pressed the spacebar. On
each trial two numbers differing in value (between 0 to
99) and font size (Arial font 60 or 20) were displayed
on each side of the fixation cross. A margin box
appeared around, and at the same time as, each
number display. Margin color determined the rule
relevant on that trial - blue: choose the smaller value,
green: choose the smaller font. Participants’ decisions
were conveyed via presses on buttons that were
spatially congruent with the correct onscreen number
(Numpad 1 for left and Numpad 2 for right on a
standard QWERTY keyboard number pad). The stimuli
remained onscreen until a response was made, and
was followed by the next trial after 1 s.
To check whether participants were keeping an
accurate 3 or 5-trial count, a probe (“?”) appeared at
the end of the block of 6-35 trials. Participants were
20
asked to key in the step number that they had just
executed before the probe appeared (e.g. if the probe
appeared after 7 trials, the correct response in a 5-trial
episode block would be 1 (1-2-3-1-2-3-1). Feedback was
given on the accuracy of probe responses (high pitch
tone for correct, low pitch tone for incorrect). This and
subsequent experiments were created in Visual Basic
and run on a Dell computer with an 85 Hz refresh rate
monitor kept at a comfortable distance from the
participant.
Procedure:
The experiment was conducted on an individual
basis in a testing room designed to minimize visual and
noise distraction. Participants were first given 7 trials
of practice on each of the two operations (value and
font judgments). They then completed a 30-trial
practice block in which the stimulus margin changed
21
its color randomly signaling the currently relevant rule.
They were then told to execute trials while keeping a
count in threes. They practiced on two such blocks
before proceeding on to the main experimental session.
Participants completed 70-120 blocks each consisting
of 6-35 trials with 3- and 5-step blocks being randomly
interlaced.
In
all
practice
and
subsequent
trials
participants were asked to respond as quickly and as
accurately as possible.
Participants:
Fifteen healthy participants (9 females) were
recruited through MRC-CBU volunteers’ panel (age 18
to 40 years). They gave written, informed consent
before the experiment, and were paid £8.50 for their
participation. All had normal or corrected to normal
vision.
22
2.1.2 Results:
Figure 2. Pattern of reaction times across the steps of the 5 and 3 step episodes (continuous line bars: rule switch trials,
dashed line bars: rule repeat trials). Note that trial 1 had the highest RT for both switch and repeat trials in both 5 and 3 step
episodes.
Table 1. Mean reaction times (ms) and accuracies (%) across rule switch and repeat trials of 3 and 5 step episodes.
Response Time
Accuracy
Effect
dfs
F
MSE
p
F
MSE
p
Serial Position (3 step)
2,28
37.9
196781
<0.001
1.55
0.001
0.23
23
Response Time
Accuracy
Effect
dfs
F
MSE
p
F
MSE
p
Rule Switch (3 step)
1,14
32.3
255789
<0.001
9.86
0.11
<0.01
Rule Switch x Serial
Position (3 step)
2,28
18.7
22258
<0.001
0.52
0.001
0.6
Serial Position (5 step)
4,56
46.9
317819
<0.001
3.07
0.002
0.02
Rule Switch (5 step)
1,14
26.0
256207
<0.001
12.4
0.012
<0.01
Rule Switch x Serial
Position(5 step)
4,56
5.7
9956
0.001
1.76
0.002
0.15
Serial Position x
Episode length
2,28
24.2
39053
<0.001
1.5
0.001
0.25
Table 2. Serial Position: Repeated measures ANOVA looking at the main effect of the position of trial within the episode on
RT and accuracy. Rule Switch: Main effect of rule switch (Switch vs Repeat trials). Serial Position x Rule Switch: Interaction
between the effects of rule switch and serial position. Serial Position x Episode length: Effect of serial position compared
across 3 step and the first three trials of 5 step episodes.
Figure 2 and tables 1 and 2 summarize the main
results. As is evident, these concur with the key
predictions: trial 1 RT be the longest, trial 1 RT for 5
trial episodes > 3 trial episodes, and switch cost be
24
absent at trial 1. The first trial of the conceived episode
did take longest to execute (table 1, and main effect of
serial position in table 2). Cohen’s d (effect size: mean
difference/standard deviation) was 1.68 and 1.99 for 3
and 5 trial episodes respectively. This trial 1 RT was
higher for 5-trial compared to 3-trial episodes (95% CI
of difference = [46, 113], table 1 and interaction
between serial position and episode length in table 2,
Cohen’s d = 1.3). While the performance on switch
trials was poorer than on repeat trials (table 2, main
effect of rule switch), this effect of rule switch differed
across the serial positions within the episode (table 2:
Rule Switch x Serial Position). Specifically, as predicted
it was absent at trial 1 (RT: paired t 14 = 1.05, p=0.3,
95% CI of difference = [-19, 54]; Accuracy: paired t 14 =
1.1, p =0.3, 95% CI of difference = [-0.02, 0.01]).
25
An interesting dissociation was that while elevated
RTs on switch-trials were accompanied by the expected
reduced accuracy (compared with repeat-trials), the
elevated RTs on the first trial of a task episode was, if
anything, associated with greater accuracy on those
trials. This was the case even though RT on switch
trials was slower only by 92 ms compared to the
difference of 250 ms between trial 1 and subsequent
trials. This is to be expected because trial 1 response
got delayed because the episodic routine had to be
assembled prior to the execution of trial 1, and not
because of additional control demands related to the
task item execution.
In the next experiment we sought to replicate the
basic findings of Experiment 1 using a different design
and to test another key prediction. The episodic routine
account claims that the entire sequence of the higher
26
task related cognitive states through which individual
trials would be executed are prospectively prepared for
at the beginning of the episode and subsequently get
instantiated
automatically
across
the
episode.
In
contrast, when trials are executed as independent
tasks and not as parts of an extended task then task
related
cognitive
states
would
be
individually
instantiated before the execution of each trial. This
would predict faster execution of trials executed as
parts of a task episode (because the task related
cognitive state through which they will be executed is
already prepared for) than identical trials executed as
independent tasks. Obviously, the RT benefit for trials
executed as part of a task would be apparent only on
trial 2 and beyond, because trial 1 RT additionally
includes the time taken in assembling the routine.
2.2 Experiment 2
27
Here
we
used
a
different
method
to
bias
participants’ conception of what constituted a task.
Series of 3 or 5 consecutive trials were grouped
together into chunks by having relatively small intertrial durations between them, while these chunks were
separated from each other by discernibly larger
durations (Figure 3). This was further reinforced by a
faded
number
in
the
stimulus
background
that
conveyed the number of steps left in the current
episode. The first trial of the 3-trial chunk had the digit
‘3’ in the background, while the second trial had ‘2’
and the third had ‘1’. Likewise in 5-trial chunks this
background digit changed 5-4-3-2-1, across its 5 trials.
Note that the participants did not have to construe the
every 3 or 5 trials as a higher task. Trials (and by
extension the experiment) could be executed perfectly
well without construing the series of trials as a task
episode.
28
Apart from blocks composed of 3 and 5 trial
episodes there was a third block type whose trials were
not organized into chunks and were instead presented
as one flat sequence. Such trials had the digit ‘1’ in the
stimulus background. We assumed that each trial of
this block type will be regarded as independent of each
other and will be executed as an independent task, in
contrast to trials of 3 or 5 trial episodes which will be
executed as parts of a higher task.
2.2.1 Methods:
Figure 3. Trial rules were the same as in Experiment 1. There were three kinds of trial blocks. (a) 3 trial episodes were created
by having small inter-trial intervals within the episode and large inter-trial intervals across episode boundaries. This was
further reinforced by the faded digit in the background that went from 3-2-1 across the 3 trials making up the episode. (b) 5
trial episodes were similar to the 3 step episodes other than consisting of 5 trials. (c) Trials of the independent trial blocks
were presented in a flat sequence with constant inter-trial interval and had the digit ‘1’ in their background that remained the
same throughout the block.
29
Stimuli and trial rules were identical to experiment
1 (Figure 3).
Before 3 and 5 step task blocks
participants saw ‘3 Step Tasks’ and ‘5 Step Tasks’
respectively, onscreen till spacebar was pressed. In
such blocks 3 or 5 successive trials were temporally
grouped together with smaller inter-trial durations
(500 ms) within the chunk and longer inter-trial
durations (2 s) between successive chunks. In addition,
the first trial of 3 trial chunks had a faded ‘3’ in the
stimulus background, the second trial had a ‘2’ and the
third had ‘1’; likewise for 5 trial chunks (5-4-3-2-1).
Before the independent trial block participants saw
“Independent Trials”, and the individual trials of this
block had the number ‘1’ in the stimulus background.
Inter-trial interval in these blocks was 500 ms. The 3and
5-trial
task
blocks
had
48
and
80
trials
respectively, and the independent trial blocks had 30
30
trials. The order of the three block types was random.
Participants did a total of 85 – 100 blocks.
Participants first did a practice block of 50 trials.
These were presented as a flat sequence (iti = 500 ms).
They then proceeded to the main experiment. They
were not informed about the temporal grouping of the
trials in the main experiment, and were just asked to
ignore the faded number in the stimulus background.
Testing conditions were same as for Experiment 1.
Participants
Eighteen healthy participants (11 females)
were recruited through MRC-CBU volunteers’ panel
(age 18 to 40 years). They gave written, informed
consent before the experiment, and were paid £8.50
for their participation. All had normal or corrected to
normal vision.
31
2.2.2 Results
Figure 4. Pattern of reaction times across the steps of the 5 and 3 step episodes (continuous line bars: rule switch trials,
dashed line bars: rule repeat trials). Independent dashed and dotted lines represent the switch and repeat trial RTs in the
independent trial blocks. Note that these are higher than switch and repeat trial RTs of both 3 and 5 step episodes.
32
Table 3. Mean reaction times (ms) and accuracies (%) across rule switch and repeat trials of 3 and 5 step episodes, and of
independent trial blocks.
Response Time
Effect
dfs
F
Serial Position (3 step)
2,34
29.6
Rule Switch (3 step)
1,17
Rule Switch x Serial
Position (3 step)
Accuracy
MSE
p
F
MSE
p
419633
<0.001
0.44
0.001
0.6
43.2
440396
<0.001
0.9
0.001
0.3
2,34
14.2
55634
<0.001
4.02
0.003
0.02
Serial Position (5 step)
4,68
31.5
199388
<0.001
1.3
0.001
0.3
Rule Switch(5 step)
1,17
39.5
758310
<0.001
9.4
0.01
<0.01
33
Response Time
Accuracy
Effect
dfs
F
MSE
p
F
MSE
p
Rule Switch x Serial
Position(5 step)
4,68
13.9
37673
<0.001
1.4
0.001
0.2
Serial Position x
Episode length
2,34
0.324
1013
0.7
1.5
0.001
0.25
Table 4. Serial Position: Repeated measures ANOVA looking at the main effect of the position of trial within the episode on
RT and accuracy. Rule Switch: Main effect of rule switch (Switch vs Repeat trials). Serial Position x Rule Switch: Interaction
between the effects of rule switch and serial position. Serial Position x Episode length: Effect of serial position compared
across 3 step and the first three trials of 5 step episodes.
Figure 4, tables 3 and 4 summarize the key results.
The first trial of the 3 and 5 trial episodes took longest
to execute (Cohen’s d: 1.3 and 1.6; table 3, and main
effect of serial position in table 4). While there was a
main effect of rule switch on performance (main effect
of rule switch in table 4), the effect varied across serial
positions (table 4: Rule Switch x Serial Position).
Specifically, as in the previous experiment it was
absent on trial 1 (RT: paired t17 = 1.7, p=0.1, 95% CI of
34
difference = [-8, 71]; Accuracy: paired t 17 = 0.8, p
=0.4, 95% CI of difference = [-0.01, 0.02]). Unlike the
previous experiment, the trial 1 RT for 5 step episodes
was not higher than 3 step episodes (95% CI of
difference = [-39, 59]; table 4 interaction between
serial position and episode length, Cohen’s d = 0.1).
Lastly, as in the previous experiment the strong
difference in RT between the first and subsequent
steps was not accompanied by changes in accuracy
(table 3, and main effect of serial position on accuracy
in table 4).
Trials conceived as parts of a task episode were
executed faster than trials conceived as independent
tasks. The dashed and dotted lines in Figure 4 mark
the RTs on rule switch and repeat trials in the
independent trial blocks. As is evident these were
higher than RTs on trials 2 and beyond of task episodes
(t17 > 5.5, p < 0.001; Cohen’s d > 1.61). Accuracies
35
however were not significantly different (table 3). This
control
benefit
also
extended
to
switch
control.
Reaction time switch cost amongst independent trials
was greater than that amongst trials executed as parts
of task episodes (176 vs 128 ms; t17 = 2.1, p = 0.06, CI
of difference = [-1, 96], Cohen’s d = 0.49). Note that in
this analysis we excluded trial 1 which showed no
switch cost.
Thus,
organization
of
trials
into
larger
task
episodes created a slower trial 1 but faster execution
of subsequent trials. Was the price paid at trial 1 offset
by gains at subsequent steps? We compared average
RT and accuracy between blocks organized into task
episodes with those that were not. Average RT on 3
and 5 trial task blocks was still significantly lower than
on the independent trial blocks (t 17 > 4.1, p < 0.001,
Cohen’s d = 0.82). Accuracy measures, however, were
not different across them.
36
These results suggested that the episodic routine
structures assembled at trial 1, that resulted in higher
trial 1 RT were indeed prospectively related to the
execution and control of the rest of the trials making
up the episode, and hence trials executed through
them were better controlled than trials executed as
independent tasks. These results also showed that the
presence of episodic routine did not decrease the
limited capacity cognitive reserves available for the
execution and control of its component trials.
Experiment 2 did not show increased RT at the
beginning of 5- compared with 3-trial task episodes. A
possible reason could have been that in the current
design the same conceived task episode was repeated
across the block, hence participants did not have to
organize their cognition actively enough for the small
temporal difference in initiating 5 and 3 trial tasks to
be evident.
37
2.3 Experiment 3
The design was identical to experiment 2 except
that both 3 and 5 trial task episodes would now occur
randomly in the same block.
2.3.1 Methods
Identical
to
Experiment
2.
The
experimental
session lasted half an hour during which participants
did an average of 250 task episodes, which consisted of
38
roughly equal number of 3 and 5 step episodes in a
random order.
Participants
22
healthy
participants
(11
females)
were
recruited through MRC-CBU volunteers’panel (age 18
to 40 years). They gave written, informed consent
before the experiment, and were paid £8.50 for their
participation. All had normal or corrected to normal
vision.
2.3.2 Results
39
Table 5. Mean reaction times (ms) and accuracies (%) across rule switch and repeat trials of 3 and 5 step episodes, and of
independent trial blocks.
Response Time
Effect
dfs
F
MSE
Accuracy
p
F
MSE
p
40
Response Time
Accuracy
Effect
dfs
F
MSE
p
F
MSE
p
Rule Switch (3 step)
1,21
39.7
577148
<0.001
2.2
27.1
0.15
Rule Switch x Serial
Position (3 step)
2,42
11.1
93740
<0.001
2.3
13.0
0.11
Serial Position (5 step)
4,84
23.5
697698
<0.001
2.5
9.9
0.05
Rule Switch (5 step)
1.21
43.1
1273614
<0.001
6.2
48.8
0.02
Rule Switch x Serial
Position (5 step)
4,84
12.4
51311
<0.001
1.9
7
0.12
Serial Position x
Episode length
2,42
11.1
35336
<0.001
0.09
0.39
0.9
Table 6. Serial Position: Repeated measures ANOVA looking at the main effect of the position of trial within the episode on
RT and accuracy. Rule Switch: Main effect of rule switch (Switch vs Repeat trials). Serial Position x Rule Switch: Interaction
between the effects of rule switch and serial position. Serial Position x Episode length: Effect of serial position compared
across 3 step and the first three trials of 5 step episodes.
Tables 5 and 6 summarize the key results. Notably,
the trial 1 RT was higher for 5 trial compared to 3 trial
episodes (table 5 and table 6, interaction between
serial position and episode length, 95% CI of difference
41
= [15, 96], Cohen’s d = 0.6). Other results were largely
a replication of the previous two experiments. The first
step of the conceived episode took longest to execute
(table 5, and main effect of serial position in table 6).
This was the case for both switch and repeat trials at
trial 1. Performance on switch trials was poorer than
on repeat trials (table 6 main effect of rule switch),
however, the effect of rule switch was not the same
across the trials making up the episode (table 6: Rule
Switch x Serial Position). Specifically, the switch cost
was absent at the first position (RT: paired t 21 = 0.89,
p=0.4, 95% CI of difference = [-43, 109]; Accuracy:
paired t21 = 0.3, p =0.7, 95% CI of difference = [-1.4,
1.9]). Lastly, accuracies were not significantly different
between the trial 1 and subsequent trials (main effect
of serial position on accuracy in table 6).
Current experiment affirmed that longer episodes
did indeed take longer to initiate, and replicated other
42
behavioral signatures seen in experiments 1 and 2 that
suggested that trials construed as parts of a task were
executed through subsuming routines assembled at the
beginning of the extended task.
2.4 Experiment 4
Whilst there is debate about the contributions of
various sources to switch-cost (repetition advantage,
inhibition of or interference from previous trial etc.;
Monsell, 2003), it is, by definition, related to the
previous trial. We have argued that the absence of
switch
cost
at
trial
1
was
related
to
the
disassembly/reassembly of episodic routine structures
at episode boundaries ‘washing out’ the existing trial
rule related configurations. This would predict that the
interference seen during Stroop task (e.g. identify the
font color of the word RED when the font is blue)
would be unaffected at the beginning of task episodes
43
because the control requirements generated by Stroop
incongruence pertain largely to the current trial and
not to the previous one.
Participants
did
a
modification
of
color-word
Stroop task (Figure 5). The experiment had two kinds
of blocks. Trials of the first one were temporally
grouped into chunks similar to that in Experiments 2
and 3. In such blocks four consecutive trials were
grouped together through smaller inter-trial interval
(500 ms), while two adjacent chunks were separated by
longer intervals (1500 ms). Furthermore, the stimulus
background had a faded digit that moved 4-3-2-1
across the four trials of the chunk.
In the second kind of blocks, trials were not
organized into higher order tasks and were presented
as a flat sequence (iti = 500 ms) with the digit ‘1’ in
the stimulus background. The second kind of blocks
44
allowed us to replicate the results of experiment 2 that
had shown faster execution of trials when they were
executed as parts of a larger task episode than when
similar trials were executed as independent tasks.
Specifically, we were curious if trials executed as parts
of an episode will show better Stroop control compared
to independent trials. Recall that in experiment 2 such
trials had shown better switch control.
2.4.1 Methods
45
Figure 5. Participants chose the color of the print of the word from the two choices below. Notion of task episodes was created
by having short inter-trial durations between the four trials making up the supposed episode and long inter-trial durations
between the consecutive trials across successive episodes. A faded number went from 4 to 1 across the four trials of these
episodes. In a second block type trials were presented as a flat sequence with constant inter-trial interval across the block. All
trials of such blocks had the number ‘1’ in their background.
On
each
trial
participants
saw
a
centrally
presented color name (‘Red’, ‘Blue’, ‘Green’, ‘White’; in
Arial font size 40) in colored prints, such that the print
and the name were congruent on 60% of trials and
incongruent on the rest (Figure 5). Participants’ chose
the color of the print from the two color words
presented in Arial black font size 20 (1 degree away
from the center and ½ degree below it) below by
pressing a button spatially congruent with their choice
46
- left: Numpad 1 (to be pressed with right index finger);
right: Numpad 2 (to be pressed with right middle
finger). One (allocated to left or right at random)
always reflected the font color. When the font and word
were incongruous, the other word always repeated the
stimulus word. On congruous trials the other word was
selected randomly from the remaining colors. Between
these two options a partially (70%) transparent digit
appeared in black. The stimuli remained on screen till
a response was made. Erroneous responses elicited a
low pitched feedback tone.
Procedure
Participants completed a 3-minute practice block of the
basic Stroop task at the beginning of the session. For
the main session they were only told to ignore the digit
in the background, and were not instructed about the
47
episodic organization of trials. Before the 4-trial blocks
the instruction screen mentioned ‘4 - step blocks’, it
remained on until the spacebar was pressed. Such
blocks consisted of 48 trials. The instruction for the flat
trial sequence block mentioned ‘independent trials’.
Such blocks consisted of 20 trials. These blocks were
interleaved. Participants did a total of 150 - 250 blocks.
Participants
Eighteen healthy participants (9 females) were
recruited through MRC-CBU volunteers’ panel (age 18
to 40 years). They gave written, informed consent
before the experiment, and were paid £8.50 for their
participation. All had normal or corrected to normal
vision.
48
2.4.2 Results
Figure 6. Pattern of reaction times across the four trials of the construed task episodes (continuous line bars: incongruent
trials, dashed line bars: congruent trials). Independent dashed and dotted lines represent the incongruent and congruent trial
49
RTs in the independent trial blocks. Note that these are higher than the corresponding trials that formed part of a larger task
episode.
Table 7. Mean reaction times (ms) and accuracies (%) across the incongruent and congruent trials corresponding to the four
steps of task episode and of independent trial blocks.
Response Time
F
MSE
Accuracy
Effect
dfs
p
Serial Position (4 step)
3,51
18.3 90838
<0.001
1.8
1821
0.2
Stroop Cost (4 step)
1,17
92.1
769757
<0.001
87.9
726102
0.3
Stroop Cost x Serial Position (4 step)
3,51
2.1
1965
2.4
2115
0.08
0.1
F
MSE
p
Table 8. Serial Position: Repeated measures ANOVA looking at the main effect of the position of trial within the episode on
RT and accuracy. Stroop Cost: Main effect of Congruence (Incongruence vs Congruence trials). Stroop Cost x Serial
Position: Interaction between the effects of congruence and serial position.
50
Figure 6 and tables 7 and 8 summarize the key
results. As before, the first trial of the episode took
longest to execute (table 7, and main effect of serial
position in table 8, 95% CI of difference = [54, 146],
Cohen’s d = 1.1). This was the case for both congruent
and incongruent trial 1. Expectedly, the performance
on incongruent trials was poorer than on congruent
trials (table 8 main effect of rule switch), but crucially
it did not vary across the episode (table 8: Rule Switch
x Serial Position). Specifically, unlike switch cost in
previous experiments, the cost of incongruence did not
disappear on trial 1 of the episode (RT: one-sample t17
= 8.8, p=<0.001, 95% CI of difference = [127, 206];
Accuracy: one-sample t17 = 4.2, p = <0.001, 95% CI of
difference = [1.7, 5.0]).
The dashed and dotted lines in Figure 6 mark the
RTs on incongruent and congruent trials in the
independent trial blocks. As is evident both were
51
higher than the incongruent and congruent trials of
task episodes. In fact the average RT on trials executed
as
parts
of
task
episodes
was
lower
than
for
independent trials even when task episode trial 1 RTs
were included in comparison (congruent trials: t 17 = 7,
p<0.001, 95% CI of difference = [89, 165]; incongruent
trials: t17 = 4.9, p<0.001, 95% CI of difference = [69,
175] ). Whilst there was a general RT advantage for
task
episode
trials,
Stroop
cost
on
RT
(RT
on
incongruent trials - RT on congruent trials) did not
differ between task episode and independent trials (t 17
= 0.3, p=0.8, 95% CI of difference = [-27, 37]).
However
Stroop
cost
on
accuracy
(Accuracy
on
Congruent trials - Accuracy on incongruent trials) was
significantly higher on independent than task episode
trials (t17 = 2.6, p=0.02, 95% CI of difference = [0.5,
5.5]). As might be expected from that result, when
Stroop
trials
were
broken
down
by
congruency,
52
accuracy was greater for task episode than individual
trials for incongruent trials (t17 = 2.7, p=0.02, 95% CI
of difference = [0.6, 5.3]), but there was no difference
on congruent trials (t17 = 0.07, p=0.9, 95% CI of
difference = [-0.5, 0.5]).
These results affirmed the prediction that Stroop
cost will not be absent on trial 1, and that trials
executed as parts of a higher task will get executed
faster and better controlled than trials executed as
independent tasks. Again, the presence of the episodic
routine did not decrease the limited capacity cognitive
reserves available for resolving Stroop interference.
Instead, analogous to experiment 2 results, Stroop
control was better on trials subsumed by the episodic
routine.
2.5 Experiment 5
53
Counting
or
countdowns
involved
in
prior
experiments can be construed as a hierarchical act,
and it is possible that the hierarchy evident in the
execution of trials in experiments 1 to 4 was a spillover
from a separate but simultaneous hierarchical task –
counting (or countdown) in 3s, 5s or 4s. Through this
experiment we clarify that this was not the case.
Instead above results were the result of the construal
of the trial series as a single task, as a consequence of
which these trials were executed through a routine
instantiating the cognitive states related to that task.
The task episodes of the current experiment did
not have a fixed number of trials and did not involve
counting or countdown. The idea of episodes was
created by the presence of an additional margin
(outside the one conveying the relevant trial rule) that
stayed on for an extended duration during which trials
54
would appear randomly at any time. Trials would not
appear after this margin had switched off. Every
episode began with the appearance of trial 1 stimulus
along with this margin (Figure 7). While the trial 1
stimuli disappeared after the response, the outer
margin remained on. Subsequent trials would appear
anytime while this outer margin was on. This margin
went off marking the end of the episode, and came
back on with the beginning of the next episode. The
notion of task episode was thus implicitly defined as
the period enveloped by the presence of the outer
margin during which a trial could appear any time.
Two kinds of episodes (short and long), framed by
different color margins (black or red), were randomly
interleaved. Short episodes lasted between 3 to 6
seconds and two to three trials would appear at any
time during this period. Long episodes lasted between
55
7 to 10 seconds and three to seven trials could appear
during this period. The inter-episode interval was fixed
at 2 s, but the inter-trial interval within an episode
varied from 50 ms to 3s in short episodes and 50 ms to
5s in long episodes. Note that in this experiment, not
only did the subjects not have any foreknowledge
about the identity and sequence of task items, they also
could not predict the number of trials and their intertrial intervals.
56
Figure 7. The outermost margin (here in red) stayed on for extended duration during which an unpredictable number of
trials would appear at random intervals. Trials would not appear when this margin was off. We hoped that participants
would construe the temporal epochs carved by the presence of such margins as the recurring task episodic units to be
executed.
2.5.1 Methods
Save for the additional outer margin, individual
trials were identical to Experiment 1 (speeded decision
as to which of two numbers had the smallest value or
font size as indicated by the color of the inner margin).
Participants did two experimental sessions each lasting
57
16 minutes, with a period of rest in between. Within a
session the two kinds of conceived episodes occurred
randomly. Participants did a 4 minute practice session
that was identical to the main experimental session.
They were not instructed about the episodic structure
of the task block and were told nothing about the
outermost margins.
Participants
Twenty seven healthy participants (17 females)
were recruited through MRC-CBU volunteers’ panel
(age 18 to 40 years). They gave written, informed
consent before the experiment, and were paid £8.50
for their participation. All had normal or corrected to
normal vision.
58
2.5.2 Results
Figure 8. Pattern of RTs across long and short episodes. Note that trial 1 RT was the highest, and longer episodes had higher
trial 1 RT.
Key results of previous experiments were again
replicated (Figure 8). The first trial of the episode took
longest to execute (F2, 52 = 26.2, p < 0.001). This trial 1
RT was longer before longer episodes (95% CI of
difference = [14, 39], Cohen’s d = 0.81; t26 = 4.2, p <
0.001). Switch cost was substantially reduced at the
first trial of the episode (Smaller episodes: F 2, 52 = 18;
Longer episodes: F6, 156 = 8.1, p < 0.001 for both). The
execution of trials construed as parts of a larger task
59
episode again showed key behavioral signatures that
suggested that these trials were executed not as
independent entities but as parts of a routine. This was
the case even though participants were unaware of the
number and sequence of task items they would be
required to perform, when those items would be
executed and the precise duration of the episode.
2.6 Experiment 6
The thesis that task episodes are executed through
a preassembled routine that unfolds into the sequence
of higher level cognitive states making up the task
episode implies that the routine will have elements
related to the entire length of the episode and not just
those related to the constituent trials. This predicts
that larger routines will be assembled before episodes
that are longer in time but have the same number of
60
trials causing longer trial 1 RT before episodes that
last longer but have the same number of trials.
Here
participants
executed
trials
that
were
identical to those in experiments 1-3 in three different
kinds of blocks - 3-trial-short, 5-trial-short and 3-triallong (Figure 9). In the first two there was no inter-trial
interval within the episodes; the next stimulus onset
immediately after response. In the 3-trial-long blocks
within-episode
inter-trial
interval
was
2
s.
This
produced task episodes with mean execution durations
of 2.6 s, 4.0 s and 6.6 s, respectively. The critical
prediction was that duration would play a separate role
to number of trials and, therefore, trial 1 RT of 3-triallong episodes would be greater than for 3-trial-short.
Another now obvious prediction was that RT for trial 1
of 5-trial-short episodes would be greater than for 3trial-short as these episodes had both more trials and
longer duration. It was less clear what would occur on
61
trial 1 of the 5-trial-short vs. 3-trial-long comparison,
where task step number and duration were pitted
against each other.
A concern was that the 2 s inter-trial intervals of
the 3-trial-long blocks would lead participants to view
these as individual trials rather than as part of a single
task. To offset this a new feature was introduced. In all
blocks participants were asked to press a ‘task end’
button at the end of each episode in the hope that this
would reinforce the construal that individual trials
were part of a larger task.
2.6.1 Methods
62
Figure 9. There were three kinds of blocks of the number-font switching task, 3-trial-short, 5-trial-short and 3-trial-long.
Inter-trial intervals within an episode in the first two were 0 s and in the third 2 s. Accordingly the effect of episode duration
could be examined separately from the number of component steps on trial 1 RT of each task episode. Inter-episode interval
was 2 s in all conditions. To encourage construal of 3- and 5-trials as a single task, participants were asked to press an
additional ‘task end’ key at the completion of each episode.
Stimuli and trial rules were identical to Experiment
1. The key differences in this experiment concerned
inter-trial intervals and the additional requirement to
press a ‘task end’ button (key ‘Z’ on QWERTY
keyboard) at the completion of each task episode. In 3trial-short and 5-trial-short blocks, within each episode,
the stimuli for the next trial onset immediately after
the response to the previous trial (Figure 9a and b).
Within 3-step-long episodes the onset of the next trial
occurred 2 s after the response to the previous trial
during which the monitor was blank (Figure 9c). In all
63
conditions there was a 2 s delay between the end of
each episode and the beginning of the next.
Participants completed a 5 minute practice session
before
the
main
experiment,
during
which
they
practiced on the three kinds of episode blocks. In the
experiment session, blocks comprised 60 trials (i.e. 20
task episodes in 3-trial-short and 3-trial-long blocks
and
12
task
episodes
in
5-trial-short
blocks).
Participants completed 30-40 blocks, the different
number reflecting how long they chose to rest between
in between blocks in the fixed-duration experimental
session.
Block
order
was
randomized
for
each
participant.
Participants
Nineteen healthy participants (11 females) were
recruited through MRC-CBU volunteers’ panel (age 18
to 40 years). They gave written, informed consent
64
before the experiment, and were paid £8.50 for their
participation. All had normal or corrected to normal
vision.
2.6.2 Results
Figure 10. Trial 1 RTs of both 5-trial-short and 3-trial-long episodes were higher than those of the 3-trial-short episodes.
As evident in Figure 10 and as would be expected
from the previous results trial 1 RT of 5-trial-short
65
episodes was longer than that of 3-trial-short episodes
(paired t18 = 3.1, p=<0.01, 95% CI of difference = [23,
123], Cohen’s d = 0.71). Critically, despite having the
same number of trials, trial 1 RT for 3-trial-long
episodes was also significantly greater than for 3-trialshort episodes (paired t18 = 2.9, p= <0.01, 95% CI of
difference = [17, 103], Cohen’s d = 0.67). When RTs
for trial 1 of the longer duration/fewer steps 3-triallong
episodes
were
contrasted
with
the
shorter
duration/more steps 5-trial-short episodes there was no
difference (paired t18 = 0.6, p=0.5, 95% CI of
difference = [-37, 64]; Cohen’s d = 0.1)
In this experiment we sought to separate out the
effects of task episode duration from the number of
steps in the task in terms of the preparation time at the
beginning of the task episodes. The results suggest
that both are important. Despite having the same
66
number of steps, the anticipated longer-duration of the
3-step-long condition produced significantly longer
preparation times than the 3-steps-short condition.
Similarly, although the duration of the 5-step-short
tasks was only moderately longer than the 3-step-short
tasks, the increase in reaction time attributable to the
greater number of steps was broadly equivalent to that
of the 3-long vs. 3-short comparison (Cohen’s d 0.71 vs.
0.67). In this case the effect of more steps/shorter
duration appears to have been balanced with the fewer
steps/longer duration as no differences were observed
in response time between the 3-steps-long and 5-stepsshort tasks.
3.Discussion
67
We showed that every time a series of otherwise
independent and unpredictable trials was construed as
a task episode, behavior suggested that these trials
were not executed independent of each other but as
parts of a common subsuming routine that was
assembled at the beginning of the supposed episode:
(1) Trial 1 of the episode had the longest RT,
suggesting the additionally time taken to assemble the
routine structure. (2) Trial 1 RT was longer before
episodes that had more number of trials or that were
temporally longer, because longer episodes required
longer routines that took longer to assemble. (3)
Switch cost due to change in task items across
consecutive trials was absent on trial 1 because the
episodic routine subsumed the execution of trials of the
episode
allowing
consecutive
trial
related
configurations to interact within a routine only and not
across them because dismantling of the routine also
68
dismantled trial rule related configurations nestled
under it. In contrast to switch cost, Stroop interference
on a trial largely comes about from issues pertaining to
that
trial
and
not
because
of
configurations/traces/associations)
issues
related
to
(e.g.
the
previous trials, hence Stroop cost was not absent on
trial 1. Additionally, we found that trials construed as
parts of a task episode were executed faster and had
better control for switching rules and overcoming
Stroop
interference
than
trials
executed
as
independent tasks. This would be expected because the
higher task related cognitive state through which trials
of the episode were executed had been prepared for as
part
of
the
routine
structure
and
instantiated
automatically across time. However, for independent
task
trials
such
states
had
to
be
individually
instantiated before each trial execution.
69
Analogous to higher trial 1 RT seen in the current
study, neuroimaging of extended tasks showed higher
and more widespread activity at
the completion of
construed task episodes suggesting that the routine
assembled
at
the
beginning
dismantled
at
completion
of
the
(Farooqui
et
episode
al.
is
2012;
Desrochers et al. 2015). The intensity and spread of
this activity is related to the hierarchical level of the
completed
episode
-
task
completion
>
subtask
completion, suggesting a hierarchical arrangement of
routines related to tasks and subtasks. Completion of a
subtask dismantles the routine related to it leaving the
overarching routine intact, hence elicits less activity
compared to the completion of the main task, which
dismantles routines at all levels. This is conceptually
similar to the current finding of absence of switch cost
across episode boundaries.
70
Neuroimaging
of
experiments
similar
to
experiments 2 and 3 (Farooqui et al. in preparation)
showed a widespread sequential increase in activity
across
trials
making
up
the
task
episode.
This
dependence of activity on the sequential position in the
construed task episode again suggested that these
trials were executed as parts of a larger routine that
changed
systematically
across
the
length
of
the
episode causing a systematic change in activity across
the sequential but identical trials forming the episode.
Episodic Routines
The current study suggested that our intuition that
we execute extended task identities may be correct
because the corresponding behavior although extended
is executed through a single act. This act generates as
a single routine the entire series of cognitive states
that will go on to subsume the execution of the
71
corresponding behavior. Importantly, this happens even
when the task items to be executed within the episode
are unknown or unpredictable, because the cognitive
states generated through this routine are related to the
task identity being executed. Hence, it is the construal
that task items to follow, whether predictable or not,
will be executed as parts of a higher task episode that
creates this subsuming episodic routine because such a
construal implies that these items will be executed
through the cognitive states corresponding to the
construed task, and as components of the larger task.
The cognitive states created through this routine may
correspond to the subjects’ notion of the task being
executed (Vallacher & Wegner, 1989) and correspond
to the widespread changes in cognition based on her
knowledge of what that task entails. This knowledge
may specify a particular step, or direct her to the
relevant
environmental
contingencies
that
will
72
determine the identity of that step, or just involve
being on-task.
For executing any extended task, cognition not
only has to execute the set of overt behavior but also
needs to execute the widespread cognitive changes
that bring about that overt behavior. Overt behavior,
after all, is the proverbial tip of the mass of changes in
cognition
that
take
place
during
any
purposeful
activity. It is well recognized that goal directed
cognition requisites specific organization that brings to
fore task relevant learnings, memories, dispositions,
knowledge and expectancies, and the corresponding
configurational
changes
in
various
perceptual,
attentional, mnemonic, and motor processes so that of
the multiple possibilities ongoing cognitive processing
takes the one that brings cognition and the state of the
environment closer to the intended goal (e.g. Bartlett,
1932; Dashiel, 1940; Miller, Galanter & Pribram, 1960;
73
Rogers & Monsell, 1995; Mayr & Keele, 2000; Logan &
Gordon, 2001). The episodic routine may instantiate
the automatic flow of task relevant knowledge and of
the widespread changes in cognition that follow from
or are linked to this knowledge – the most basic of
which is the conscious knowledge of the task being
executed.
In the current experiments participants knew that
the task consisted of a particular number of trials and a
particular period of time (or a probabilistic estimate of
these, e.g. in experiment 5), and that the component
trials involved (e.g.) right handed response, attention
on the numbers appearing in the area around the
center of the screen, decisions that compared the two
numbers. It is possible that the sequence of cognitive
states
instantiated
corresponded
to
the
by
the
subjects’
episodic
knowledge
routine
of
the
construed task and expectations related to that task. It
74
therefore
could
have
included
increased
motor
preparation of the right hand and relevant attentional
and rule set on central vision, either throughout the
episode or specifically during instances when trials
were expected and less so when inter-trial intervals
were expected.
If, additionally, the identity and sequence of task
items to be executed on individual trials was known
beforehand as is the case in task sequence studies, the
routine could additionally have specified the identity
and sequence of individual task items, and would have
created more specific preparatory states (trial related
cognitive set). If, further, a foreknowledge of what
motor actions to make across time was also available,
the routine would have acted as a motor program
(Keele, Cohen & Ivry, 1990).
The presence of foreknowledge about the steps
affects the nature of prior preparation and the
75
additional content of the higher task related cognitive
states specified by the episodic routine and not the
presence of the episodic routine and the task related
cognitive state specified by it. The phenomenon in the
current study is, thus, a more general version of the
phenomena
seen
when
explicit
task
or
motor
sequences are executed (Rosenbaum et al. 1983;
Schneider & Logan, 2006). The principle underlying all
of these is the hierarchical nature of cognitive action
whereby abstract and temporally extended actions
aimed at higher level tasks subsumes and sets the
context for more concrete and temporally limited
actions aimed at subtasks. Construal of task - i.e. what
is the task to be executed - determines the nature of
the higher level action that organizes cognition across
time and sets the ground for lower level and more
concrete
action,
and
therefore
may
be
the
key
76
determinant
of
cognitive
organization
during
purposeful behavior.
Episodes in Cognition
Analogous to the current suggestion that purposive
behavior
may
be
executed
through
episodes
corresponding to extended task constructs, Zack and
Tversky (2001) suggested that perception occurs
through episodic which they termed ‘events’. When
participants
viewed
a
video
(e.g.
of
preparing
breakfast) they reliably organized their viewing into
episodes
constructs
corresponding
(e.g.
to
preparing
temporally
tea,
toasting
extended
bread).
Importantly, akin to Farooqui et al. 2012, greater
activity was elicited at the boundaries of higher level
perceptual episodes (e.g. preparing tea) than at the
boundaries of a lower level episodes (e.g. boiling
water; Zacks et al. 2001). They further suggested that
77
perceptual episodes (or ‘events’) may be periods
wherein a subsuming perceptual construct provides a
prediction
template
to
interpret
the
incoming
sensations and that these templates get changed at
episode boundaries (Kurby & Zacks, 2007). This again
is analogous to our claim that during the execution of
task episodes the subsuming episodic routine following
from the active task construct provides the template
for relevant processing and action generation. We
would further predict that the higher level cognitive
states related to the perceptual construct would be
instantiated as a routine.
The notion of an extended period of behavior being
instantiated as a single routine has in the past been
used in situations that are habitual and rigid whereby
once
instantiated
the
corresponding
behavior
automatically runs off to completion. In contrast, we
have evidenced abstract routines that unfolded into a
78
sequence of cognitive states that could subsume
different kinds of behavior. These routines were open
in that they could not run-off on their own but
depended on more deliberative trial executions to
move forward in time.
Other studies have also evidenced the presence of
such abstract routines that neither correspond to a
predetermined sequence of behavior nor run-off on
their own.
Cushman and Morris (2015) recently
provided formal evidence that a series of sub-goals
may get assembled into a routine for selection as a
single
entity
while
their
deliberative
decisions/actions.
suggestion
that
(implementation
future
intention)
execution
required
Gollwitzer’s
(1999)
self-regulatory
goals
may
be
automatically
implemented is also akin to an abstract routine e.g. If
my implementation intention is to be friendly to an
unpleasant person in a meeting I may behave in a
79
myriad ways to fulfill this goal. When I meet him I may
smile, do an extra effort at striking and maintaining a
conversation, modulate my emotions and so forth. The
sequence of behavior I end up undertaking to appear
friendly could not have been pre-specified by the
implementation intention because the actual behavior
depended on situations that occurred in the meeting,
the response the supposed unpleasant person gave and
so forth. At the same time, if I am socially adept, I
would not have to make an active high level decision
before each of step of the sequence of polite behavior.
Most importantly, this notion of task episode
execution is in line with emerging ways of thinking
about goals. Tasks and goals are intimately related
constructs. Tasks culminate in goals, and connect the
intending of a goal to its achievement. A wide variety
of frameworks accept that goals are important for the
control and execution of tasks that lead to their
80
achievement (James, 1890; Lewin, 1926; Greenwald,
1972; Prinz, 1987; Jeannerod, 1988; Meyer & Kieras
1997; Gollwitzer & Sheeran, 2006; Anderson, 2014),
and hence may correspond to some cognitive entity
that is active during the task episode (James, 1890;
Kruglanski & Kopetz, 2009). The episodic routine may
be the cognitive bridge that connects goal to the period
of its execution by organizing cognition across time
and ensuring its goal directedness from the highest
level.
Automatic and deliberative aspects of cognition are
frequently
considered
as
separate
systems
(Kahnemann, 2011; Evans & Stanovich, 2013). A sharp
distinction is usually made between situations where
extended behavior comes about through deliberate
decisions made at each step, and those where the
extended behavior is selected as a routine (e.g. Dolan
& Dayan, 2013). In contrast, current study suggested
81
the
simultaneous
existence
of
automatic
and
deliberative aspects of cognition. The deliberative
execution of randomly switching rules on individual
trials was subsumed by an automatically unfolding
episodic routine that even eased the deliberativeness
of the trials it subsumed.
Alternate Accounts
Hierarchy accounts of goal directed behavior differ
in the form of hierarchy emphasized (Broadbent, 1977;
Cohen, 2000; Botvinick, 2008; Badre & D’Esposito,
2009; Shallice
&
Cooper, 2011). Frequently,
the
emphasis is placed on the hierarchical nature of task
item related knowledge and representations in memory
(e.g. Bower, 1970; Minsky, 1975; Schank & Abelson,
1977; Cooper & Shallice, 2006; Schneider & Logan,
2006). In these accounts beginning a task requires
accessing/establishing
task
item
related
82
representations in long term/working memory and
hence the delay at step 1. Likewise, retrieving higher
task related representation is thought to remove
memory traces related to lower level task items
causing absence of item level switch cost at position 1.
The success of such accounts in explaining the
current results depends upon on their proposed nature
of task item representations. Accounts that take these
representations to be abstract and propositional cannot
explain the current observations that were made in
absence
of
propositional
representations
corresponding to individual trials.
Other versions of such accounts may claim that
what
is
hierarchically
propositional
task
stored
item
in
memory
representations
is
not
but
the
episodic memory of task execution (e.g. Medin &
Shaffer, 1978; Jacoby & Brooks, 1984; Logan, 1988;
Trammell, 1997). It is possible that instances of task
83
execution generate a whole host of episodic (explicit
and implicit) memories (e.g. Jacoby & Brooks, 1984;
Logan, 1988), with re-execution strengthening the
common elements across different iterations of the
task episode and whittling away those that were
specific
to
individual
instances,
and
thus
create
episodic memories of task execution. Later execution
of task episodes will occur through the recall and
instantiation of the episodic memory of that task
episode causing the signs of hierarchical cognition.
This account, however, does not explain what
makes the segment of executed behavior to be
represented as a single episodic entity in the first place
for the memory of its execution to be represented as
one chunk. It presupposes it. The current account
claims that construal of a segment of behavior as a
single task results in its execution as a single episodic
entity, which consequently may facilitate the memory
84
generated by that segment to be stored as discrete
episodic chunks.
Some accounts emphasize the hierarchical nature
of control processes in terms of their relational and
decision complexity e.g. single level decision (red color
→ left key press) vs two-level decisions (if circle then
red color → left key press, but if square then red color
→ right key press) (Christoff et al. 2009; Badre &
D’Esposito, 2009). In the current study, however,
individual trials of an episode were identical in these
aspects.
Other
accounts
have
emphasized
the
temporally extended nature of higher level control
referred to as episodic control (e.g. Koechlin &
Summerfield, 2007; Braver, 2012). Such control may be
needed to maintain rules and decisions that are unique
to an episode, or to maintain an ongoing task in
abeyance while a branch of the task is being executed.
In most of the current experiments all trials of an
85
experiment block were identical in these terms, and
the boundaries of the conceived episodes were not
accompanied by changes in control of the kind
emphasized by these accounts.
Conclusion
In summary, we showed that a series of otherwise
independent trials when construed as a task is
executed
as
a
single
cognitive
act
through
a
preassembled episodic routine. We suggested that this
routine creates the series of cognitive states related to
the
construed
task
identity
being
executed.
We
speculate that such episodic routines represent the
means
through
which
cognition
is
sequentially
86
organized across time in line with the temporally
extended nature of the task and goal being pursued.
4.References
Anderson, J. R., Bothell, D., Lebiere, C., & Matessa, M. (1998). An integrated
theory of list memory. Journal of Memory and Language, 38(4), 341-380.
Anderson, J. R. (2014). Rules of the mind. Psychology Press.
Anderson, J. R., & Matessa, M. (1997). A production system theory of serial
memory. Psychological Review, 104(4), 728.
Badre, D., & D’Esposito, M. (2009). Is the rostro-caudal axis of the frontal lobe
hierarchical? Nature Reviews. Neuroscience, 10(9), 659–69.
Bartlett, S. F. C., & Bartlett, F. C. (1932). Remembering: A Study in
Experimental and Social Psychology. Cambridge University Press
Botvinick, M. M. (2008). Hierarchical models of behavior and prefrontal
function. Trends in Cognitive Sciences, 12(5), 201–8.
Bower, G. H. (1970). Organizational factors in memory. Cognitive Psychology,
1(1), 18–46.
87
Braver, T. S. (2012). The variable nature of cognitive control: a dual
mechanisms framework. Trends in Cognitive Sciences, 16(2), 106–113.
Broadbent, D. E. (1977). Levels, hierarchies, and the locus of control.
Quarterly Journal of Experimental Psychology, 29(2), 181–201.
Christoff, K., Keramatian, K., Gordon, A. M., Smith, R., & Mädler, B. (2009).
Prefrontal organization of cognitive control according to levels of
abstraction. Brain Research, 1286, 94–105.
Cohen, G., (2000). Hierarchical models in cognition: Do they have
psychological reality? European Journal of Cognitive Psychology, 12(1), pp.1–
36.
Cooper, R. P., & Shallice, T. (2006). Hierarchical Schemas and Goals in the
Control of Sequential Behavior. Psychological Review, 113(4), 887-916.
Dashiell, J. F. (1940). A neglected fourth dimension to psychological research.
Psychological Review, 47(4), 289.
De Jong, R., (1995). The Role of Preparation in Overlapping-task Performance.
The Quarterly Journal of Experimental Psychology Section A, 48(1), pp.2–25.
Desrochers, T. M., Chatham, C. H., & Badre, D. (2015). The Necessity of
Rostrolateral Prefrontal Cortex for Higher-Level Sequential Behavior.
Neuron, 87(6), 1357–68.
Dezfouli, A., Lingawi, N. W., & Balleine, B. W. (2014). Habits as action
sequences: hierarchical action control and changes in outcome value.
Philosophical Transactions of the Royal Society of London. Series B,
Biological Sciences, 369(1655).
Dolan, R. J., & Dayan, P. (2013). Goals and habits in the brain. Neuron, 80(2),
312–25.
Evans, J. S. B. T., & Stanovich, K. E. (2013). Dual-Process Theories of Higher
Cognition: Advancing the Debate. Perspectives on Psychological Science, 8(3),
223–241
Farooqui, A. A., Mitchell, D., Thompson, R., & Duncan, J. (2012). Hierarchical
organization of cognition reflected in distributed frontoparietal activity.
The Journal of Neuroscience, 32(48), 17373–81.
Gobet, F., & Simon, H. A. (1998). Expert chess memory: revisiting the
chunking hypothesis. Memory (Hove, England), 6(3), 225–55.
Gollwitzer, P. M. (1999). Implementation intentions. Strong effects of simple
plans. American Psychologist, 54(7), 1999, 493
Gollwitzer, P. M., & Sheeran, P. (2006). Implementation Intentions and Goal
Achievement: A Meta‐analysis of Effects and Processes. Advances in
Experimental Social Psychology, 38, 69–119.
Greenwald, A. G. (1972). On doing two things at once: time sharing as a
function of ideomotor compatibility. Journal of experimental psychology, 94(1),
52.
88
Henry, F.M. & Rogers, D.E., (1960). Increased Response Latency for
Complicated Movements and A “Memory Drum” Theory of Neuromotor
Reaction. Research Quarterly. American Association for Health, Physical
Education and Recreation, 31(3), pp.448–458.
Jacoby, L. L., & Brooks, L. R. (1984). Nonanalytic Cognition: Memory,
Perception, and Concept Learning. Psychology of Learning and
Motivation, 18, 1–47.
James, W., (1890). The Principles of Psychology, Volume 1, H. Holt.
Jeannerod, M. (1988). The neural and behavioural organization of goaldirected movements. Oxford University Press.
Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux
Keele, S. W., Cohen, A., & Ivry, R., (1990). Motor programs: Concepts and
issues. In Jeannerod, M. (Ed.). (1990). Attention and performance XIII: Motor
representation and control (Vol. 13). Lawrence Erlbaum.
Klapp, S. T., Anderson, W. G., & Berrian, R. W. (1973). Implicit speech in
reading: Reconsidered. Journal of Experimental Psychology, 100(2), 368.
Koechlin, E., Ody, C., & Kouneiher, F. (2003). The architecture of cognitive
control in the human prefrontal cortex. Science, 302(5648), 1181-1185.
Kruglanski, A. W., & Köpetz, C. (2009). What is so special (and nonspecial)
about goals. The psychology of goals. Guilford Press, New York, NY, 27-55.
Kurby, C.A. & Zacks, J.M., (2008). Segmentation in the perception and memory
of events. Trends in cognitive sciences, 12(2), pp.72–9.
Lewin, K., (1926). Vorsatz, Wille und Bedürfnis. Psychologische Forschung,
7(1), pp.330–385.
Lien, M. C., & Ruthruff, E. (2004). Task switching in a hierarchical task
structure: evidence for the fragility of the task repetition benefit. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 30(3), 697.
Logan, G. D. (1988). Toward an instance theory of automatization.
Psychological Review, 95(4), 492–527.
Logan, G.D., (2004). Working memory, task switching, and executive control in
the task span procedure. Journal of experimental psychology. General,
133(2), pp.218–36.
Luria, R., & Meiran, N. (2003). Online order control in the psychological
refractory period paradigm. Journal of Experimental Psychology. Human
Perception and Performance, 29(3), 556–74
Mayr, U., & Keele, S. W. (2000). Changing internal constraints on action: the
role of backward inhibition. Journal of Experimental Psychology: General,
129(1), 4.
Medin, D. L., & Schaffer, M. M. (1978). Context theory of classification
learning. Psychological Review, 85(3), 1978, 207-238
89
Meyer, D. E., & Kieras, D. E. (1997). A computational theory of executive
cognitive processes and multiple-task performance: Part I. Basic mechanisms.
Psychological review, 104(1), 3.
Miller, G. A., Galanter, E., & Pribram, K. H. (1986). Plans and the Structure of
Behavior. Adams Bannister Cox.
Minsky, M. (1975). A Framework for Representing Knowledge. In P. Winston
(Ed.), The Psychology of Computer Vision, McGraw-Hill
Monsell, S., (2003). Task switching. Trends in cognitive sciences. 7(3), pp.134–
140.
Neill, W. T. (1997). Episodic retrieval in negative priming and repetition
priming. Journal of Experimental Psychology. Learning, Memory, and
Cognition, 23(6), 1291–1305.
Newell, A., & Simon, H. A. (Herbert A. (1972). Human problem solving.
Prentice-Hall.
Prinz, W. (1987). Ideo-motor action. Lawrence Erlbaum Associates, Inc.
Rogers, R. D., & Monsell, S. (1995). Costs of a predictible switch between
simple cognitive tasks. Journal of experimental psychology: General, 124(2),
207.
Rosenbaum, D. A., Kenny, S. B., & Derr, M. A. (1983). Hierarchical control of
rapid movement sequences. Journal of Experimental Psychology. Human
Perception and Performance, 9(1), 86–102.
Schank, R. C., & Abelson, R. P. (1977). Scripts, plans, goals, and
understanding : an inquiry into human knowledge structures. L. Erlbaum
Associates.
Schneider, D.W. & Logan, G.D., (2015). Chunking away task-switch costs: a
test of the chunk-point hypothesis. Psychonomic bulletin & review, 22(3),
pp.884–9.
Schneider, D. W., & Logan, G. D. (2006). Hierarchical control of cognitive
processes: Switching tasks in sequences. Journal of Experimental Psychology:
General, 135(4), 623.
Schneider, D. W., & Logan, G. D. (2007). Retrieving information from a
hierarchical plan. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 33(6), 1076.
Shallice, T., & Cooper, R. (2011). The organisation of mind. Oxford University
Press.
Sigala, N., Kusunoki, M., Nimmo-Smith, I., Gaffan, D., & Duncan, J. (2008).
Hierarchical coding for sequential task events in the monkey prefrontal
cortex. Proceedings of the National Academy of Sciences, 105(33), 1196911974.
90
Sternberg, S., Monsell, S., Knoll, R. L., & Wright, C. E. (1978). The latency and
duration of rapid movement sequences: Comparisons of speech and
typewriting. Information processing in motor control and learning, 117-152.
Stokes, M. G., Kusunoki, M., Sigala, N., Nili, H., Gaffan, D., & Duncan, J.
(2013). Dynamic coding for cognitive control in prefrontal cortex. Neuron,
78(2), 364-375.
Vallacher, R. R., & Wegner, D. M. (1989). Levels of personal agency: Individual
variation in action identification. Journal of Personality and Social Psychology,
57(4), 660.
Zacks, J. M., & Tversky, B. (2001). Event structure in perception and
conception. Psychological Bulletin, 127(1), 3–21.
Zacks, J. M., Braver, T. S., Sheridan, M. A., Donaldson, D. I., Snyder, A. Z.,
Ollinger, J. M., … Raichle, M. E. (2001). Human brain activity time-locked
to perceptual event boundaries. Nature Neuroscience, 4(6), 651–655.
91