You are on page 1of 7

Acta Psychologica 164 (2016) 3945

Contents lists available at ScienceDirect

Acta Psychologica
journal homepage: www.elsevier.com/locate/actpsy

Contingency learning is not affected by conict experience: Evidence


from a task conict-free, item-specic Stroop paradigm
Yulia Levin a,, Joseph Tzelgov b
a
b

Department of Psychology and Zlotowski Center for Neuroscience, Ben-Gurion University of the Negev, Israel
Department of Psychology and Zlotowski Center for Neuroscience, Department of Brain and Cognitive Sciences, Ben-Gurion University of the Negev, and Achva Academic College, Israel

a r t i c l e

i n f o

Article history:
Received 12 February 2015
Received in revised form 14 December 2015
Accepted 15 December 2015
Available online xxxx
Keywords:
Automaticity of reading
Item-specic proportion congruent effect
Informational conict
Task conict
Contingency learning
Cognitive control

a b s t r a c t
A contingency learning account of the item-specic proportion congruent effect has been described as an associative stimulusresponse learning process that has nothing to do with controlling the Stroop conict. As supportive evidence, contingency learning has been demonstrated with response conict-free stimuli, such as
neutral words. However, what gives rise to response conict and to Stroop interference in general is task conict.
The present study investigated whether task conict can constitute a trigger or, alternatively, a booster to the
contingency learning process. This was done by employing a task conict-free condition (i.e., geometric
shapes) and comparing it with a task conict condition (i.e., neutral words). The results showed a signicant
contingency learning effect in both conditions, refuting the possibility that contingency learning is triggered by
the presence of a task conict. Contingency learning was also not enhanced by the task conict experience, indicating its complete insensitivity to Stroop conict(s). Thus, the results showed no evidence that performance optimization as a result of contingency learning is greater under conict, implying that contingency learning is not
recruited to assist the control system to overcome conict.
2015 Elsevier B.V. All rights reserved.

1. Introduction
Reading is an acquired human ability to decode and interpret visual
lexical symbols. In adults, this ability is known to be automatic, that is, it
occurs whenever a lexical stimulus is encountered. The most dramatic
demonstration of the automaticity of the reading process is an interference effect obtained in the Stroop task (Stroop, 1935). In this task, participants have to name the color of visually presented words (e.g., blue for
the stimulus RED presented in blue ink) while ignoring their meaning
(e.g., the word RED). There is no need to read the words to accomplish
the task and yet reading occurs, as evidenced by slower response times
for incongruent stimuli (e.g., RED in blue ink) than for neutral letter
strings (e.g., XXXX in blue ink). The fact that reading takes place in
spite of the fact that it is not required, and even interferes with performance, demonstrates its automaticity (Perlman & Tzelgov, 2006).
The interference, or conict, produced by the automatic performance of the irrelevant reading task has been shown to be a target of
cognitive control. That is, when a conict becomes too strong, cognitive
control is able to reduce it. Much evidence for that ability of the cognitive system came from a bulk of studies that manipulated the proportion of congruent vs. incongruent stimuli to control the Stroop effect

Corresponding author at: Department of Psychology, Ben-Gurion University of the


Negev, P.O.B. 653, Beer Sheva 84105, Israel.
E-mail address: levinyu@post.bgu.ac.il (Y. Levin).

http://dx.doi.org/10.1016/j.actpsy.2015.12.009
0001-6918/ 2015 Elsevier B.V. All rights reserved.

(e.g. Logan, 1985; Logan & Zbrodoff, 1979). The main nding of these
studies, or what is known as the list-wide proportion-congruent effect, was that the magnitude of the observed interference effect was
smaller when the experienced conict was too strong (i.e., large proportion of incongruent trials in the list).
Several models have been proposed to explain the mechanism by
which conict is reduced in the Stroop task (a conict-monitoring
framework; Blais, Robidoux, Risko, & Besner, 2007; Botvinick, Braver,
Barch, Carter, & Cohen, 2001; Botvinick, Cohen, & Carter, 2004; De
Pisapia & Braver, 2006). According to Botvinick et al.'s (2001);
Botvinick et al.'s (2004) conict-monitoring architecture, increasing
the proportion of incongruent trials raises the amount of (response)
conict (i.e., stronger competition between the response activated by
the color-naming process and the irrelevant response activated by reading). The elevation in conict is detected by the conict-monitoring
unit, which in turn signals units responsible for control exertion. The
control is achieved through focusing attention on the relevant task.
This way the irrelevant reading task does not get much attention and
the conict it produces is considerably reduced.
It has also been proposed that the control system is not just able to
reduce the conict accumulated at the list level, but is also exible
enough to reduce the conict produced by specic items in the list
(Bugg, Jacoby, & Chanani, 2011; Bugg, Jacoby, & Toth, 2008; see also
Blais et al., 2007). The item-specic proportion-congruent effect
(Jacoby, Lindsay, & Hessels, 2003; Jacoby, McElree, & Trainham, 1999)
demonstrates that when the proportion of incongruent stimuli is

40

Y. Levin, J. Tzelgov / Acta Psychologica 164 (2016) 3945

manipulated at the level of specic words, the words mostly presented


as incongruent stimuli tend to produce less interference than those
mostly presented as congruent stimuli.
1.1. Cognitive control, learning and what is in between
The process of cognitive control is assumed to proceed in a way that
can be described as automatic, that is, without assuming any hidden
agency deciding when, where and how to intervene. As such, for the
last couple of decades there has been some tension in this research
eld to differentiate between the automatic control and simple learning mechanisms, or to dene how much the former may be relying on
the latter.
According to the proposal of Verguts and Notebaert (2008, 2009), a
simple learning process may be in fact in service of control. Specically, their proposal holds that the goal of control (i.e., conict reduction) can be achieved through associative (Hebbian) learning that
binds together all currently active (i.e., task-relevant) representations.
That is, according to this account the control is actually based on a learning process. It is important for the present discussion to note that except
for extending the general conict-monitoring theory (Botvinick et al.,
2001, 2004) by explaining how the system knows where to intervene,
it shares most of its other features. Thus, when the conict is sufciently
reduced, less learning occurs, which means learning in this situation is
dependent on and guided by the magnitude of the experienced conict.
In contrast, there are suggestions that learning does not represent a
mechanism in-service-of control, but separate cognitive phenomena
that sometimes might mimic the effects of control. That is, learning
is assumed to produce an independent (confounding) effect on reaction
time (RT) that happens to look like the effect of conict reduction attributed to cognitive control. In the context of the Stroop task, Schmidt,
Crump, Cheesman, and Besner (2007) proposed that the item-specic
proportion-congruent effect might be driven by such learning that has
nothing to do with (controlling the) conict. This contingency learning
account of the item-specic proportion-congruent effect is based on
the fact that in the original study of Jacoby et al.'s (2003), as well as in
the recent replications (Hazeltine & Mordkoff, 2014), faster responses
were also observed for congruent items in a mostly congruent condition, as compared to the condition where the probability of an item
appearing in a congruent or incongruent color was equal.1 This result
cannot be accounted for by assuming the intervention of cognitive control, since congruent items do not produce response conict, and therefore are not able to engage control (see also Levin & Tzelgov, 2014).
According to the contingency learning account (Schmidt, 2013a, 2013b;
Schmidt & Besner, 2008; Schmidt et al., 2007; for contrasting views
see Bugg & Hutchison, 2013; Bugg et al., 2011; Hutchison, 2011; see
also Abrahamse, Duthoo, Notebaert, & Risko, 2013; Atalay & Misirlisoy,
2012; Bugg, 2014), the item-specic proportion-congruent effect is better described as a speeding-up observed for the words frequently
appearing in a specic (be it congruent or incongruent) color, and is
due to the fact that manipulation of proportions at the item level creates
contingencies between specic words and responses. These contingencies are learned and subsequently used to predict responses. For example, if RED frequently appears in blue ink, the learned association would
be if the word is RED then push the blue button. Note, in contrast to
the learning-based control view (Verguts & Notebaert, 2008, 2009),
contingency learning is not assumed to be aided by response conict,
but rather to represent a general ability to bind stimuli and responses
on the basis of their existing correlations. The mechanism of contingency learning as implemented in the parallel-episodic processing
model (Schmidt, 2013a) has no feature that is able to measure the response conict, nor has it a property allowing for allocation of attention

A 50/50 condition in Jacoby et al.'s (2003) study.

in an adaptive manner, and yet it successfully simulates the pattern of


the item-specic proportion-congruent effect.
However, according to recently reported data, which will be
discussed shortly, there might be a third type of controllearning relationship that comes right in between the two aforementioned proposals
and which is at the focus of the present study. Recent studies showed
that implicit learning processes might not be completely independent
of conict as suggested by Schmidt et al. (2007) for contingency learning. However, the way they depend on conict does not t the
learning-based control put forward by Verguts and Notebaert (2008,
2009) (i.e., conict-monitoring framework) either. Deroost,
Vandenbossche, Zeischka, Coomans, and Soetens (2012) presented a
probabilistic sequence of the colors in the Stroop task, which was implicitly learned by the participants. They found that sequence learning
did not help to reduce the conict (i.e., the Stroop effect). Stroop conict
however, was shown to enhance the expression2 of learning: the acquired sequence knowledge was used more under conict
(i.e., incongruent) conditions than under conict-free (i.e., congruent
and neutral) conditions. Boosting effects of conict on implicit learning
have also been reported in other studies. Deroost and Soetens (2006)
observed a larger sequence learning effect for participants who were
trained with incompatible (i.e., conicting) than compatible stimulus
response mappings. Similarly, Zhao, Ngo, McKendrick, and TurkBrowne (2011) found that engaging in a secondary (i.e., interfering)
task during the training phase, as opposed to passive viewing, improved
statistical learning. Finally, Vandenbossche, Coomans, Homble, and
Deroost (2014) reported a larger sequence learning effect for aged
adults under high-interference (i.e., a dual task performed in the same
modality) than under low-interference (i.e., a cross-modal dual task)
training condition.
To summarize, the implicit sequence learning was not found to
serve control by reducing the conict (Deroost et al., 2012) as assumed
for learning-based control (Verguts & Notebaert, 2008, 2009). Yet, the
observed enhancement and stronger reliance on implicit learning in
conict environments (Vandenbossche et al., 2014; Zhao et al., 2011;
see also Deroost & Soetens, 2006; Koch, 2007, Exp. 1) speaks for the possibility that implicit learning processes might nevertheless contribute to
cognitive control. However, this may happen not by reducing the conict but according to Deroost et al. (2012), through optimization of the
task performance: Optimization of task performance was accomplished
by an increased reliance on implicit sequence knowledge under high
conict. This indicates that implicit learning processes can be exibly recruited to support cognitive control (p. 15). This idea might seem novel
in the domain of cognitive control, since the latter is traditionally described as the process that is sensitive to the amount of conict and
that aims to minimize this conict when it gets too strong. However,
considering the control process more broadly makes it is perfectly
clear that the reduction of the conict is only needed to ensure a good
level of performance in the ongoing task. Stated otherwise, the nal
goal of the control process is to protect the performance from the
conict-related decline. Bugg's (2014) study provides empirical support
for such a view of cognitive control. It was shown in a series of experiments that in a high-conict context, cognitive control was only engaged as a last resource, when stimulusresponse associations did
not allow maintaining a sufcient level of performance. Thus, the magnitude of the conict seems only to matter when it has a detrimental effect on performance. This emphasizes the importance of the
performance rather than conict per se in the context of control
engagement.
One way to preserve the required performance when conict arises,
as suggested by the conict-monitoring theory, is by reducing the

2
As opposed to acquisition of learning that was not affected by the amount of conict
(manipulated by the proportion of congruent trials) at the training phase.

Y. Levin, J. Tzelgov / Acta Psychologica 164 (2016) 3945

conict. This is obviously the most efcient way that solves the problem
from the rootif you want to protect the performance from interference, eliminate the cause of the interference. However, as the results
of the aforementioned studies suggest, there might be another way to
assist maintaining good performance in the face of conict, namely,
recruitment of the implicit learning processes and stronger reliance on
what was learned by these processes. For example, learning what button should be pushed when the word that appears is red
(i.e., contingency learning) allows shortening the time that is needed
to respond, thus promoting better performance. Such performanceboosting learning seems to be especially useful when performance has
been damaged by conict. That is, it is conceivable that in conict situations, the control system, along with its attempts to reduce the conict,
might also engage/enhance such an implicit learning process that may
help to maintain performance in a faster and resource-saving way.

1.2. Possible contribution of the contingency learning to cognitive control


Regarding Stroop contingency learning, it has been claimed that this
learning process does not represent learning-based control (as proposed by Verguts & Notebaert, 2008, 2009). This claim was supported
by empirical data demonstrating a contingency learning effect with
neutral color-unrelated words (Schmidt et al., 2007). By showing that
the effect of contingency learning was observable even when the congruency variable was not manipulated, the authors claimed that it had
to be regarded as an independent process that had nothing to do with
control. One of the aims of the current study is to complete the attempt
of Schmidt et al. to disregard contingency learning as a mechanism contributing to cognitive control. Whereas Schmidt et al. only considered
one way in which learning might contribute to controlthe way suggested by Verguts and Notebaertwe suggest that there is another option that should be considered as well. Specically, we investigated
whether contingency learning in the Stroop task might represent the
assisting type of the controllearning relationship, where the contingency learning is recruited not to reduce the conict, but to help the
control system optimize performance (Deroost et al., 2012) in some of
the trials.
What gives rise to the latter possibility is a notion of the multipleconict nature of Stroop interference. Using neutral words in Schmidt
et al.'s (2007) and Schmidt and Besner's (2008) experiments did not
eliminate the Stroop conict completely. This is because response conict is not the only conict known to contribute to Stroop interference
(Goldfarb & Henik, 2007; MacLeod & MacDonald, 2000; for contrasting
views see Melara & Algom, 2003). In fact, response conict is a direct
outcome of the parallel activation of two color concepts in the semantic
network, which is frequently referred to as the informational conict
(Goldfarb & Henik, 2007; MacLeod & MacDonald, 2000).3 The activation
of one color concept represents the retrieval of the word's meaning as a
result of the reading process, whereas the activation of the second color
concept represents processing of the color of the ink the word is written
in. Note, however, that even when the reading does not result in activation of the irrelevant color concept, for example, when color-unrelated
stimuli such as neutral words (e.g., DOG) or letter strings (e.g., LGFD/
XXXX) are used, the interference effect, though of smaller magnitude,
is still observed (Brown, 2011; Klein, 1964; Sharma & McKenna,
1998). This fact indicates that the main origin of Stroop interference is
not an informational incompatibility that leads eventually to the

3
Note that in a typical Stroop experiment, each color concept requires a different vocal
response, so in such experiments, response conict and informational conict are confounded (but for a manual response design allowing decoupling of this confound see De
Houwer, 2003). Thus, in the present article we do not distinguish between these two types
of conict and when mentioning response conict we refer to both the informational
and response conicts.

41

response conict, but a competition between two possible tasks that


one can perform when the stimulus has lexical propertiesreading,
which is automatically activated while not being a part of the task requirement (Tzelgov, 1997), and the color-naming task.4 The competition between the two tasks has been conceptualized by MacLeod and
MacDonald (2000) as well as by Goldfarb and Henik (2007) to be the
task conict, the existence of which has been well-documented by behavioral (Entel, Tzelgov, Bereby-Meyer, & Shahar, 2014; Kalanthroff,
Goldfarb, & Henik, 2013) as well as neuroimaging data (Bench et al.,
1993; see also Aarts, Roelofs, & van Turennout, 2009; Steinhauser &
Hbner, 2009).
As evident from this discussion, interference due to task conict can
be obtained as long as the stimulus can be read, regardless of whether it
is related to color or not (for an analysis further delineating the contribution of each type of conict to the Stroop effect see Levin & Tzelgov,
2015). Hence, using color-unrelated words in Schmidt and Besner's
(2008) and Schmidt et al.'s (2007) studies eliminated the response conict but not the main component of the Stroop interferencethe task
conict that was experienced by the participants in every trial. Therefore, in this experiment the contingency learning effect was only observed under a conict condition. This raises a question whether
constant presence of the task conict in Schmidt and Besner's and
Schmidt et al.'s experiments could enhance (or even trigger) the contingency learning process in order to optimize performance by facilitating
the RTs on some of the trials. Answering this question and refuting the
possibility that contingency learning may be involved in assisting control by optimizing performance requires evidence that it can also be observed in a Stroop situation where no task conict is produced by the
stimuli. To that end, in one condition we employed an item-specic paradigm with stimuli that could not be read (i.e., geometric shapes), and
the second condition was identical to that used by Schmidt et al.
(2007) and used neutral words as stimuli. Comparing the performance
in these conditions would allow rst, to test whether contingency
learning is triggered5 by conict experience. If that is the case, we
should not expect to observe the contingency learning effect in the
shapes condition, but only in the words condition. Second, if the contingency learning effect is observed with shapes stimuli, it is possible to
test whether its effect, as has been shown for other implicit learning
processes, is boosted by the conict experience. The latter would express itself in a larger contingency learning effect under a task conict
(i.e., neutral words) condition. Results supportive of at least one of
these hypotheses would indicate that contingency learning is recruited
as a performance-optimizing tool assisting the control system to overcome conict.

2. Method
2.1. Participants
Forty-two undergraduate students at Ben-Gurion University of the
Negev (25 females and 17 males, mean age = 25 years old, SD = 2.3),
who were native speakers of Hebrew, participated in the experiment
and were paid 25 NIS. All participants reported having normal or
corrected-to-normal vision acuity, as well as normal color vision. No

4
Note, by main origin we do not mean the task conict contributes most of the Stroop
effect in a classic color-word task, but that it has a key role in initiation of the interference.
Simply put, without engaging in the irrelevant reading task, there would be no Stroop interference no matter what stimulus type (incongruent/neutral word) is used.
5
Based on the results of Deroost et al. (2012) showing that conict only affects the expression but not the acquisition of learning, we believe that the triggering is a less likely
scenario than the enhancement. However, since we use another paradigm than the one
used originally by Deroost et al., it would not be experimentally correct not to consider this
possibility as well.

42

Y. Levin, J. Tzelgov / Acta Psychologica 164 (2016) 3945

one reported having a learning disability or attention decit/hyperactivity disorder.

2.2. Materials
For half of the participants the stimuli were four-letter neutral words
(plane, cake, doll, cat) in Hebrew, and for the other half the stimuli were
geometric shapes (circle, triangle, rhombus, rectangle). All word stimuli
were printed in 42-point David font and subtended a visual angle of 1.3
degrees vertically and 2.9 degrees horizontally from a viewing distance
of 60 cm. The shape stimuli subtended a visual angle of 1.7 degrees vertically and 1.6 degrees horizontally. Importantly, both stimulus conditions were matched with respect to the amount of perceptual
information. Thus, each shape was made up of relatively the same number of pixels as the mean number of pixels within the words (1152.75
and 1117.25 pixels for shapes and words, respectively).
Stimuli could appear in one of the four colors: red, blue, green,
and yellow. Color was not allowed to repeat in consecutive trials.
There were 384 trials in the experiment. In the high-contingency condition, each of four words/shapes was paired with a specic (randomly selected for each item) color in 72 trials, resulting in a total
of 288 high-contingency trials (75%). In the low-contingency condition,
each of four words/shapes was presented 8 times in each of the three
remaining colors, resulting in a total of 96 low-contingency trials
(25%).

2.3. Procedure
The stimuli were presented on a Dell 19-inch monitor with a resolution of 1280 1024 pixels. Stimulus presentation was controlled by
OpenSesame 2.8.0 software (Matht, Schreij, & Theeuwes, 2012). Participants sat approximately 60 cm from the computer screen. They
were told to name the color of the stimulus appearing at the center of
the screen as accurately and as fast as possible. Responses were made
manually by pressing a key on a keyboard: a for blue, z for green,
k for red, and m for yellow. The response buttons were marked
with appropriate colored stickers.
The experiment started with a 32-trial practice session. During the
practice, each stimulus appeared equally often in each color. Following the practice, participants performed 384 experimental trials. Participants were given two 2-second breaks during the experiment
session.
The experimental trial started with a xation white cross that
remained at the center of the screen until the participant pressed the
SPACE bar with their thumb. The xation was followed by a blank
screen for 500 ms. After that the target appeared and remained visible
until a response was made or for 2000 ms. For every trial in which the
participant was not fast enough to respond, a feedback message was
presented on screen for 1000 ms: You did not respond! Please
focus! A trial ended with a blank black display that was presented for
300 ms. In addition, feedback was given to the participants each time
they made a mistake: You made a mistake! Please focus! The feedback
message appeared on the screen for 1000 ms.

2.4. Design
The design of the experiment included two variables. Stimulus category (neutral words/shapes) was treated as a between participant factor. Participants were assigned to each level of the stimulus category
in a counterbalanced manner. The second variable was contingency
level (high contingency/low contingency) and was manipulated within
participants.

3. Results and discussion


One participant was replaced because he mistakenly quit the experiment before the end. The data collected from each participant was
trimmed so that no RT outliers (i.e., RTs b 250 ms and RTs =
2000 ms6) were included in the analysis of variance (0.06% of the
data). RTs of error trials were also excluded from the analysis of variance. Due to a very low error rate (0.8% of the data) in the present experiment, no separate analysis was conducted for accuracy. All effects
were tested at the signicance level () of .05.
The two-way contingency X stimulus category analysis of variance
revealed a signicant main effect for contingency, showing that across
stimulus category, RTs for high-contingency items were faster than for
low-contingency items, F (1, 40) = 50.6, MSE = 215, 2p = .56,
p b .001. No main effect was obtained for stimulus category, F (1,
40) = 2, MSE = 17,494, 2p = .05, p = .164.7 The interaction between
the two variables was signicant, F (1, 40) = 6.3, MSE = 215, 2p =
.13, p = .016, demonstrating a twice as large contingency learning effect
for shapes, F (1, 40) = 46.2, MSE = 215, 2p = .54, p b .001, than for
word stimuli, F (1, 40) = 10.6, MSE = 215, 2p = .21, p = .002 (see
Fig. 1).
First of all, the present data provide an additional replication of previous ndings (Schmidt & Besner, 2008; Schmidt et al., 2007), demonstrating that contingency learning does not depend on the presence of
response conict caused by incongruent color words, but is also evident
with response conict-free neutral words. That is, our data support the
notion that contingency learning does not function as a learningbased control mechanism that aims to reduce Stroop interference.
As for the critical question raised by the current study, namely,
whether constant experience of the task conict caused by neutral
words in Schmidt and Besner's (2008) and Schmidt et al.'s (2007) studies could enhance or even trigger the contingency learning, the results
do not supportive either of these possibilities. The contingency learning
effect was observed with unreadable task conict-free geometric
shapes. Hence, the present study shows that although contingency
learning in the Stroop task was originally demonstrated under a condition that included task conict, it was likely not dependent on
(i.e., triggered by) this conict experience. As for the less extreme possibility, according to which contingency learning, similar to other implicit
learning processes (Deroost & Soetens, 2006; Deroost et al., 2012;
Vandenbossche et al., 2014; Zhao et al., 2011), can be enhanced by conict experience, our data proved to be inconsistent as well. Contrary to
the ndings of the previous studies on implicit learning, we found a
smaller rather than larger contingency learning effect in a conicting
(i.e., words) than in a conict-free (i.e., shapes) condition. Finding a
contingency learning effect in both a conict and a conict-free condition with no indication of the enlargement of the learning effect as a
consequence of the constant conict experience, refutes the possibility
that contingency learning is involved in what we called an assisting
type of controllearning relationship. A demonstrated, immunity of
the contingency learning process to the presence of the (task) conict
implies that it is not recruited by the control system as a tool allowing
improvement of performance in the face of conict. In that sense, our
study completes the attempt of contingency learning proponents
(Schmidt & Besner, 2008; Schmidt et al., 2007) to disregard contingency
learning as a conict-dependent process, by showing that contingency
learning not only does not represent learning-based control, but it also
does not contribute to the control process by its ability to facilitate performance, and thus counteract the effect of conict.
As already mentioned, in contrast to the studies on other implicit
learning processes that showed a larger learning effect under conict

RTs slower than 2000 ms were not allowed by the experiment program.
Note, the insignicance of this effect is attributed to a smaller sensitivity of the
between-participant design.
7

Y. Levin, J. Tzelgov / Acta Psychologica 164 (2016) 3945

Fig. 1. Response latencies for contingency and stimulus category. Vertical bars represent
standard error of the mean.

conditions, in our study a larger contingency learning effect was observed under a conict-free condition. One possible reason for that inconsistency might be that in contrast to other implicit learning
processes that are reported as not being related to such high-order cognitions such as working memory (Siegelman & Frost, 2015) and attention resources (Jimnez & Mndez, 1999; Turk-Browne, Junge, &
Scholl, 2005), contingency learning might depend on them. It has
been shown, for example, that working memory load weakens the contingency learning effect (Schmidt, De Houwer, & Besner, 2010). Loading
working memory, however, means reducing the amount of attention
resources devoted to the task/process of interest. Therefore, contingency learning might be sensitive to the amount of available attention
resources. This is consistent with the smaller contingency learning effect found in our study for word stimuli. Words automatically trigger
another taskreadingwhereas shapes do not. Performing an additional task consumes attentional resources that otherwise would be devoted to the learning of contingencies. Such sensitivity of the
contingency learning with respect to available attention resources,
which of course should further be investigated in future studies,
seems to be consistent with demonstrated immunity of the contingency learning to conict experience. That is, since the cognitive system
ideally maintains good performance at minimal cost (for a discussion
see Goldfarb & Henik, 2014), it might be protable for the cognitive system to enhance only those implicit learning processes that do not require much cognitive resources. If contingency learning indeed
consumes attention resources then enhancing it in conict situations
may provide no advantage or may even cause more harm to performance, which explains why the contingency learning effect was not enlarged under a conict condition in the present study.
Another issue that has already been partially acknowledged in the
introduction section is the apparent similarities existing between contingency learning and what is known as implicit learning. The latter
includes various learning processes such as probabilistic sequence
learning (Nissen & Bullemer, 1987), articial grammar learning
(Reber, 1967) and statistical learning (Saffran, Aslin, & Newport, 1996)
that refer to the basic ability of the cognitive system to pick up the regularities existing in a continuous environment. Implicit learning is traditionally described as an associative, unintentional process that
proceeds automatically, as by product of mere exposure (Saffran,
Johnson, Aslin, & Newport, 1999) for the purpose of generating expectations (Siegelman & Frost, 2015; for review see Cleeremans,
Destrebecqz, & Boyer, 1998). Thus, for example, in a serial-response

43

task (Nissen & Bullemer, 1987) a target (e.g., an asterisk) appears in a


serial fashion in one of four (or more) possible locations on the screen.
Participants are asked to indicate the location where the target appeared by pushing a key representing that location. Unbeknownst to
the participants, the target follows a predictive (deterministic or probabilistic) sequence. This sequence is unintentionally learned by the participants, as indicated by the gradual improvement of performance
with practice.8 However, when the sequence is unexpectedly changed
to a random sequence, the last two or three blocks of the response RTs
are elevated. The RT elevation demonstrates, therefore, the effect of mismatch between the expected location of the target, based on the learned
sequence, and the real location. The former effect is usually considered
to be the effect of learning acquisition, whereas the effect of changing
to a random sequence demonstrates the retrieval/expression of the acquired (sequence) knowledge. As one can note, learning the sequence
and using the acquired knowledge is highly protable in that it maximizes performance. Using the terminology proposed by Perlman and
Tzelgov (2006), most of the examples of implicit learning represent incidental learning. Incidental learning is learning that is acquired unintentionally, when no explicit instructions are given, but it is benecial
to the task intentionally performed. The usefulness of incidental learning to task performance is the main feature allowing differentiating it
from automatic learning, which is also acquired unintentionally but
does not offer any performance advantage. In fact, as the results of Perlman and Tzelgov show, automatic learning acquisition and expression
may occur even when they interfere with ongoing performance (see
also Poznanski & Tzelgov, 2010 for articial grammar learning
paradigm).
Within this terminology, it is clear that contingency learning in the
Stroop task represents another exemplar of incidental implicit learning.
There are obviously no instructions given to participants to attend to
regularities in the co-appearance of specic words and colors in the
Stroop task, and yet those are noticed and very quickly learned. According to Schmidt et al. (2010), it only takes 18 trials to acquire such knowledge. The acquired contingency knowledge is then used to predict
responses, and therefore to optimize performance, which is consistent
with the denition of incidental learning.
However, despite the aforementioned theoretical-level similarities, there are fundamental differences between the Stroop contingency
learning and implicit learning paradigms. The most obvious difference is
the nature of a developed association. In the implicit learning literature,
it has been shown that even when the possibility of creation of a pure
stimulusresponse association is eliminated by mapping multiple responses to the same response key, the stimulusstimulus association
can underlie the learning effect (e.g., Deroost et al., 2012). Whether
the same is true for the Stroop contingency learning paradigm is still
to be shown in future studies. Schmidt et al. (2007, Experiment 4)
showed that contingency learning proceeds exclusively through association between a word and a key response, but not between a word and a
color-naming response. According to these ndings, no contingency
learning should be evident with vocal responses, for example, where
the association can only be made with a color-naming response. If no
contingency learning is observed with vocal responses, such a nding
would make an interesting departure from the implicit learning
view of the contingency learning process, according to which any regularities in the environment can be automatically picked up by the cognitive system. Of course, such an investigation requires a design that
would allow disentangling the process of learning acquisition and the
process of the retrieval/expression of what was learned.

8
Another variation of this paradigm employs a between-subject design in which the effect of sequence learning is estimated by comparing the performance of the group that is
exposed to the sequenced target appearance with the performance of the group exposed
to a random target appearance.

44

Y. Levin, J. Tzelgov / Acta Psychologica 164 (2016) 3945

Another difference between the contingency learning paradigm and


sequence learning paradigm relates to the nature of the generated predictions in temporal terms. Hence, what underlies the sequence learning effect, for example, is the development of an association between
the identity of the target in the current trial and its identity in the subsequent one. Hence, the predictions made based on the learned sequence are relevant to the next trial. This is not the case, however, in
the color-word contingency learning as discussed here, where the
learned stimulusresponse association is relevant to the response
made in the current trial. Importantly, the latter is allowed by the fact
that in this paradigm, the employed stimuli are two-dimensional and
possess a task-irrelevant (word) dimension, which by itself constitutes
another aspect of dissimilarity. In contrast to the sequence learning paradigm, it is the irrelevant (distracting) dimension of the stimulus that
becomes associated with a specic response in the Stroop contingency
learning paradigm.
It is still to be understood whether these dissimilarities might or
might not be signicant when relating and possibly equating contingency learning with implicit learning phenomena. For now, it seems
that it cannot be expected that ndings from one paradigm would be
necessarily replicated in the other. In this context, the present study
can be viewed as providing new information that adds to the similarities
between the contingency learning and other implicit learning processes. As our data show, contingency learning does not require special
conditions, such as conict experience or a presence of an additional
task it can develop with unreadable, conict-free stimuli, such as geometric shapes. This nding is consistent with the denition of implicit
learning according to which it is a basic human ability to pick up the regularities from the environment through mere exposure to them. In this
sense, contingency learning is not different from, for example, statistical
learning found in infants who manage to learn what sequences in continuous speech represent words. There is no conict experience that can
push the newborns to engage in such a process; obviously no instructions or other extrinsic motivators are given, and yet the learning occurs. According to the present results, contingency learning shares this
feature of acquisition by mere exposure as well. However, as we mentioned previously, other possible moderators such as response modality
should be investigated as well to support the automaticity of colorword contingency learning.
4. Conclusions
Schmidt and Besner (2008) and Schmidt et al. (2007) explicitly
claimed that the mechanism underlying contingency learning does
not depend on the presence of Stroop conict, meaning that contingency learning is not a mechanism recruited to help control, but is an independent process that functions in parallel to it (Schmidt, 2013a,
2013b). However, until now, it has only been demonstrated that contingency learning is independent of response conict, meaning that it does
not represent learning-based control (Verguts & Notebaert, 2008,
2009). In the present study, we investigated whether the constant presence of a task conict in the reported experiments (i.e., Schmidt &
Besner, and Schmidt et al.) could enhance or even trigger contingency
learning. Such ndings would mean that contingency learning does depend on conict experience but not in the sense proposed by the
conict-monitoring theory (e.g., a learning-based control). According
to Deroost et al. (2012), instead of regulating the magnitude of the experienced conict, some implicit learning mechanisms might be recruited by the control system to counteract detrimental effects of
conict on performance (i.e., an assisting form of controllearning relationship). The results showed that contingency learning was neither
triggered nor boosted by the presence of the task conict. The present
results imply that contingency learning is completely independent of
any conict experience, and therefore does not represent learningbased control nor the assisting form of the controllearning
relationship.

References
Aarts, E., Roelofs, A., & van Turennout, M. (2009). Attentional control of task and response
in lateral and medial frontal cortex: Brain activity and reaction time distributions.
Neuropsychologia, 47, 20892099.
Abrahamse, E. L., Duthoo, W., Notebaert, W., & Risko, E. F. (2013). Attention modulation
by proportion congruency: The asymmetrical list shifting effect. Journal of
Experimental Psychology. Learning, Memory, and Cognition, 39, 15521562.
Atalay, N. B., & Misirlisoy, M. (2012). Can contingency learning alone account for itemspecic control? Evidence from within- and between-language ISPC effects. Journal
of Experimental Psychology. Learning, Memory, and Cognition, 38, 15781590.
Bench, C. J., Frith, C., Grasby, P., Friston, K., Paulesu, E., Frackowiak, R., & Dolan, R. (1993).
Investigations of the functional anatomy of attention using the Stroop test.
Neuropsychologia, 31, 907922.
Blais, C., Robidoux, S., Risko, E. F., & Besner, D. (2007). Item-specic adaptation and the
conict-monitoring hypothesis: A computational model. Psychological Review, 114,
10761086.
Botvinick, M. M., Braver, T. D., Barch, D. M., Carter, C. S., & Cohen, J. D. (2001). Conict
monitoring and cognitive control. Psychological Review, 108, 624652.
Botvinick, M. M., Cohen, J. D., & Carter, C. S. (2004). Conict monitoring and anterior cingulate cortex: An update. Trends in Cognitive Sciences, 8, 539546.
Brown, T. L. (2011). The relationship between Stroop interference and facilitation effects:
Statistical artifacts, baselines and a reassessment. Journal of Experimental Psychology.
Human Perception and Performance, 37, 8599.
Bugg, J. M. (2014). Conict-triggered top-down control: Default mode, last resort, or no
such thing? Journal of Experimental Psychology. Learning, Memory, and Cognition, 40,
567587.
Bugg, J. M., & Hutchison, K. A. (2013). Converging evidence for control of color-word
Stroop interference at the item level. Journal of Experimental Psychology. Human
Perception and Performance, 39, 433449.
Bugg, J. M., Jacoby, L. L., & Toth, J. P. (2008). Multiple levels of control in the Stroop task.
Memory and Cognition, 36, 14841494.
Bugg, J. M., Jacoby, L. L., & Chanani, S. (2011). Why it is too early to lose control in accounts
of item-specic proportion congruency effects. Journal of Experimental Psychology.
Human Perception and Performance, 37, 844859.
Cleeremans, A., Destrebecqz, A., & Boyer, M. (1998). Implicit learning: News from the
front. Trends in Cognitive Sciences, 2, 406416.
De Houwer, J. (2003). On the role of stimulusresponse and stimulusstimulus compatibility in the Stroop effect. Memory and Cognition, 31, 353359.
De Pisapia, N., & Braver, T. S. (2006). A model of dual control mechanisms through anterior cingulate and prefrontal cortex interactions. Neurocomputing, 69, 13221326.
Deroost, N., & Soetens, E. (2006). The role of response selection in sequence learning. The
Quarterly Journal of Experimental Psychology, 59, 449456.
Deroost, N., Vandenbossche, J., Zeischka, P., Coomans, D., & Soetens, E. (2012). Cognitive
control: A role for implicit learning? Journal of Experimental Psychology. Learning,
Memory, and Cognition, 38, 12431258.
Entel, O., Tzelgov, J., Bereby-Meyer, Y., & Shahar, N. (2014). Exploring relations between
task conict and informational conict in the Stroop task. Psychological Research (Epub ahead of print).
Goldfarb, L., & Henik, A. (2007). Evidence for task conict in the Stroop effect. Journal of
Experimental Psychology. Human Perception and Performance, 33, 11701176.
Goldfarb, L., & Henik, A. (2014). Is the brain a resource-cheapskate? Frontiers in Human
Neuroscience, 8, 857. http://dx.doi.org/10.3389/fnhum.2014.00857.
Hazeltine, E., & Mordkoff, J. T. (2014). Resolved but not forgotten: Stroop effect dredges up the
past. Frontiers in Psychology, 5, 1327. http://dx.doi.org/10.3389/fpsyg.2014.01327.
Hutchison, K. A. (2011). The interactive effects of listwide control, item-based control, and
working memory capacity on Stroop performance. Journal of Experimental Psychology.
Learning, Memory, and Cognition, 37, 851860.
Jacoby, L. L., McElree, B., & Trainham, T. N. (1999). Automatic inuences as accessibility
bias in memory and Stroop tasks: Toward a formal model. In D. Gopher, & A. Koriat
(Eds.), Attention and performance XVII: Cognitive regulation of performance: Interaction
of theory and application (pp. 461486). Cambridge, MA: MIT Press.
Jacoby, L. L., Lindsay, D. S., & Hessels, S. (2003). Item-specic control of automatic processes: Stroop process dissociation. Psychonomic Bulletin & Review, 10, 638644.
Jimnez, L., & Mndez, C. (1999). Which attention is needed for implicit sequence learning? Journal of Experimental Psychology. Learning, Memory, and Cognition, 25, 236259.
Kalanthroff, E., Goldfarb, L., & Henik, A. (2013). Evidence for interaction between the stop
signal and the Stroop task conict. Journal of Experimental Psychology. Human
Perception and Performance, 39, 579592.
Klein, G. S. (1964). Semantic power measurement through the interference of words with
color-naming. The American Journal of Psychology, 77, 576588.
Koch, I. (2007). Anticipatory response control in motor sequence learning: Evidence from
stimulusresponse compatibility. Human Movement Science, 26, 257274.
Levin, Y., & Tzelgov, J. (2014). Conict components of the Stroop effect and their control.
Frontiers in Psychology, 5, 463. http://dx.doi.org/10.3389/fpsyg.2014.00463.
Levin, Y., & Tzelgov, J. (2015). What Klein's semantic gradient does and does not really
show: Decomposing Stroop interference into task and informational conict components.
(Manuscript in preparation).
Logan, G. (1985). Skill and automaticity: Relations, implications and future directions.
Canadian Journal of Psychology, 39, 367386.
Logan, G., & Zbrodoff, N. J. (1979). When it helps to be misled: Facilitative effects of increasing the frequency of conicting stimuli in a Stroop-like task. Memory and
Cognition, 7, 166174.
MacLeod, C. M., & MacDonald, P. A. (2000). Interdimensional interference in the Stroop
effect: Uncovering the cognitive and neural anatomy of attention. Trends in
Cognitive Sciences, 4, 383391.

Y. Levin, J. Tzelgov / Acta Psychologica 164 (2016) 3945


Matht, S., Schreij, D., & Theeuwes, J. (2012). OpenSesame: An open-source, graphical experiment builder for the social sciences. Behavior Research Methods, 44, 314324.
Melara, R. D., & Algom, D. (2003). Driven by information: A tectonic theory of Stroop effects. Psychological Review, 110, 422471.
Nissen, M. L., & Bullemer, P. (1987). Attentional requirements of learning: Evidence from
performance measures. Cognitive Psychology, 19, 132.
Perlman, A., & Tzelgov, J. (2006). Interactions between encoding and retrieval in the domain of sequence-learning. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 32, 118130.
Poznanski, Y., & Tzelgov, J. (2010). Modes of knowledge acquisition and retrieval in articial grammar learning. The Quarterly Journal of Experimental Psychology, 63,
14951515.
Reber, A. S. (1967). Implicit learning of articial grammars. Journal of Verbal Learning and
Verbal Behavior, 6, 855863.
Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274, 19261928.
Saffran, J. R., Johnson, E. K., Aslin, R. N., & Newport, E. L. (1999). Statistical learning of tone
sequences by human infants and adults. Cognition, 70, 2752.
Schmidt, J. R. (2013a). The Parallel Episodic Processing (PEP) model: Dissociating contingency and conict adaptation in the item-specic proportion congruent paradigm.
Acta Psychologica, 142, 119126.
Schmidt, J. R. (2013b). Questioning conict adaptation: Proportion congruent and Gratton
effects reconsidered. Psychonomic Bulletin & Review, 20, 615630.
Schmidt, J. R., & Besner, D. (2008). The Stroop effect: Why proportion congruent has nothing to do with congruency and everything to do with contingency. Journal of
Experimental Psychology. Learning, Memory, and Cognition, 34, 514523.
Schmidt, J. R., Crump, M. J. C., Cheesman, J., & Besner, D. (2007). Contingency learning
without awareness: Evidence for implicit control. Consciousness and Cognition, 16,
421435.

45

Schmidt, J. R., De Houwer, J., & Besner, D. (2010). Contingency learning and unlearning in
the blink of an eye: A resource dependent process. Consciousness and Cognition, 19,
235250.
Sharma, D., & McKenna, F. (1998). Differential components of the manual and vocal
Stroop tasks. Memory and Cognition, 26, 10331040.
Siegelman, N., & Frost, R. (2015). Statistical learning as an individual ability: Theoretical
perspectives and empirical evidence. Journal of Memory and Language, 81, 105120.
Steinhauser, M., & Hbner, R. (2009). Distinguishing response conict and task conict in
the Stroop task: Evidence from ex-Gaussian distribution analysis. Journal of
Experimental Psychology. Human Perception and Performance, 35, 13981412.
Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of
Experimental Psychology, 18, 643662.
Turk-Browne, N. B., Junge, J. A., & Scholl, B. J. (2005). The automaticity of visual statistical
learning. Journal of Experimental Psychology. General, 134, 552564.
Tzelgov, J. (1997). Specifying the relations between automaticity and consciousness: A
theoretical note. Consciousness and Cognition, 6, 441451.
Vandenbossche, J., Coomans, D., Homble, K., & Deroost, N. (2014). The effect of cognitive
aging on implicit sequence learning and dual tasking. Frontiers in Psychology, 5, 154.
http://dx.doi.org/10.3389/fpsyg.2014.00154.
Verguts, T., & Notebaert, W. (2008). Hebbian learning of cognitive control: Dealing with
specic and nonspecic adaptation. Psychological Review, 115, 518525.
Verguts, T., & Notebaert, W. (2009). Adaptation by binding: A learning account of cognitive control. Trends in Cognitive Sciences, 13, 252257.
Zhao, J., Ngo, N., McKendrick, R., & Turk-Browne, N. B. (2011). Mutual interference between statistical summary perception and statistical learning. Psychological Science,
22, 12121219.

You might also like