Existing so-called conflict-based models of executive control aim to explain how an agent, without constantly controlling its own processing (what is both cognitively costly and inefficient), can know when to apply strong control, but when to withdraw it. These models predict that the strength of control is adjusted proportionally to the level of conflict among competing stimuli/response tendencies. However, so far the conflict-based models were verified with the use of relatively simple experimental paradigms, like the Stroop task. In the present study, we extended the effect of evoked conflict on the strength of executive control, exerted by participants, to a more realistic task (the search of information in a portal-like browser). The results indicate that also semantic conflicts (incompatible meaning of subsequent messages) can mobilize executive control, and help people to cope with experienced distraction and difficulty.
A crucial human mental faculty that is intensively studied in
cognitive science/neuroscience is executive control (also
called cognitive control). It allows humans to direct and
coordinate their thoughts and actions in a flexible and novel
way, in order to reach adopted goals, even in face of
conflicting stimulation and strongly learned but inadequate
response tendencies. The important role of control in human
behavior becomes clearly visible in situations when such a
control has been disrupted (e.g., due to illness, aging, etc.),
and agents are no longer able to inhibit intruding thoughts or
responses, prevent perseveration, overcome salient
distraction, switch between alternative tasks, or plan their actions
Recent research efforts aim to explain how the mind/brain
is able to internally control its own cognitive processes,
without positing any vague and homuncular constructs like
will, person, or self. One important conclusion from this line
of research states that cognitive control most likely is not a
function of one dedicated cognitive subsystem, but it seems
to emerge from the complex interactions between diverse
mechanisms/processes
Dunbar, & McClelland; Gray, 2007), to cognitive
neuroscience
However, highly controlled (goal-focused) processing is
cognitively and energetically costly
One solution to this paradox assumes that an agent just
monitors some simple global signal (simple enough not to
require any complex processing), which acts as a heuristic
for the evaluation of how strong control is needed in a
particular situation. It has been proposed that such a signal
can rely on various measures of conflict (incongruency,
incompability) between thoughts/actions that can be
potentially applied in a given situation
Since very beginnings of psychological research, the role
of conflict in mediating control was studied in natural
settings (henceforth we will call such settings realistic
tasks). For example, Kurt
However, one disadvantage of studying the relationship
between perceived conflict and executive control is the fact
that realistic tasks usually do not provide sufficient
experimental control required to carry out more fine-grained
cognitive and neurofunctional research, and in particular –
to verify precise computational models of control. Thus, this
type of research usually uses simpler laboratory tasks, like
the Stroop task and its variants
The crucial observation in Stroop, called the congruency
effect, consists of increased response latency in incongruent
trials, for example when the color denoted by a word
mismatches the ink color, compared to RT in neutral trials,
for instance when the color of a color-unrelated string, like
‘XXXXX’, has to be named. The effect is even larger if the
incongruent trials are compared to trials in which ink color
and the word meaning match (to congruent trials). Because
of its simplicity (simple stimuli displayed, and only a few
vocal/manual responses required), the Stroop task (and
similar tests) have been widely used to examine the theories
and models
Two experimental effects found in Stroop studies were
especially interpreted as resulting from differences in the
evaluated conflict level. The Gratton effect
Moreover, the congruency effect can be decreased by an
increasing proportion of incongruent trials in the sequence
The aim of the present study is to show that the above mentioned relationship between conflict and control can also be found in a more complex and ecologically valid test of executive control, that is, in a realistic task. At the same time, this task will still be computer-administered, thus potentially allowing for precise manipulations of task parameters (e.g., proportions of certain stimuli, presentation times, the nature of evoked conflict, feedback, etc.).
First, if conflicts evoked within such a task affect the indices of executive control performance, this fact will imply that the predictions of conflict-based models of executive control observed so far can be generalized onto more complex and higher-level processes, supporting the psychological plausibility of these models. Also, the novel knowledge about operation of executive control in realistic tasks will allow us to design such tasks in a better way (e.g., in a way in which they impose less load on executive control or working memory), so it will have important practical implications.
Second, the conflicts evoked in our task will consist of the
semantic incongruency between presented stimuli, whereas
these incongruent stimuli will not yield incompatible motor
responses (they will just lead to cognitive dissonance). As
so far most of conflict-based models of control accounted
only for conflicts at the stimulus
Especially, our own computational model
One challenge for a study of executive control in natural settings is the design of a realistic task that, on one hand, can be applied using a computer, and requires relatively simple reactions (e.g., with a mouse), whereas on the other hand it is still ‘realistic’, in the sense that it resembles activities that most of people do for a certain part of a day at their work or at home. Our choice was a tool that requires both searching and reading the short portions of information (both textual and graphical) within a simplified internet portal, in order to fulfill a task of gathering as much relevant knowledge on a given (realistic) problem as possible, and eventually answering one precise question regarding that problem. The crucial manipulation in such a task consists of introducing a certain amount of semantic incompatibility between target passages of text (some passages negate others). We expected increased semantic incompatibility (i.e., the conflict on a conceptual level) to increase the strength of exerted control, which in effect would help to deal with a higher distraction – a factor that likely would affect negatively the goal-relevant performance, if not prevented by strong executive control.
A total of 46 women and 36 men participated (82 people). All of them were recruited via adds on social networking webpages. Mean age was 22.8 years (SD = 3.38, range 18 – 38). For a two-hour session each participant received the equivalent of ten euro in local currency. All participants had normal or corrected-to-normal vision.
The screen in the task was composed of 3 × 3 matrix of locations. The task consisted of four problems. In a particular problem, the initial screen consisted of nine messages. Each message was placed in one of the matrix cells. In subsequent cycles, every 5 s, a random message was substituted with another message. In total, 100 messages were presented in one problem (including the initial messages).
Messages could belong to one of four categories. Regular target messages were short passages of text (not shorter than 120 characters) providing an information relevant for a problem to be solved (see below). For instance, a regular target message A could state that ‘company X expects more sales next year and prepares for that fact’. However, a certain number of target messages (conflicting target messages) negated regular target messages that directly preceded them (e.g., message B: ‘X expects less sales next year and will cut costs’, directly following – that is, not separated by any other target message – message A).
Another category were distractor messages, which were text messages (30%) which conveyed information superficially associated with the problem, but in fact irrelevant for it (e.g., ‘sales employees of X won soccer cup in the 2013 sales departments competition’), attractive graphics (30%; either funny cartoons or erotic images of young pretty women/handsome men), or text jokes (40%). Distractor messages were intended to capture attention of participants, what might result in missing target messages, as the latter disappeared from the screen after certain time (depending on the number of cycles that it was displayed for).
The last category were noise messages, which conveyed either text information irrelevant for the problem, but in no way conflicting or distracting (e.g., ‘several national parks have been founded in Poland in recent years’), or images of supposedly not distracting objects and landscapes. The use of both the distractor and noise messages made the contents of the task relatively similar to internet portals, which usually contain a lot of irrelevant textual and graphical information. The example screen of the task, including all types of messages, is presented in Fig. 1.
The task of each participant was to monitor and read messages that can be potentially informative with regard to the problem presented to her or him in an instruction. Participants were also instructed that they have to confirm with the computer mouse the fact that a certain message is a message conveying an important knowledge on the problem (by clicking on that message). At the beginning of the experiment, the participants were informed that after the computerized part of the test they would be provided with messages they chose, and they would have to answer a question about presented problems. Answering the question consisted of providing the subjective probability of the confirmative answer to this question.
The number of conflicting target messages and the number of distractor messages were two crucial task parameters. In the no-conflict condition, there were 30 regular target messages defined for a particular problem, but no conflicting target messages were presented. In the conflict condition, half of 20 regular target messages (randomly picked up from the pool of 30 messages) was followed by the corresponding conflicting target message (so, there were also 30 targets in total, but some their pairs were mutually incompatible semantically). In the lowdistraction condition, there were 10 distractor messages/ images in a run, whereas in the high-distraction condition as much as 60 such messages/images were presented. In order to obtain the 100-message/image sequences, in the former condition 60 noise messages/images were used, whereas in the latter – 10 such messages/images were included. For each participant and problem, the distractor and noise messages/images were picked up on random from a pool of 1186 distractor and 1500 noise messages/images.
The problems were formulated as follows: ‘On a basis of information provided in a task, please …’: • analyze new investment of IT company X in a mobile phone system, and judge the probability that X will increase its headcount due to this investment; • describe how the human cortex works; • tell how computer processor works; • evaluate what factors have the most important role in supporting the existing political system in Ukraine. Noteworthy, in order to be maximally interesting for participants, the problems pertained to diverse topics.
Thus, in the present experiment, the independent variables were: semantic conflict (either present or absent), and distraction (either low or high). For each participant, all possible problems and conditions were combined on random, resulting in the 2 × 2 ‘within-subjects’ design. First, we expected that distraction would significantly decrease performance accuracy (i.e., people will be looking at erotic pictures or jokes instead of selecting the target messages). Second, we expected that the magnitude of the distraction effect (i.e., accuracy on low distraction minus high distraction condition) in the no-conflict condition would be attenuated by increased control in the conflictcondition (i.e., people, after detecting conflicts, would focus more on the task, and would better ignore distractors). Thus, we expected the two-way interaction analogous to the Gratton effect in Stroop.
Participants were tested in a large, dimly lit room, in groups of up to ten people. Standard PC workstations with 17’ LCD monitors were used for the test. Each participant occupied a visually isolated desk, and she or he was asked to adopt the most comfortable sitting position.
The primary dependent variable (DV) was the proportion
of missed regular target messages (i.e., error rate) in each
problem, corrected (i.e., increased) by the weighted
proportion (with the weight reflecting the ratio of targets to
non-targets; i.e., noise and distractor messages) of
incorrectly identified non-targets
.41 (.20) .36 (.15) cCoonnfflliicctticnogndtaitrigoents in .46 (.24) .45 (.22) Table: Mean error rate (and SD) in all conditions of the task. Data regarding regular target messages were submitted to ANOVA. In the case of errors, two factors yielded significant main effects. First, in the high distraction condition participants missed target messages more often (M = .39) than in the low distraction condition (M = .28), F(1, 81) = 55.59, p < .001, η2 = .41. This fact implied that superficially similar texts, funny cartoons, and erotic images, originally aimed to capture people’s attention, indeed diverted participants from fulfilling the task, and constituted the substantial source of interference for the executive system to cope with. Second, there was no significant difference in errors between the conflict and noconflict conditions, F(1, 81) = 0.12, meaning that conflict did not affect the accuracy of recognition of regular target messages per se.
As expected, in the conflict condition participants missed the conflicting target messages more often (M = .46) than they missed the regular target messages (M = .31), F(1, 81) = 58.80, p < .001, η2 = .42. This effect indicates that they indeed detected semantic incompatibility between consecutive target messages, and often decided that a incompatible message was irrelevant for the solution of the current problem (so they did not click on it).
In light of our hypotheses, the most important effects pertained to the two-way interactive effect of factors, which was significant F(1, 81) = 15.24, p < .001, η2 = .94. Tukey’s HSD test showed that high and low distraction conditions differed both in no-conflict (p < .001) and conflict (p = .008) conditions as well as conflict and no-conflict conditions differed both in low distraction (p = .009) and high distraction (p = .009) conditions. The interaction is presented in Figure 2.
Our hypotheses assumed that increased perception of conflict, evoked by placing the semantically incompatible messages within the stream of information presented to participants, would affect the effects possibly yielded by factors that load executive control mechanisms, which might be responsible for dealing with our realistic task. We obtained strong evidence in favor of this hypothesis. Increased distraction made people to respond less correctly, but this effect was attenuated by increased conflict. In line with the conflict-based models of control, we attribute this interaction to additional strength of control, which was ‘mobilized’ and, thus, control became more effective when participants started facing semantic conflicts.
Thus, these results are pretty analogous to the Gratton effect observed in various tests of executive control. However, as far as we know, now for the first time they have been observed within a much more complex task than such tests, that is, within a task that in a way resembles natural situations of information acquisition and selection.
The theoretical implications of the present work pertain to
the verification of the above mentioned conflict-based
models of executive control. Extending their predictions to a
(more or less) realistic task suggests that perceived conflict
(in stimulation or between mental representations/response
tendencies) may indeed be a type of signal that is evaluated
for the sake of optimizing the strength of exerted control.
Thus, our study seems to extend and generalize predictions
of the conflict-based models of control. Especially, the
results in some way support a key assumption of our own
conflict-based model, which predicts that not only
response
However, it must be noted that conflicts may not be the
only type of signal that can regulate executive control. Other
accounts, for instance models that in regulating control rely
on the learned (via reinforcement learning) likelihood of
negative outcomes like errors or risky actions
In conclusion, the present study in an original way combined the precise manipulation of factors possibly affecting the workings of human executive control mechanisms with the relatively complex, higher-level realistic task. Future steps in the present line of research should extend the examination of variables possibly influencing executive control, which are based on evoked conflict, to even more realistic settings. In this regard, the development of virtual reality platforms constitutes a very promising research tool that should be further exploited. Knowledge on factors negatively (or positively) affecting the internal control of human cognitive processing in natural settings may also help to design better human-computer interfaces, vehicle cockpits, etc.
This work was sponsored by the National Science Centre of Poland (grant no. 2011/01/D/HS6/00467). We are grateful to Jolanta Wójcik for designing the problems and conducting the experiment, and to Jaroslav Sadowsky for programming the experimental task.