Introduction
Cognitive control constitutes an integral part of life in a society where one not only needs to follow rules and suppress impulsive behaviors in public environments, but also to ignore distractions and noise in one’s own private environment. “Noise” seems to gain more and more ground in the natural as well as the digital context, making numerous daily tasks such as navigating through a website while bombarded by visual and auditory advertisement an arduous journey of cognitive demands. To adapt and perform in a wide variety of contexts, cognitive control is required and recruited every step of the way. To study the capacity for cognitive control in the lab, real life situations that require selective attention towards a target while suppressing distractions are simulated through conflict tasks, also sometimes referred to as distracter interference tasks or stimulus response compatibility tasks (Eriksen & Eriksen,
1974; Simon & Rudell,
1967; Stroop,
1935). A key question in this line of research is to what extent and under which conditions cognitive control can be exerted effectively under changing contexts. In the current study, we examine the flexibility of cognitive control by studying how conflict adaptation effects transfer within and across different conflict tasks of varying similarity.
Conflict tasks aim to measure the level of engagement of cognitive control through the presentation of relevant information in combination with irrelevant information and require the participant to respond only to the target information. The relevant and irrelevant dimensions can activate the same or different response tendencies, giving rise to the so-called congruent and incongruent conditions of which the latter is thought to trigger the experience of conflict that is usually accompanied by more errors and longer reaction times compared to congruent conditions; a performance difference termed the congruency effect (Egner,
2007; Kornblum et al.,
1990). Examples of classic conflict tasks are the Stroop (
1935), Flanker (Eriksen & Eriksen,
1974) and Simon (Simon & Rudell,
1967) tasks. While these are often grouped together as conflict tasks, there is now widespread appreciation that they are distinct in many ways. For instance, the conflict arising in Simon tasks is traditionally associated with stimulus-response location, whereas in Stroop tasks it is associated with stimulus-stimulus dimensional overlap (Kornblum et al.,
1999). Further, across the different tasks automatic and controlled processes are thought to unfold in different temporal scales (Ulrich et al.,
2015), and the impact of relevant and irrelevant information appears to be distinct (Mackenzie et al.,
2022). In the current study, this diversity in conflict tasks was exploited to create more and less similar task pairings and investigate the flexibility of cognitive control processes.
In addition to participants’ performance being affected by the current trial’s congruency, studies employing conflict tasks have reported sequential modulations of the size of the congruency effect depending on the previous trial’s congruency (e.g. Gratton et al.,
1992). This more dynamic view of cognitive control has been termed conflict adaptation. More specifically, a reduced congruency effect following an incongruent trial, as compared to following a congruent trial, is a phenomenon now commonly referred to as the congruency sequence effect or Gratton effect (Braem et al.,
2014; Egner,
2007). The congruency sequence effect (from here on referred to as the CSE) has been studied as a marker for conflict adaptation and was originally thought to arise because people expect the same type of trial to repeat and prepare for this (Gratton et al.,
1992). Two types of CSE have been identified: the so-called within-task CSE (when the effect is observed across trials of the same conflict task) and the across-task CSE (when the effect is observed across trials of different conflict tasks). The latter is thought to reflect cognitive control transcending task context and as such reflects a surprising degree of flexibility of cognitive control.
A variety of theories and models have been suggested with respect to the source, underlying mechanism and specificity of conflict adaptation. The most prominent theories can be summarized in two groups according to their core concepts (Weissman et al.,
2014): the ‘trial by trial modulation of cognitive control’ account and the ‘learning and memory account’. The former assumes the adaptive top-down allocation of attentional resources and examples of this category include the original proposition by Gratton et al. (
1992) as well as the well-known conflict-monitoring theory (Botvinick et al.,
2001). The learning and memory account assumes the effect arises because associations of different task features are learned and stored into memory and comprises theories such as feature integration (Hommel,
2004) and contingency learning (Schmidt & DeHouwer,
2011). While the different theoretical accounts disagree on the determinant factors and processes involved in the occurrence of a CSE, they offer a rich ground for research into the different mechanisms underlying conflict processing. It should, however, be noted that testing these different theoretical models is not the aim of the current study.
While observations of within-task CSEs have been reliably replicated across studies employing several variations of the classic conflict tasks, the findings for across-task CSEs are inconsistent and there remains debate regarding which shared or distinct features between tasks play a role in the presence of the CSE. In an effort to address this issue, Braem et al., (
2014) suggested a framework in which a U-shaped function describes the relationship between the observation of an across-task CSE and context similarity between tasks. This function reflects that when two tasks can be identified as either highly similar or highly dissimilar with respect to their task sets, they are likely to lead to an across-task CSE. The reasoning behind this hypothesis, given that intuitively one may expect a linear rather than U-shaped relation, originates from whether the two contexts or features interfere with each other when they are both active in working memory. The assumption is that if the contexts differ to such an extent as to not interfere with each other, then both contexts can be maintained simultaneously in working memory. On the contrary, in the case of partial but not complete overlap between two contexts, interference will hinder co-activation of the two contexts, allowing for only one context to be available therefore inhibiting transfer effects between contexts. This framework is also in line with other conflict adaptation models and memory theories namely, theory of event codes (Hommel,
2004), task set level control theory (Hazeltine et al.,
2011) and adaptation by binding theory (Verguts & Notebaert,
2009). The design of the current study was inspired by the U-shaped function proposed by Braem et al. (
2014); we aimed to systematically manipulate the similarity of two tasks that were to be processed in alternation and examine to what extent conflict adaptation across tasks occurred.
Response dynamics and mouse-tracking
In previous years, studies focusing on the mechanisms underlying decision-making assumed that the procedures of perception, programming a response, and executing a response constitute a linear sequence of events. A growing body of research, however, now proposes that these do not unfold sequentially and can influence each other throughout the decision-making process (Erb,
2018; Freeman et al.,
2011; Gallivan et al.,
2018). Studies employing both behavioral and neurophysiological methods have in fact shown that the same underlying system of neurons is responsible for sensory, cognitive and motoric components of the decision-making process and are simultaneously active during the process (e.g., Song & Nakayama,
2009). A considerable amount of literature employing mouse-tracking in a variety of domains including language (e.g., Spivey et al.,
2005), social cognition (e.g., Freeman et al.,
2008) and conflict processing (e.g., İkizoğlu & Çakır,
2021; Yamamoto et al.,
2016; Ye & Damian,
2022) converge in finding that measures derived from the manual dynamics of a response, e.g., the mouse trajectories, are sensitive enough to capture small effects which often escape simple reaction time measures. In their study using Flanker, Simon and Spatial Stroop tasks, Ye and Damian (
2022) found that trajectory measures as well as initiation times could sensitively capture previously experienced conflict. İkizoğlu and Çakır (
2021) observed larger stimulus response compatibility effects in mouse-tracking measures than in response time measures when studying different versions of the Simon task. Employing a reverse Stroop task, Yamamoto et al. (
2016) demonstrated that while interference effects were reflected in the response trajectories, facilitation effects were not. While conventional keyboard response measurements reflect the presence and ultimate resolution of conflict, mouse-tracking parameters can more directly capture the complexity of the cognitive processes underlying conflict; for example, how each task feature may contribute to response selection and at which temporal stage (Freeman,
2018; Hermens,
2018; Kieslich et al.,
2020; Schoemann et al.,
2020).
A variety of parameters have been analyzed in mouse- and other hand-tracking methods that are thought to provide additional insights into decision-making processes as they unfold. Three such movement variables, namely, the Movement Time (MT), Initiation Time (IT) and Maximum Absolute Deviation (MAD), are assessed in this study and will be presented here. The MT reflects the time interval between stimulus presentation and response completion and is therefore analogous to response times. The IT measures the time elapsed between the presentation of the stimuli to the start of the response movement and the MAD measures the degree of deviation of the cursor from the direct path to the final target (Wirth et al.,
2020). Erb et al. (
2016) propose that when measured in conflict tasks the IT and MAD reflect two processes, namely the response threshold adjustment process and the controlled selection process, respectively. The response threshold adjustment process refers to a temporary inhibition of all possible motor outputs in response to conflict. The controlled selection process has been suggested to reflect the ongoing competition between targets, with greater conflict resulting in greater attraction towards the incorrect response, translating into larger movement curvatures or degrees of deviation (Erb & Marcovitch,
2018,
2019; Erb et al.,
2018). Notably, these mouse-tracking measures have primarily been studied within the context of a single conflict task. Here, we extend this approach to evaluate the sensitivity of these measures to detect across-task conflict adaptation effects.
The current study
The aim of the present study is to investigate how flexibly cognitive control can adapt to different contexts. To this end, performance was examined in response to and following conflict in three experiments that varied the context similarity of the tasks employed. In each experiment a different pair of tasks was presented either alternating on a trial-by-trial basis in mixed blocks, or separately in single blocks. The different pairs aimed to establish gradual degrees of similarity with respect to relevant and irrelevant features. By manipulating the similarity between these dimensions, we can observe the extent to which changes in context impact conflict-induced adjustments in performance. Moreover, this study employed the dynamic response method of mouse-tracking, encompassing measures of both time and trajectory, which allowed for fine-grained observation of conflict processing and may inform future studies’ approach in evaluating conflict processes and resolution. The present study originates from a larger project studying the development of cognitive control; here the results of the adult age group are reported.
To assess performance adjustments in response to and following conflict, the presence of the CSE was assessed in the context of single tasks and across different tasks. Based on previous evidence (Hommel,
2004; Kerns et al.,
2004; Notebaert et al.,
2006; Wühr,
2005) we expected to observe within-task CSEs in all three experiments. Following up on Braem et al.’s (
2014) proposed U-shaped function describing the relationship between context similarity and across-task CSEs, we expected to observe across-task CSEs in MT only in the case of very high (Experiment 1) and very low (Experiment 3) context similarity between tasks. Given that the majority of conflict adaptation studies have utilized keypress responses, there was less evidence on which to base our hypotheses of mouse-tracking measures. However, based on Erb and Marcovitch (
2019) we anticipated the IT to be only affected by the current and previous trial’s congruency, but CSEs to also be reflected in the MAD.
General discussion
The goal of this online mouse-tracking study was to investigate the different contexts across which cognitive control can be recruited and exerted. More specifically, we investigated whether conflict adaptation, as measured by the CSE, can transfer within and across conflict tasks that differ in the extent to which the relevant and irrelevant dimensions overlap. Participants responded via computer mouse and the measures accuracy, movement time, initiation time and maximum absolute deviation were collected and analyzed within each and across different combinations of adapted Simon and Stroop tasks. Our hypotheses for the presence of across-task conflict adaptation in each of the three experiments were inspired by a review by Braem et al., (
2014) where it was proposed that across-task CSEs are more likely to be observed when the two tasks’ context similarity can be characterized as either very high (Experiment 1) or very low (Experiment 3) compared to cases of intermediate or partial similarity (Experiment 2). Overall, the results delivered mixed support for our hypotheses with fairly reliable within-task conflict adaptation, but more varied evidence regarding the transfer of conflict adaptation across different tasks (see Table
4 for a summary of the experimental features and results).
Table 4
Summary of experiment features and CSE results from Experiments 1 to 3
1 | Animal Simon | Arrow Simon | Yes | Yes | x MT | ✓ MT | x MT | ✓ MT |
x IT | x IT | x IT | x IT |
✓ MAD | ✓ MAD | ✓ MAD | ✓ MAD |
2 | Size Simon | Arrow Simon | No | Yes | ✓ MT | ✓ MT | x MT | ✓ MT |
x IT | x IT | ✓ IT | x IT |
✓ MAD | ✓ MAD | x MAD | ✓ MAD |
3 | Animal Simon | Animal Stroop | No | No | ✓ MT | ✓ MT | x MT | x MT |
x IT | x IT | x IT | x IT |
✓ MAD | ✓ MAD | ✓ MAD | ✓ MAD |
Let us first consider the across-task data pattern from movement times and maximum absolute deviation values. It was hypothesized that both very high and very low similarity conditions would enable the transfer of conflict adaptation effects; the movement time data were consistent with this in the high similarity condition (Experiment 1) but not in the low similarity condition (Experiment 3). Moreover, and inconsistent with our hypothesis, an across-task CSE in movement times was also reflected in one task order in the medium similarity condition (Experiment 2) where the task features only partially overlapped. Interestingly, the maximum absolute deviation values sometimes provided a different picture. That is, trajectory measures seemed to capture transfer effects across all three experiments and most task orders (with the exception of Size Simon-Arrow Simon in Experiment 2). Taken together, it can be concluded that in all three experiments, conflict experienced in a prior different task influenced response selection processes in the subsequent trial to some extent, and as such our results are not consistent with the hypothesis regarding across-task conflict adaptation.
The time taken to initiate a response movement, reflected in our initiation time measure, was mostly affected by the current and previous trial’s congruency. This is consistent with Erb et al’s (
2016) interpretation of initiation time reflecting a response threshold adjustment process resulting in generalized inhibition of all motor output. While Erb et al. (
2016) studied these measures within the context of a single task at a time, here we have replicated those results within a single task and extended the finding to the context of alternating tasks. Surprisingly, however, initiation time seems to have also captured an across-task CSE in Experiment 2, in the case of Size Simon trials followed by Arrow Simon trials. What is more, in that task order there was no evidence of conflict adaptation in the maximum absolute deviation values. As mentioned previously, one could speculate a trade-off between the processes captured by the two measures. This may call into question the interpretation of initiation time as reflecting a Response Threshold Adjustment process, as it suggests that the time taken to initiate a movement can also be influenced by the particular combination of current and previous trial’s congruency. Even though the majority of across-task conflict adaptation effects observed here were found primarily in trajectory measures (i.e. MAD values) and not time measures (e.g. IT), a fact that could suggest the characterization of CSEs as a measure involved mainly in the controlled selection stage of processing; future research that directly investigates cases in which time but not trajectory measures reflect conflict adaptation effects could shed more light on the source, locus and processes underlying conflict adaptation.
One reason that our results do not fully support Braem et al.’s (
2014) model could be that our experiments did not encompass the entire spectrum of context similarity. More specifically, our classification of the task combinations in Experiments 2 and 3 as medium and low similarity, respectively, could be challenged. Braem and colleagues’ examples of low similarity
1 task combinations include a sentence processing task combined with a color Stroop task and a gender flanker task combined with a letter flanker task. While on the surface these task pairings may appear more dissimilar than those we selected, we do consider them comparable to our Experiment 3. We postulated that the task combination we selected in Experiment 3 is low in similarity, since the two tasks (Animal Simon and Animal Stroop) differ in terms of the source of conflict as well as memory demands. More specifically, in the Animal Stroop task long-term memory regarding the animals’ real-life sizes is required to resolve the conflict, which is in direct contrast with the Animal Simon task where the information causing conflict as well as the information needed to resolve it is immediately available visually in every trial. Furthermore, it is still unresolved whether the conflict adaptation effects found in the studies listed by Braem in this category (e.g., Kan et al.,
2013) may be restricted only to very specific types of language-processing conflicts since other efforts have not replicated those results (e.g., Dudschig,
2022; Simi et al.,
2022).
In composing task combinations in the current study, we classified context similarity based on shared relevant and irrelevant dimensions across the tasks. However, this could be too simplistic and perhaps more attention should be given to some additional concepts such as that of task space, the exact source of conflict and how conflict is resolved in each task. According to Xiong and Proctor (
2018), task space, task set and relations between stimulus and response (S-R relations) are distinct terms with task set associated with task-relevant information, S-R relations associated with mappings between stimulus and response and task space transcending both task set and S-R relations and including task irrelevant information. Especially in complex situations, task space seems like a promising multidimensional candidate to facilitate the definition of task context borders. With respect to sources of conflict, in his review Egner (
2008) notes that sources of conflict can be shared across distinct tasks and that even in the case of distinct sources of conflict, conflict types may still converge at later stages of processing, such as the behavioral output. Interestingly, our mouse-tracking results could be considered consistent with this idea, as sequence effects appear sometimes earlier and sometimes later in the processing stream. All in all, it is clear that the concept of context similarity is still to be precisely defined and should be further addressed in future studies. For now, we can conclude that the occurrence of across-task conflict adaptation cannot be attributed to a simplistic distinction based on shared relevant and irrelevant dimensions as we adopted here.
Another possible reason that our results do not fully support Braem and colleagues’ model could be related to response methods; in the studies reviewed by Braem et al. button responses were employed, whereas we employed mouse-tracking. It should be noted, that even though reaction times as collected through button presses cannot fully map onto movement times as measured through mouse-tracking, Braem et al.’s proposed model had not restricted its application to a specific response method. Our findings, however, can also trigger the question of whether the response mode employed in conflict tasks could directly affect conflict processing and response strategy. It is commonly observed that with hand-tracking response methods, error rates are greatly reduced since there is the opportunity to correct a response (as reflected in the maximum absolute deviation values) before its completion, resulting in a higher number of trials available for data analysis compared to button-response methods. As such, arguably more of the trials on which the most conflict is experienced remain in the dataset when responses are given via hand-tracking methods compared to when responses are given via button presses. To directly investigate whether there are differences in conflict processing depending on the response method it would be interesting to compare different response methods within the context of a single task or combination of tasks. Such studies could help us understand the impact of these design decisions on conflict processing.
Another design decision that may influence the transfer of conflict adaptation effects across tasks is the sequence in which tasks are presented, namely in a predictable or unpredictable manner. When reviewing conflict adaptation studies that employed either alternating or randomly intermixed designs, no notable pattern could be discerned with respect to the presence of transfer effects. It has, however, been suggested that task predictability may lead to more salient task boundaries and therefore less transfer of conflict adaptation effects (Hazeltine et al.,
2011). In addition to context similarity (Braem et al.,
2014), task predictability may be a further important influence on across-task conflict adaptation that is certainly deserving of further investigation.
An unanticipated and to our knowledge novel finding in this study, was the potential effect of task order on across-task conflict adaptation, which adds to the complexity of conflict processing. This complexity could entail now not only the type or source of conflict of the current task and the congruency of the previous task but also the type or source of conflict of the previous task. Interestingly, additional tests revealed that this trend could not have been driven by differences in the size of the congruency effect. For example, conflict adaptation effects across tasks were observed consistently in both movement times and maximum absolute deviation values when Arrow Simon trials were followed by either Animal Simon (Experiment 1) or Size Simon trials (Experiment 2). However, only in Experiment 2 was there a larger congruency effect in the Arrow Simon task than the other task, as calculated from movement times in single blocks (Experiment 1: Arrow Simon
ηp2 = 0.48, Animal Simon
ηp2 = 0.66; Experiment 2: Arrow Simon
ηp2 = 0.74, Size Simon
ηp2 = 0.56). Perhaps the task order effect on across-task conflict adaptation is attributable to the different time courses of automatic versus controlled processes in each task impacting the degree to which effects transfer to the other task. To develop a full picture, additional studies applying modelling approaches such as the Diffusion Model for Conflict tasks (Ulrich et al.,
2015) could address some of these interpretations or provide further insights into other mechanisms involved.
Attention should additionally be drawn to the fact that task performance and hence the presence or absence of behavioral effects, depend on which task representations are formed, assuming multiple representations can correspond to one task (Kleinsorge,
2021). It has even been demonstrated that this can depend on how the tasks are presented to the participants, through for example instructions (Kleinsorge,
1999). In the current study, we kept the type of instruction constant across all three experiments (tasks were introduced separately and participants were informed they would sometimes be presented within the same block). As such, this cannot be an explanation for our varied findings. Nevertheless, the representations every participant forms about each task separately and in combination with another task cannot always be manipulated with precision by the experimenter, leaving the question of whether two tasks are perceived as separate entities or as one single task unresolved in most circumstances. Clarifying the definition of context similarity and task representation seems to be an important next step for this field of research. There are different possible approaches to tackle this question, such as reviewing the already large literature on CSEs or designing new experiments to manipulate different candidate features of context similarity and identify crucial ones. The empirical experimental approach taken here is just one attempt.
It is of note that the novel tasks developed and employed in this online study elicited all the hallmarks of conflict tasks (CE, within-task CSEs). With respect to confound minimization even though there was no direct repetition of stimuli, since this study employed mouse-tracking and aimed at studying conflict processing children as young as 6 years old, other confounds such as response repetition and lateralization were not controlled for. We are not aware of mouse-tracking studies with conflict tasks that also circumvent the problem of response repetition and lateralization. Indeed, given the nature of the core mouse-tracking measures (such as trajectory) that are of interest in these studies, the options for more complex designs with multiple stimulus–response mappings are limited. It is, however, an important limitation to acknowledge due to concerns about the interpretation of CSEs in designs that are not confound-minimized. Nevertheless, there seems to be consensus in the field that while confounds may enhance CSEs, these effects are still observed in fully confound minimized designs (e.g. Koob et al.,
2023). Moving forward, an important aim for future research is to develop confound minimized designs that can be implemented for mouse-tracking and child studies without compromising the translational value of the findings.
In summary, conflict adaptation effects within each of our novel tasks were consistently observed across all experiments, while the transfer of these effects across the different task combinations varied depending on measurement. That is, time measures provided only partial support for our hypothesis and trajectory measures reflected transfer effects across all experiments independent of task combination. This heterogeneity of findings across different measures that capture various aspects of the unfolding decision-making process underlines the importance of applying more sensitive measurement tools to evaluate conflict adaptation. Furthermore, while it is true that variety in findings with respect to across-task transfers is not new to this line of literature, the systematic variation of across-task similarity within one study and the addition of mouse-tracking measurements is. In order to disentangle the meaning and underlying mechanisms of these effects, the conditions under which such transfers occur have to be identified and these conditions need to be addressed more systematically within and across different studies. The current evidence shows that previous conflict affects subsequent conflict processing in all three task combinations, which could reflect a remarkable degree of flexibility of cognitive control in adapting to highly variable and changing environments.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.