Conflict and error adaptation in the Simon task

Abstract

We present recent empirical and theoretical advances in conflict and error monitoring in the Simon task. On the basis of the adaptation by binding account for conflict adaptation and the orienting account for post-error slowing, we predict a dissociation between conflict and error monitoring. This prediction is tested and confirmed as conflict adaptation is task-specific while post-error slowing is not.

Introduction

Cognitive control refers to information processing adjustments in order to optimize task performance, which typically occurs after problems were encountered. One task that has been extremely useful in the study of such control-related adjustments is the Simon task because of its compatibility manipulation (Simon, 1990). The typical problems that call for an adjustment in the Simon task are response conflict and errors. Response conflict occurs on incompatible trials (e.g., stimulus on left location requiring a right response) because on these trials, two opposing responses are activated; one on the basis of the relevant stimulus information, usually color or shape, and one on the basis of the irrelevant stimulus location (e.g., Zorzi & Umiltá, 1995). In most cases response conflict does not result in an error, but sometimes the activation of the incorrect response is so strong that an error is observed. Obviously, errors also occur on compatible trials but less often so and for various reasons (e.g., fluctuations in attention).

In the present paper we address the question whether errors and conflict trigger the same behavioral adjustments. The dominant perspective on conflict and error monitoring is that both processes are supported by similar brain areas (Kerns, Cohen, MacDonald, Cho, Stenger, & Carter, 2004) and involve similar computations (Botvinick et al., 2001, Yeung et al., 2004). We challenge this mainstream point of view and propose that behavioral adaptation after errors differs from adaptation after conflict. We will review the literature on conflict adaptation specifically focusing on the Simon task and present our computational model for conflict adaptation (Verguts & Notebaert, 2008, Verguts & Notebaert, 2009). We will subsequently present recent work on post-error slowing and describe our orienting account (Notebaert et al., 2009). Interestingly, both accounts predict a dissociation between conflict and error monitoring. This prediction is finally tested in an experiment where Simon and SNARC trials are randomly presented (Notebaert & Verguts, 2008).

Conflict adaptation was initially demonstrated in the flanker task where the flanker effect was smaller after incongruent trials (> < >) than after congruent trials (> > >; Gratton, Donchin & Coles, 1992). Similarly, conflict adaptation in the Simon task was suggested by the observation that the Simon effect was smaller after incompatible trials than after compatible trials (Notebaert et al., 2001, Stürmer et al., 2002). One explanation for this Gratton effect is that participants increase control after incompatible trials, for instance, by blocking the processing pathway that processes the irrelevant spatial information (Stürmer et al., 2002).

Importantly, in the standard Simon task with two colors, two stimulus locations and two response locations, the data pattern can also be explained in terms of feature repetition and integration effects (Notebaert et al., 2001, Hommel et al., 2004). Today, there are different versions (e.g., Mayr et al., 2003, Nieuwenhuis et al., 2006) of this alternative account but the main idea is that the data pattern (reduced compatibility effects after incompatible trials) can equally well be described in terms of stimulus and response repetition and alternation effects. More particular, the two trial sequences that benefit from ‘conflict adaptation’ are compatible–compatible and incompatible–incompatible sequences. In a Simon task, and in any other congruency task, these and only these sequences include exact stimulus repetitions (location and color repetition) and complete stimulus alternations (location and color alternation), and it has been demonstrated that complete repetitions and complete alternations are faster than partial repetitions. This effect is typically explained in terms of feature integration effects (Hommel et al., 2004). Very reasonably, researchers raised the question why we should need an extra mechanism (conflict adaptation) for explaining behavioral data that are already explained by a mechanism that also explains different effects?

The discussion whether this behavioral pattern reflects conflict adaptation or merely repetition effects continues. In our opinion, however, there are some indications that the effect is not entirely due to repetition effects. A first indication is delivered by electrophysiological studies. Stürmer et al. (2002) showed that the Simon effect disappeared (or even reversed) after incompatible trials. Most interestingly, the data also demonstrated a modulation of the lateralized readiness potential (LRP), which is hard to explain with feature integration and repetition effects. Typically, the LRP shows initial incorrect response activation on incompatible trials, which is then later followed by activation of the correct response. Stürmer et al. showed that this initial incorrect response activation disappeared after incompatible trials. They explained the results in terms of suppression of the irrelevant route after incompatible trials. Additionally, Stürmer, Redlich, Irlbacher, and Brandt (2007) demonstrated that administering transcranial magnetic stimulation (TMS) on left dorsolateral prefrontal cortex (DLPFC) 300–500 ms preceding the next stimulus abolishes the sequential modulation of the Simon effect.

Further support for conflict adaptation was provided by studies demonstrating the behavioral effect in situations where repetitions were excluded. Wühr (2005) investigated this by using four stimulus locations and two response locations. Red and green stimuli were presented on the corners of an imaginary square with a vertical response dimension operated with the left and right hand. Half of the participants responded to the upper key with the left hand and to the lower key with the right hand while this was reversed for the other half of the participants. The reaction times confirmed conflict adaptation in the sense that conflict adaptation was observed for trial types that did not differ in stimulus or response overlap (e.g., when all trials were complete alternations). Akcay and Hazeltine (2007) reached similar conclusions in a Simon task with 4 stimuli, 4 locations and 4 responses.

In a further attempt to disrupt all overlap between two consecutive Simon trials, Akcay and Hazeltine (2008; exp. 4) combined two Simon tasks. One was a letter classification task with four letters and the other a color discrimination task with three colors. One task had a vertical stimulus and response arrangement while the other had a horizontal arrangement and both tasks were presented on one side of the fixation cross. The pattern of data showed conflict adaptation on same-side sequences while no conflict adaptation was observed for opposite-side sequences. Because the trials included in this analysis did not contain stimulus or response repetitions, the authors concluded that sequential modulations on the same-side sequences were due to control, which was recruited locally, within one type of Simon task.

In sum, conflict adaptation has been observed in a variety of tasks where feature repetitions were excluded. The study by Akcay and Hazeltine (2008) on the other hand suggests that some kind of overlap between trial n-1 and trial n is required, in the sense that conflict in a letter Simon task, does not modulate performance in a subsequent color Simon task. The authors point out that this indicates that cognitive control is a local effect, as opposed to a global (task-aspecific) effect. A similar conclusion was obtained in the study of Notebaert and Verguts (2008). In this study a Simon task was combined with a SNARC task (spatial numerical associations of response codes). The SNARC effect is the observation that small numbers are responded to faster with the left response key and large numbers faster with the right response key (Dehaene, Bossini, & Giraux, 1993). In this particular experiment, participants had to respond to the orientation of the numbers with a left or right response key. The Simon task either used the same relevant information (orientation of a laterally presented X) or different relevant information (color). Both tasks were randomly intermixed and conflict adaptation across the tasks (conflict in Simon trial reduces SNARC effect and conflict in SNARC reduces Simon effect) was only observed in the condition where both tasks used the same relevant information, further supporting the notion of local or task-specific control.

In order to explain local or task-specific control effects Verguts & Notebaert, 2008, Verguts & Notebaert, 2009 proposed the adaptation by binding account, which integrates the repetition/integration perspective and the conflict adaptation perspectives. The basic principle is that associations connecting active (stimulus, response, or other) units are strengthened proportional to the amount of conflict on the current trial. For example, when Task 1 is presented, the S–R connections of Task 1 are more active than the S–R connections of Task 2. Consequently, when conflict occurs, the S–R connections of Task 1 will be strengthened to a larger degree than the connections of Task 2. As a result, control is implemented in a task-specific way. A similar associative mechanism was proposed by Blais et al., 2007, Davelaar & Stevens, 2009. Important for the present purposes is that adaptation by binding generates a dissociation between conflict and error monitoring. Before we elaborate on this issue we will first briefly describe recent theoretical advances in the domain of error adaptation.

Although error adaptation can take on many forms, we will focus here on post-error slowing, the observation that RTs are slower after errors than after correct trials. In the Simon task, this was demonstrated by Ridderinkhof (2002). It is generally thought that post-error slowing reflects a strategic adjustment following an error in order to avoid another error (e.g., Botvinick et al., 2001, Ridderinkhof, 2002). Problematic for explanations in terms of strategic adjustments, however, is that usually accuracy does not improve following errors. In Ridderinkhof (2002) for instance, there was no difference between post-error and post-correct accuracy and others have reported worse performance after errors than after correct trials (e.g., Notebaert et al., 2009). We therefore recently proposed an alternative account for post-error slowing (Notebaert et al., 2009). We postulated that post-error slowing could be explained in terms of an orienting response towards infrequent events, which errors typically are. Using a color discrimination task, we demonstrated that post-error slowing depends on the frequency of an error, not on the error information itself. Post-error slowing was observed only when errors were infrequent; when instead correct trials were infrequent, post-correct slowing was observed. Moreover, we observed similar slowing after completely irrelevant unexpected signals. We interpreted this in terms of an orienting response towards the unexpected signals that delays processing of the next stimulus. In a follow-up study, we demonstrated that the size of the feedback-related P3 and not the error-related or feedback-related negativity predicted the RT on the following trial. As the frontocentral P3 is generally considered as an index of orienting, this was in line with the orienting account (Nunez Castellar, Kühn, Fias, & Notebaert, 2010).

Both on the basis of adaptation by binding (conflict adaptation) and the orienting account (error adaptation), a dissociation between conflict and error adaptation is predicted. The adaptation by binding account states that active associations will be strengthened when conflict is detected. This principle leads to increased control after correct (incompatible) trials but does not necessarily improve performance after incorrect trials. Indeed, in incorrect trials, incorrect stimulus–response representations are presumably more active than correct representations, so their associations will be strengthened, leading to worse performance. We assume that binding also operates after errors, it is only very hard to deduce which were the active representations. Most likely, there is a lot of variability with respect to the activation patterns that lead to an error, blurring the effect of binding after errors.

Also on the basis of the orienting account, a dissociation is predicted. If post-error slowing is caused by the detection of an unexpected event which triggers an orienting response towards this event and delays subsequent stimulus processing, it should not matter what task follows this period of distraction. More specifically, the orienting account predicts that post-error slowing should not be task-specific, meaning that an error in Task 1 should result in post-error slowing in the subsequent Task 2. We therefore predict that post-error slowing is not task-specific and should occur for any sequence of tasks. In order to investigate this prediction, we measured post-error slowing in data (Notebaert & Verguts, 2008) where we reported across-task conflict adaptation when both tasks shared the relevant dimension, and no such effect when they did not. In contrast to conflict adaptation, we predict post-error slowing in both conditions. Moreover, on the basis of the orienting account, we also predict that the size of post-error slowing will be related to the accuracy level of each participant, in the sense that participants who make more errors will be less surprised by an error. In particular, we predict a negative correlation between error percentage and post-error slowing.