Sequential modulations of the Simon effect depend on episodic retrieval.

Sequential modulations of conflict effects, like the reduction of the Simon effect after incompatible trials, have been taken to reflect the operation of a proactive control mechanism commonly called conflict monitoring. However, such modulations are often contaminated by episodic effects like priming and stimulus-response feature integration. It has previously been observed that if the episodic representation of a conflicting trial is altered by rotating the stimulus framing 180(∘) around its axis, the subsequent "conflict adaptation" pattern is eliminated. In Experiment 1, we replicate the findings and provide the basic episodic interpretation. In Experiment 2, we extend the framework to include rotations of 90(∘), and verify that the episodic effects generalize to scenarios of neutral compatibility. Finally, in Experiment 3, we add complete, 360(∘) rotations, and show that the episodic manipulation by itself does not eliminate the conflict adaptation patterns - as long as conditions favor episodic retrieval. The experiments are argued to demonstrate that an episodic account of the conflict adaptation effect can most parsimoniously account for the behavioral effects without relying on higher order cognition. Accordingly, we conclude that conflict adaptation can be understood either as critically depending on episodic retrieval, or alternatively reflecting only episodic retrieval itself.

INTRODUCTION

The time it takes to act is strongly affected by the compatibility between the stimulus and response (Fitts and Seeger, 1953). Simon and Rudell (1967), for example, showed that processing the location of a stimulus automatically triggers a response toward the source of the stimulus, resulting in frequent errors and increased reaction latencies if a stimulus appears in a location opposite to the response. This effect later became known as the Simon effect (coined by Hedge and Marsh, 1975), and is one of the more popular effects amongst the range of stimulus-response compatibility phenomena (see for an overview Alluisi and Warm, 1990). Like the Stroop effect (Stroop, 1935), and the flanker-compatibility effect (Eriksen and Eriksen, 1974), the Simon effect follows the general rule that if a task-irrelevant dimension of a stimulus suggests a different response than the relevant dimension, performance will be impaired.

CONFLICT CONTROL

The location of a stimulus may attract a response in the wrong direction, a word’s meaning sometimes prompts an incorrect verbalization and a peripheral stimulus can distract by cueing inaccurate actions, yet, overall, we are capable of withstanding temptation and can carry out Simon, Stroop, and Eriksen tasks eventually. Models of cognition typically account for this ability by implementing a function which detects and resolves the conflicting responses using executive or conflict control. A conflicting readiness of the motor cortex can indeed be detected using EEG (Stürmer et al., 2002), which might act as a trigger for the cognitive system to utilize cortical areas associated with cognitive control. Then, the control itself could be achieved by facilitating task-related information, thereby supporting the correct response (e.g., Botvinick et al., 1999; Egner and Hirsch, 2005). Alternatively, incorrectly triggered response alternatives could be actively suppressed, similarly biasing the response competition (e.g., Band et al., 2003; Ridderinkhof et al., 2004). Either approach thus assumes that the cognitive system continuously monitors for conflict and enhances control upon detection.

One of the most important sources of evidence for these conflict-control models is provided by the so-called “Gratton Effect,” concerning sequential effects in conflict-inducing tasks (Sheth et al., 2012). This was named after Gratton et al. (1992), who showed that the impact of response-compatible and incompatible flankers on performance is reduced in trials that follow trials with incompatible flankers as compared to trials with compatible flankers. Comparable observations have been made with the Stroop task (e.g., Kerns et al., 2004) and the Simon task (e.g., Praamstra et al., 1999), with either effect being reduced, eliminated and sometimes even reversed after incompatible trials. These observations have been taken as evidence that facing a conflict trial induces an increase of cognitive control, which then proactively facilitates the resolution of conflict in the next trial – resulting in the observed reduction of subsequent conflict effects.

EVENT FILES

Later considerations and findings have, however, raised some doubts on the interpretation of sequential conflict effects as reflecting a universal, conflict monitoring function. As pointed out by Mayr et al. (2003) and Hommel et al. (2004), sequential relationships between compatibility and incompatibility are naturally confounded with particular patterns of stimulus and response repetitions and alternations. Approximately half a century of research on the effect of priming shows that simply repeating a stimulus or response markedly affects reaction times (Bertelson, 1963; Meyer and Schvaneveldt, 1971) and during sequence modulations, such effects are always present. Given that the combinations of stimulus and response repetitions are not equally distributed across the possible transitions between compatibility conditions, it is possible that at least some sequential modulation effects are due to feature – rather than conflict – repetition (Mayr et al., 2003; see also Figure row 1).

Sequence of events in five trial pairs and coding in terms of conflict (Control account) and features (Event File account). Given that the participant responds right (R) to circles and left (L) to stars, the initially compatible (c) trial (S1) in pair 2 is followed by an incompatible (I) trial (S2), which usually results in increased reaction times (RT: ++). Increased reaction times are also predictable in this scenario from an Event File perspective, as the shape is repeated (=) between trials, but not (≠) the position (Pos). This holds for trial pairs 1, 3, and 4 as well. However, divergent predictions were based on the gradual rotation as is depicted in row 5: whereas nothing changes in terms of conflict, the Event File model would predict performance gains (RT: --).

Sequences of stimulus-response combinations do not only invite priming effects. According to Treisman and colleagues, processing the features of an object leads to a binding of the corresponding feature codes (e.g., Treisman, 1996). They found that the priming effect obtained by repeating an object is enhanced if this object also appears in the same location, suggesting that object features get bound to location codes (Kahneman et al., 1992). Hommel (1998) and Hommel et al. (2001) extended this concept to include action and suggested that object features and action features may be spontaneously integrated into what they call event files. If, for instance, a stimulus like a cup of coffee is accompanied by an action like grasping or drinking, the codes of the sensory features (BROWN, WARM, etc.) become integrated with action features (moving the hand toward the object, the typical type of grasp, etc.), resulting in an event file of “drinking coffee.” If one or more features are encountered again the event file is retrieved automatically (cf., Logan, 1988) in a kind of pattern-completion process. This commonly leads to benefits (e.g., if recognition is hampered by suboptimal visibility), but to costs if some features are repeated but now combined with other features. Thus, perceiving the coffee cup again and immediately retrieving the drinking action may lead to an unpleasant surprise if the contents turned cold in the meantime.

These partial repetition costs also come into play during sequential conflict studies. Hommel et al. (2004) showed that partial-overlap costs are commonly confounded with the sequence of compatible and incompatible trials in the Simon task (and other conflict tasks). For instance, if a participant responds with left to a right-sided stimulus, the event is incompatible (I), whereas responding left to a left stimulus is compatible (C). Consider how each of these cases could affect subsequent conflict, such as when a right response would be required to a left location. In terms of the conflict adaptation effect, a compatible trial followed by an incompatible trial (c-I) results in slow reaction times, but an incompatible trial followed by another incompatible trial (i-I) does not. However, if we deconstruct each sequence in terms of their features, it turns out that in the C-I case (e.g., right| right → left| right), one features is repeated, but the other one is not, resulting in partial repetition costs and slow performance (see Figure row 2). In the I-I case, either the features both change (left| right → right| left, see Figure row 4), or the features both repeat (left| right → left| right), so there would be no partial repetition costs and faster performance is predicted. Thus, a “conflict adaptation” pattern is elicited by matters entirely unrelated to the change in conflict itself.

EXCLUDING REPETITIONS IN CONFLICT ADAPTATION

Given that conflict tasks rely on the interrelationship of stimulus and response features, the confound between feature repetitions and the repetitions of compatibility conditions is to some degree unavoidable – at least if the original tasks are left more or less intact. A common workaround solution is to add a layer of complexity to the simple tasks by adding constraints to the randomized design. For instance, some studies have considered only those conditions where no stimulus or response feature is repeated (e.g., Akçay and Hazeltine, 2007) and, given that sequential effects were still obtained, been taken to demonstrate purely executive effects. Even though this approach seems straightforward, it creates two somewhat related problems.

One problem is that excluding any feature overlap between two successive stimuli or stimulus-response episodes does not exclude proactive effects of episodic integration and retrieval (Dutzi and Hommel, 2009). Consider stimuli that vary on two dimensions, such as the visual letters “X” and “O” appearing in red or green. According to the available models of feature integration (e.g., Duncan and Humphreys, 1989), processing a green “X,” for instance, would lead to the competition between codes of the colors GREEN and RED and between codes of the shapes X and O. Collecting sufficient visual evidence should provide sufficient support for GREEN and X, which helps them to outcompete the possible alternatives. Now consider that you process the green X after having seen a red O. If having processed the red O led to the integration of the codes RED and O (Kahneman et al., 1992), they can be assumed to act as a unit and engage in what Duncan (1996) and Duncan et al. (1997) called integrated competition. This has advantages for the integrated unit if the stimulus it encodes is repeated but a competitive disadvantage if the stimulus changes: having integrated RED and O makes it easier to reject them as a unit (Duncan and Humphreys, 1989). Any loss of RED in the competition with GREEN will propagate to and thus weaken O as well, and losses of O in the competition with X will propagate to and weaken RED. Empirical evidence for this mechanism has been obtained in search tasks, where non-targets can be more easily rejected if they share features that do not overlap with the target, so that they can be grouped together and rejected as a group (Duncan and Humphreys, 1989). Also in line with expectations from an integrated competition approach is the observation that alternating all features and aspects of a stimulus or stimulus-response episode leads to performance that is as good as (Hommel, 1998) and sometimes even better than performance with complete repetitions (Hommel and Colzato, 2004; Colzato et al., 2008). In any case, it seems clear that avoiding feature overlap between successive trials does not allow one to exclude contributions from feature integration and episodic retrieval.

Another problem in restricting analyses to alternation trials is that this amounts to selecting a single data point which ignores all other interactions between repetitions and alternations of features and leaves out the possibility that control and retrieval might interact, as suggested by Spapé and Hommel (2008). In their study, participants responded to high and low-pitched tones by saying “high” and “low,” while ignoring voices saying “high” and “low.” Unsurprisingly, this created a Stroop-like effect if a word was incompatible with the tone. A typical sequential modulation effect was also obtained, with reduced Stroop effects after incompatible trials. However, if the voice changed between the two successive trials, no such effect remained. Thus, they argued, control information was integrated with the episodic context – i.e., the voice. Only if the episodic context was retrieved did control adaptation affect performance.

Task-switching studies provide support for this interpretation. While switching to a new task is difficult in general (Allport et al., 1994), switching costs are particularly pronounced if the current stimulus was previously encountered in a competing task (e.g., Waszak et al., 2003). This suggests that stimuli and task information are integrated into episodic bindings that are retrieved if the stimulus is encountered another time – which is beneficial if the task has remained the same (as is usual in everyday life) but problematic if the task is different.

Another converging line of evidence comes from task-switching studies that employ a type of conflict task. Evidence from such tasks suggests that a task-switch can result in an elimination of the conflict adaptation effect (Kiesel et al., 2006). In the absence of any similarity between tasks, however, Notebaert and Verguts (2008) observed no sequential effects, suggesting to them that conflict-monitoring acts locally (see also Blais et al., 2007).

To summarize, there are reasons to assume that at least some of the effects that are commonly taken to reflect adaptive control actually reflect stimulus-response integration. These effects cannot be avoided by restricting one’s analyses to alternation trials.

AIM OF STUDY AND RATIONALE

The aim of the present study was to investigate the relationship between adaptive control processes and episodic retrieval in producing sequential modulations in a conflict task, and to test the hypothesis that the former may depend on the latter. To do this, we used an effect that has before been shown to selectively affect episodic retrieval. We replicate previous findings that demonstrate how this simple, episodic manipulation can have strong effects on conflict control. Here, and throughout the article, we provide a side-by-side comparison of the episodic effect in terms of conflict control and feature integration. We then report two additional experiments which confirm separate predictions that relate episodic retrieval to conflict control. In Experiment 2, we show how feature integration effects can be found even if stimuli are repeated in entirely new positions of neutral compatibility. In Experiment 3, we show that the manipulation itself does not eliminate the conflict adaptation pattern by providing evidence that under conditions that favor episodic retrieval, this pattern can be re-established. Accordingly, each experiment concerns the primacy of episodic retrieval: if the conflict adaptation is determined by episodic retrieval, we should conclude it to either require, or be redundant to, episodic retrieval. But let us first consider the episodic effect under consideration.

Already as part of the first studies on feature integration by Kahneman et al. (1992), it has been shown that if a cue is displayed within a bounding box, a priming effect can be observed if a probe appears inside the same box, even if the box has gradually moved between prime and probe to the new position. Thus, feature integration theory allows that the letters and boxes were bound into enduring representations that were updated along with the boxes’ gradual change of position. Simplifying the task considerably, we showed that action features are likewise bound with the object: after a rotation of the stimulus presentation along its axis, a repetition of the action and shape still resulted in performance benefits (Spapé and Hommel, 2010).

Spapé et al. (2011) made use of this effect by transforming the object-reviewing or tracking task into a sequential Simon task. Figure gives a brief overview of the task and of how the conditions relate to the conflict-control and event-file accounts. If we imagine the task requires a right response for each circle that is portrayed, and a left one for each star, the first row shows a cI trial sequence (a compatible initial trial, S1, followed by an Incompatible paired trial, S2) – which typically elicits maximal errors and reaction times. Since the same shape (shape=) is displayed in a new location (pos≠), the event file account likewise predicts poor performance for the same trial sequence. In the second row – and, in fact, in all 8 different combinations of trials (see Hommel et al., 2004) – the same configurations of features cannot distinguish between the two accounts.

However, by gradually having the two boxes exchange positions, as schematically portrayed in the third row, the same predictable effect demonstrated by Spapé and Hommel (2010), should occur. That is, the circle initially displayed right should, following the gradual migration, be represented on the left side, resulting in a complete repetition, which usually results in performance benefits. Meanwhile, there is no reason to assume that the gradual change affects control processes: if registering conflict upon R1 selection suppresses or prevents the processing of stimulus location on S2 presentation, this should not be affected by the presence or absence of a rotation of actually task-irrelevant boxes in between two trials. Therefore, the episodic manipulation of rotating the boxes was not predicted to affect performance from a conflict-control account but only from an event file perspective.

Although neither the boxes, nor the rotation thereof, was relevant to the task or conflict, Spapé et al. (2011) observed strong effects of rotation on conflict adaptation in terms of behavior, as well as event related potentials of the EEG known to be involved in attention and control (the N2) and response readiness (the lateralized readiness potential, LRP): nearly all costs associated with conflict adaptation were removed. However, they did not report the event file analyses, making it difficult to assess whether all predictions from that perspective were fulfilled. Secondly, they also note some curious effects that do not immediately follow from a pure event-file approach. For instance, though they reported that rotation disrupted conflict-adaptation effects, the conflict adaptation effect in terms of psychophysiological indicators did not reverse. That is, in terms of LRP and N2, rotation, the compatibility effect no longer depended on the preceding trial. One could therefore argue that the rotation, rather than causing a change in represented position, induced a cognitive reset, undoing both the conflict adaptation and the event file.

In order to better understand the effects of rotation on both conflict control and event files, we report in this series of experiments both types of analysis side-by-side. In the conflict-control analysis, we examined the data in terms of the sequence of compatibility conditions, testing whether S1 incompatibility would reduce the S2 compatibility effect, and aiming to replicate that rotation, a factor that should be meaningless from a conflict-control point of view, affected the interaction between S1 and S2 compatibility. In the event-file analysis, we examined whether response-repetition and stimulus-location would interact in the standard partial repetition-cost pattern: complete repetitions and alternations should result in better performance than repeating one feature, but not the other (Hommel and Colzato, 2004). Of particular theoretical interest here was whether rotation would tend to eliminate these effects (as the two-event-files account of Spapé and Hommel, 2010, would suggest) or even reverse their sign (as a one-file extension of the approach of Kahneman et al., 1992, might imply).

In Experiment 1, we will first demonstrate these two types of analyses in a straight replication study. In the subsequent two experiments, we continue this pattern but extend it to new territories. In Experiment 2, we show the effects of rotating conflicting objects to completely new (in event file terms) and sometimes neutral (in conflict terms) locations by including rotations of 90∘. Finally, in Experiment 3, we allow rotations to return to their initial position as well, using a full 360∘ rotation which completes the picture and confirms that the rotation by itself does not destroy the conflict adaptation effect.

EXPERIMENT 1

In a Simon task, participants respond to a non-spatial stimulus feature by carrying out a left or right response, whilst ignoring the irrelevant location of the stimulus. In our version, participants responded to circles and stars by pressing a left or right key. The stimuli appeared in the left or right of two constantly visible boxes. Trials were presented in pairs, so that one circle or star was presented (S1) and responded to (R1) before a second circle or star (S2) appeared to signal a second response (R2). The boxes remained visible in between the two trials of a pair and were rotated by 180∘ in 50% of the trials.

METHOD

Participants

Eighteen students from Leiden University voluntarily participated in this experiment in exchange for money or course credits.

Apparatus and stimuli

Stimuli were presented on a flat-screen 17′′ CRT monitor in 800 × 600 pixels resolution and a refresh-rate of 120 Hz. A Pentium-IV dual 1.67 GHz PC running E-Prime (1.1, SP3) on Windows XP SP2 controlled stimulus-presentation and recorded reactions via the USB connected keyboard. The boxes were gray (RGB value of 128, 128, 128), black-lined squares of 60 × 60 pixels or approximately 32 × 32 mm presented against a silver (RGB value of 191, 191, 191) background. The target itself was also 60 × 60 and was either a circle or a four-pointed star. Boxes were presented 180 pixels (approximately 96 mm) left and right from the center of the screen and also kept at this distance during the gradual shifts in location.

Procedure

As outlined in Figure , a fixation cross was presented for 500 ms, after which the two boxes were presented on the left and right of the screen, one of them containing the target shape (S1) to which participants were required to respond. After 500 ms, the targets were no longer shown on the screen. In the “static” condition, the boxes stood still, without targets, for 800 ms, whereas in the rotating condition, they rotated around their axis at a speed of approximately 4∘ with each 44 ms. After the 800 ms, both in the static condition and the rotating condition, the boxes were presented for another 200 ms before the second target (S2) was presented. S2 was shown for 700 ms before a screen with feedback informed the participant of the performance. This last screen also comprised the inter-trial interval and was shown for 1100 ms.

Schematic depiction of the trial-sequence of two example trials. After presenting a fixation crosshair, two boxes were presented for 500 ms in the left and right of the screen, one containing the shape (S1) to which participants were required to respond. In the “static” condition (left), an inter-stimulus interval (ISI) followed in which the boxes stood still for 800 ms, whereas in the rotating condition, they rotated around their axis during this ISI. In both conditions, the boxes were statically presented for another 200 ms before the second target (S2) was shown. S2 was shown for 700 ms before an inter-trial interval of 1100 ms ended the trial.

Following the instruction, the first 20 trials of the experiment were considered practice of acquiring the mapping between circles or stars with a or