Acta Psychologica 159 (2015) 52–60

Contents lists available at ScienceDirect

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

Incorrect predictions reduce switch costs Thomas Kleinsorge ⁎, Juliane Scheil Leibniz Research Centre for Working Environment and Human Factors, Germany

a r t i c l e

i n f o

Article history: Received 12 December 2014 Received in revised form 26 March 2015 Accepted 20 May 2015 Available online xxxx Keywords: Task switching Cognitive control Conflict adaptation

a b s t r a c t In three experiments, we combined two sources of conflict within a modified task-switching procedure. The first source of conflict was the one inherent in any task switching situation, namely the conflict between a task set activated by the recent performance of another task and the task set needed to perform the actually relevant task. The second source of conflict was induced by requiring participants to guess aspects of the upcoming task (Exps. 1 & 2: task identity; Exp. 3: position of task precue). In case of an incorrect guess, a conflict accrues between the representation of the guessed task and the actually relevant task. In Experiments 1 and 2, incorrect guesses led to an overall increase of reaction times and error rates, but they reduced task switch costs compared to conditions in which participants predicted the correct task. In Experiment 3, incorrect guesses resulted in faster performance overall and to a selective decrease of reaction times in task switch trials when the cue-target interval was long. We interpret these findings in terms of an enhanced level of controlled processing induced by a combination of two sources of conflict converging upon the same target of cognitive control. © 2015 Elsevier B.V. All rights reserved.

1. Introduction The ability to act flexibly in accordance with a permanently changing environment is one of the core requirements of human behavior needed every day. To cope with changing demands, cognitive control is needed. Cognitive control comes in many guises, and the last couple of years have seen a surge in interest in how different manifestations of cognitive control relate to each other. One important distinction with respect to the study of cognitive control is between conditions affording the engagement of cognitive control like stimulus incongruence, changes in processing requirements, or errors, and manifestations of an engagement of cognitive control like sequential modulations of congruency effects, reductions of switch costs, or post-error adaptations. Unfortunately, this distinction often becomes blurred when studying sequential adaptations in order to elucidate dynamic adjustments of cognitive control. One example is the study of congruency sequence effects (CSEs) that has garnered a lot of empirical effort in recent years (for reviews, cf. Duthoo, Abrahamse, Braem, Boehler, & Notebaert, 2014; Egner, 2014). The signature of CSEs is a modulation of a congruency effect that depends on the congruency of the preceding trial, with the congruency effect being reduced after incongruent as compared to congruent trials. One dominant account of such effects in terms of the conflict-monitoring theory (cf. Botvinick, Braver, Barch, Carter, & Cohen, 2001) posits that incongruency in a preceding trial triggers the engagement of cognitive ⁎ Corresponding author at: Leibniz Research Centre for Working Environment and Human Factors, Ardeystraße 67, D-44139 Dortmund, Germany. E-mail address: [email protected] (T. Kleinsorge).

http://dx.doi.org/10.1016/j.actpsy.2015.05.006 0001-6918/© 2015 Elsevier B.V. All rights reserved.

control, resulting in a reduced congruency effect in the following trial. Accordingly, incongruent trials are both triggers for engaging cognitive control as well as targets for controlled processing. Using basically the same task as the trigger as well as the target for the engagement of cognitive control has resulted in extended controversies regarding the (relative) contribution of bottom-up versus top-down factors in the dynamic regulation of cognitive control (cf. Egner, 2014). Trying to tease these factors apart by using different tasks as triggers and targets for control has resulted in an amazingly heterogeneous picture, with sometimes subtle differences between conditions determining whether adjustments of control occur or not. For example, Kim and Cho (2014) investigated the CSE across two different flanker-compatibility tasks that were presented alternately in a trial-by-trial manner. When participants responded in both tasks with four fingers of the same hand, a significant CSE accrued. However, when the two tasks engaged fingers of different hands, no CSE could be observed. Although the exact reasons for these divergent observations are not entirely clear up to now, it is likely that a crucial factor consists of the degree to which both tasks are represented in an overlapping manner within the same task-control structure (Kim & Cho, 2014). In the present study, we attempted to vary the degree to which the engagement of control was triggered by manipulating two sources of conflict related to the same target of control, namely the identity of the task to be performed on the next trial. This was done by complementing a task-switching procedure with the requirement to explicitly guess aspects of the forthcoming task. With respect to task switching, most theories assume that the main challenge of the cognitive system consists of the overcoming of cognitive settings and action tendencies (‘task sets’) that were induced by

T. Kleinsorge, J. Scheil / Acta Psychologica 159 (2015) 52–60

the recent performance of another task (for a review, cf. Vandierendonck, Liefooghe, & Verbruggen, 2010). Thus, when there is a requirement to perform another task a conflict arises between the action tendencies carried over from the preceding trial and the requirements stemming from the actual task. On the other hand, we assumed that explicitly guessing the next task is associated with enhancing the activation of the corresponding task set. If the task to be performed is different from the one that was explicitly indicated, this mismatch should induce another kind of conflict. We do not assume that the ‘repetition bias’ induced by the recent performance of a task and the biased expectancy induced by explicit guessing converge upon exactly the same level of task conflict in terms of the source of conflict; however, we hypothesize that both types of conflict are resolved by strengthening the actually relevant task representation, resulting in some kind of convergence in terms of the target of cognitive control. This, in turn, led us to predict that incorrect guesses would result in a reduction of switch costs because the conflict induced by an incorrect guess should increase the amount of cognitive control needed to resolve the conflict induced by the requirement to switch the task. Preliminary evidence for the viability of this reasoning comes from a recent study by Duthoo, De Baene, Wühr, and Notebaert (2012). In this experiment, participants switched among two tasks and were required to predict the next task during the inter-trial interval. A main observation as reported by the authors consisted of a disappearance of switch costs when participants predicted a task alternation. However, because these authors analyzed their data in terms of the factor ‘repetition predicted’ versus ‘switch predicted’ (instead of correct versus incorrect prediction, as we did in the experiments reported below), the interaction in the focus of the present study went unnoticed by these authors. If the data reported by Duthoo et al. are analyzed in the way we analyzed our data, an interaction of task prediction and task transition emerges (W. Duthoo, personal communication, November 13, 2013): In terms of RTs, mean switch costs for correctly predicted trials amounted to 95 ms, whereas mean switch costs for incorrectly predicted trials amounted to only 61 ms. Even more pronounced was the effect on ERs. Mean error switch costs amounted to 2.3% for correctly predicted trials, whereas a switch benefit of 1.1% accrued for incorrectly predicted trials. Thus, these observations are in line with our assumption of a switch-cost reducing effect of incorrect task predictions. In the experiments reported below, participants switched among four tasks. In the main phase of the first two experiments, participants were asked to make explicit predictions about the forthcoming task during the interval separating the performance of two trials. Given that the sequence of tasks was completely random and participants were informed about this, these task predictions were based on guessing. Predicting the wrong task was assumed to result in a performance decrement in the first place. However, based on the considerations outlined before, we assumed that this performance decrement would be offset to some degree by a transient boost of cognitive control which should facilitate the performance of task switches more than the performance of task repetitions, resulting in a reduction of task switch costs. We also varied the duration of the cue-target interval (CTI) in order to capture the dynamics of the effect of incorrect guesses. With a short CTI, there is a relatively long interval during which participants can prepare for a guessed task, whereas preparation for the correct task as indicated by the precue is rather restricted. Conversely, with a long CTI, there is relatively much time to counteract effects of an incorrect guess. If, for example, participants inhibit the previous task in case of an incorrectly guessed task switch, the expected slowdown of task repetitions that were guessed to be task switches should be more pronounced with a short compared to a long CTI. The first two experiments differed with respect to their motor requirements. In Experiment 1, participants indicated their guess by pressing the central key of one of four rows of keys that were assigned to one of the four tasks each. As a consequence, in case of an incorrect

53

guess participants were required to respond to the imperative stimulus by pressing one of two keys that were associated with another task (for details, see below). This factor was removed in Experiment 2 in which task indications and task responses were performed with nonoverlapping sets of keys that were operated with different hands. 2. Experiment 1 2.1. Method 2.1.1. Participants 19 right-handed subjects (7 male, 12 female) with normal or corrected-to-normal vision participated. Their mean age was 24.7 years (range: 19–30). 2.1.2. Stimuli, tasks, and apparatus Imperative stimuli consisted of digits from the range 1–9 (excluding 5) and the letters A, B, G, E, N, O, S, and U. Each digit was about 7 mm high × 4 mm wide. Digits and letters were presented side by side, their position was chosen randomly on every trial. Task precues consisted of a dark blue square, diamond, circle or triangle surrounding the position of the imperative stimulus with a size of about 15 cm × 15 cm. There were four tasks, two of them regarding the digit and two regarding the letter. The numerical judgment tasks either concerned the magnitude (smaller vs. larger than five) or the parity of the digits. The magnitude task was indicated by the diamond, the parity task was indicated by the circle. The letters had to be judged regarding their position in the alphabet (first or second half, indicated by the triangle) or regarding whether it is a consonant or a vowel (indicated by the square). Stimuli were presented centrally on a 17″ monitor in black on light-gray background. Viewing distance was not controlled, but equally given with approximately 60 cm. The response device consisted of a custom-built keyboard (cf. Fig. 1) connected to a Fujitsu Esprimo P700 that was equipped with an external data acquisition module (National Instruments NI USB-6431). The response device registered not only the pressing but also the release of each individual key with a precision of about 1 ms. Each row of three keys was assigned to one of the tasks during the whole course of the experiment. The central key of each row was used to indicate the guessing response and was attached with a sticker that depicted the task cue associated with the respective task (see Fig. 1). The two outer keys of each row were used to respond to the imperative stimulus. Participants were instructed to use only the index finger of their right (dominant) hand for responding. 2.1.3. Design and procedure The experiment consisted of three phases. The first and third phases were designed as the usual cuing-variant of the task switching paradigm. During the second phase, participants additionally had to guess at the beginning of each trial on which of the tasks they would have to perform in this trial. Switching probability was .5 during the whole experiment. At the beginning of the experiment, participants were provided with on-screen instructions in which the tasks and the meaning of the task cues were explained. The first phase, in which no guessing was required, consisted of three blocks of 120 trials each. The response–stimulus interval (RSI), separating the response in trial n-1 from the onset of the imperative stimulus in trial n, was set to 1100 ms in the first and third phases of the experiment and to 3600 ms during the second (guessing) phase. In case of an error, error feedback was presented for additional 1000 ms; in case of reaction times (RTs) slower than the RT deadline of 2500 ms, RT feedback was presented for additional 1000 ms. Two CTIs of 200 and 1000 ms were employed during the whole experiment, with the duration of the CTI being evenly and pseudo-randomly distributed across all tasks. At the start of the second phase, participants were instructed to guess on every trial which task

54

T. Kleinsorge, J. Scheil / Acta Psychologica 159 (2015) 52–60

they were expected to really guess the next task. In addition, they were told that all tasks would occur with equal frequency and that switching probability was .5. The second phase consisted of seven blocks of 120 trials each. At the beginning of the third phase, participants were informed that, as in phase 1, no guessing was required. The third phase consisted of two blocks of 120 trials each. 2.2. Results

Fig. 1. Illustration of the response panel used in the experiment.

will have to be performed next. This was done by pressing the central key of the row assigned to this task. During the first 2500 ms of the RSI, a guessing request (‘Please guess now’) was presented on the screen. Within this time, participants had to press one of the guessing keys and to hold it down until the presentation of the digit/letter. This was done in order to make sure that participants did not revise their guessing during this interval. If the key was dropped earlier nevertheless, error feedback was presented and the trial was canceled. If the guessing key was not pressed until 2500 ms had elapsed, participants received feedback (‘Too slow! Please guess faster’) until one of the guessing keys was pressed, but the trial was not canceled. After the guessing interval of 2500 ms had elapsed, the trial continued as during the first phase: First, a fixation mark and after that, the precue was presented. Fixation mark and precue summed up to 1100 ms, the CTIs being the same as in the first phase. After that, the imperative stimulus was presented and had to be judged regarding the task that was indicated by the precue, irrespective of the task guessed by the participant. That is, participants were instructed that after they had placed their index finger on one of the central keys, the two outer keys of this row were the response keys for any task, meaning that correct responding was based solely on the ‘leftness’ or ‘rightness’ of their response. This way it was ascertained that the movement requirements for making a correct response did not differ for correctly and incorrectly guessed tasks. For example, if the magnitude task was guessed and the precue indicated the parity task, participants had to judge the parity of the digit but to respond with the keys labeled “b5” or “N5” in Fig. 1. Participants were explicitly told that the task sequence was completely random, that no regularity had to be ‘detected’ and that

The first block of the first and second phase was considered as practice and was excluded from analyses. Trials with RTs exceeding 2500 ms and trials following an error were discarded, as were error trials from the analysis of reaction times. Our main interest related to the observations of the second phase. However, before examining these data in detail, we checked inasmuch guessing the next task in the second phase was associated with a commitment to the guessed task by comparing the data from correctly guessed trials from that phase with the corresponding data from the first and the third phase. This was based on the rationale that commitment to the guessed task should reduce the effect of a prolongation of the CTI because preparation for the guessed task should initiate preparatory processes already in advance of the presentation of the precue. This was indeed the case, resulting in a significant interaction between Phase and CTI, F (2, 36) = 34.13, p b .001, MSe = 5011: Whereas a prolongation of the CTI reduced RT by 306 ms from 1280 ms to 974 ms in the first phase, and by 213 ms from 1023 ms to 810 ms in the third phase, the corresponding reduction amounted to only 117 ms (1123 vs. 1006 ms) in the second phase of the experiment. The net effect of CTI differed significantly among all phases, all ps b .001. The interaction between Phase and CTI was also significant in ER data, F (2, 36) = 8.28, p b .01, MSe = 15,552. While there was a significant CTI effect of 4.4% in the first phase, the CTI effects of the second (−0.8%) and third phase (1.1%) failed to reach statistical significance (p N .45 and p N .24, respectively). Coming to the main analyses of the data of the second phase, the number of trials with RTs exceeding 2500 ms amounted to 1.3%, and the number of trials discarded because of an error in the preceding trial was 6.4%. Mean individual RTs and error rates (ERs) were subjected to analyses of variance with the within-subjects factors CTI (200 ms vs. 1000 ms), Task Transition (repetition vs. switch), and Guessing Accuracy (correct vs. incorrect guess). Note that RT, as usual, is defined as the time between the onset of the imperative stimulus and the participant's response. For the RT data (cf. Table 1 and Fig. 2), all main effects turned out to be significant. Task switches went along with higher RTs (1181 ms) compared to task repetitions (1067 ms), F (1, 18) = 78.83, p b .001, MSe = 6656. Prolonging the CTI significantly reduced RTs from 1215 ms to 1030 ms, F (1, 18) = 102.26, p b .001, MSe = 12,655. Guessing the correct task was associated with faster RTs (1082 ms) compared to wrongly guessed tasks (1163 ms), F (1, 18) = 31.85, p b .001, MSe = 7750. Furthermore, all two-way interactions were significant. Most importantly, guessing the wrong task significantly Table 1 Experiment 1. Mean RT and ER as a function of Guessing Accuracy, Task Transition, and CTI. Standard errors of the means are presented in parentheses. Correct guess

CTI 200 ms CTI 1000 ms M CTI 200 ms CTI 1000 ms M

Incorrect guess

Task repetition

Task switch

Task repetition

Task switch

1056 (47) 958 (45) 1007 (64) 1.4 (0.5) 3.1 (0.5) 2.2 (0.5)

1233 (53) 1083 (55) 1158 (74) 4.9 (1.5) 4.7 (1.0) 4.8 (1.1)

1225 (48) 1016 (44) 1121 (63) 4.2 (0.8) 3.3 (0.7) 3.7 (0.9)

1346 (48) 1064 (48) 1205 (66) 5.6 (1.0) 3.7 (0.8) 4.6 (1.2)

M

1215 (94) 1030 (93) 4.0 (1.0) 3.7 (1.2)

T. Kleinsorge, J. Scheil / Acta Psychologica 159 (2015) 52–60

55

affected task repetitions more than task switches because it resulted in an additional requirement to change the row of keys, whereas such a change was required by task switches anyway. Therefore, we replicated Experiment 1 but separated the response sets for indicating the guessed task and responding to the imperative stimuli in a way that participants always used the same two keys when responding to the imperative stimulus. 3. Experiment 2 3.1. Method 3.1.1. Participants 20 right-handed subjects (3 male) with a mean age of 23.9 years (range: 19–29) participated. All had normal or corrected-to-normal vision.

Fig. 2. Experiment 1: Mean reaction time (RT) and error rate (ER) as a function of Task Transition and Guessing Accuracy. Error bars represent standard errors of the means.

reduced switch costs from 151 ms to 84 ms, resulting in a significant interaction between Task Transition and Guessing Accuracy, F (1, 18) = 11.14, p b .005, MSe = 3776. Furthermore, the interaction between CTI and Task Transition turned out to be significant (due to higher switch costs of 149 ms with short CTIs compared to 87 ms for long CTIs), F (1, 18) = 25.97, p b .001, MSe = 1436, as did the interaction between CTI and Guessing Accuracy, F (1, 18) = 22.95, p b .001, MSe = 6127. The benefit of correctly guessing the next task amounted to 141 ms for short CTIs, compared to 19 ms for long CTIs. The threeway interaction of all factors failed to reach significance, F (1, 18) = 0.50, p N .48. For the ER data, significant switch costs emerged, F (1, 18) = 9.14, p b .01, MSe = 12,826, due to higher ERs for task switches (4.7%) compared to task repetitions (3.0%). Furthermore, the interaction between Task Transition and Guessing Accuracy approached statistical significance, F (1, 18) = 4.15, p b .06, MSe = 6,074. Correctly guessing the next task resulted in higher ER switch costs (2.6%) compared to incorrect guesses (0.9%). 2.3. Discussion The main finding of the present experiment consists of a pronounced reduction of switch costs in trials in which participants made an incorrect prediction about the upcoming task. This observation is in line with our assumption that an incorrect prediction generates an error signal which results in an increased level of cognitive control. Specifically, incorrect guesses resulted in a pronounced increase of RT in case of a task repetition (169 ms with a short CTI, 58 ms with a long CTI). This compares to costs of 113 ms (short CTI) and − 19 ms (long CTI) in case of incorrectly guessed task switches. The interaction of Task Transition and Guessing Accuracy was not modulated by the duration of the CTI, with the ratio of the costs incurred by wrong guessing of task repetitions and switches being approximately 3:2 across CTIs. A somewhat peculiar feature of Experiment 1 consisted of the fact that in case of an incorrect guess, participants were required to respond by pressing keys that were associated with another task by instruction as well as by their vicinity to the central key used for indicating a guess of this other task. This feature was introduced to induce participants to (more or less literally) ‘stick to’ the guessed task, but may have resulted in an additional difficulty that predominantly affected wrongly guessed task repetitions, namely a loss of repetition priming on the motor level. That is, whereas incorrect guesses generally resulted in a requirement to use the ‘wrong’ response keys, this may have

3.1.2. Stimuli, tasks, and apparatus These were the same as in Experiment 1 except for the following differences. For guessing the next task, the central keys of the response device used in Experiment 1 had to be pressed with the left index finger. For the response to the imperative stimulus, participants had to use the left arrow key of a German QWERTZ keyboard for vowels, letters from the first half of the alphabet, even digits and digits smaller than five. The right arrow key was used for responding to consonants, letters from the second half of the alphabet, odd digits and digits larger than five. Responses to the imperative stimulus had to be executed with the right index finger. 3.1.3. Design and procedure These were the same as in Experiment 1. Due to the use of the same response set for all tasks, no additional movement had to be executed in case of a wrongly guessed task repetition. 3.2. Results The first block of the first and second phases was considered as practice and was excluded from analyses. Trials with RTs exceeding 2500 ms and trials following an error were discarded, as were error trials from the analysis of reaction times. As in Experiment 1, we checked in a first step whether guessing the next task resulted in enhanced commitment for the respective task. For this purpose, we compared correctly guessed trials with the corresponding data from the first and third phase. For RT data, the interaction between Phase and CTI was significant, F (2, 38) = 18.44, p b .001, MSe = 87,062. Whereas prolonging the CTI reduced mean RTs by 243 ms from 1216 ms to 973 ms in the first phase and by 216 ms from 930 ms to 714 ms in the third phase, the reduction amounted to only 120 ms (from 1038 ms to 918 ms) in the second phase. The net CTI effect of the second phase significantly (ps b .001) differed from that one in the first and third phase, whereas the difference between the latter two failed to reach significance (p N .20). For ER data, the corresponding interaction was not significant (p N .82), but the data pattern mirrored RT data (net CTI effects of 3.8%, 2.8%, and 3.4% for the three phases, respectively). In the second phase, the number of trials with RTs exceeding 2500 ms was 0.5%. 7.8% of the trials were discarded due to an error in the preceding trial. Mean individual RTs and error rates (ERs) were subjected to analyses of variance with the within-subjects factors CTI (200 ms vs. 1000 ms), Task Transition (repetition vs. switch), and Guessing Accuracy (correct vs. incorrect guess). In the analysis of RTs, the main effects of CTI, F (1, 19) = 92.28, MSe = 10,024, and of Task Transition, F (1, 19) = 81.50, MSe = 7817, turned out to be significant, both ps b .0001. A prolongation of the CTI from 200 to 1000 ms reduced RT from 1090 to 938 ms. A task switch increased RT to 1077 ms, compared to a mean RT of 951 ms with task

56

T. Kleinsorge, J. Scheil / Acta Psychologica 159 (2015) 52–60

repetitions. These switch costs were reduced from 157 to 96 ms when the CTI was prolonged from 200 to 1000 ms, F (1, 19) = 8.20, MSe = 4501, p b .01. More importantly, guessing an incorrect task reduced switch costs from 152 to 100 ms, F (1, 19) = 11.44, MSe = 2372, p b .01 (cf. Fig. 3). Furthermore, the CTI modulated the effect of Guessing Accuracy, with incorrect guesses being associated with a cost of 58 ms with a short CTI, but with a benefit of 25 ms when the CTI was long, F (1, 19) = 18.20, MSe = 3762, p b .01. Descriptively, this benefit was confined to task switches (cf. Table 2). The three-way interaction of all factors failed to reach significance, F (1, 19) = 0.14, p N .71. The corresponding analysis of ERs yielded only two main effects of CTI, F (1, 19) = 14.43, MSe = 18,359, p b .01, and Task Transition, F (1, 19) = 12.84, MSe = .18, p b .01. A prolongation of the CTI reduced mean ER from 7.3 to 4.9%. A task switch increased ER to 7.3%, compared to a mean ER of 4.9% with task repetitions. It should be noted, however, that incorrect guesses reduced ERs in switch trials on a descriptive level. 3.3. Discussion Experiment 2 replicated the main finding of Experiment 1 in terms of a reduction of switch costs by incorrect guessing: Whereas switch costs amounted to 152 ms in case of correct guesses, switch costs were reduced to 100 ms in case of incorrect guesses. These numbers correspond well to the 151 and 84 ms observed in Experiment 1. Amazingly, however, this neat replication was accompanied by rather different effects as a function of Guessing Accuracy and its interaction with CTI. Compared to Experiment 1, in Experiment 2 the cost of incorrect guesses amounted to 58 ms with the short CTI (compared to 141 ms in Experiment 1), and turned into a benefit of 25 ms with the long CTI (compared to a cost of 19 ms in Experiment 1). In a joint analysis of the RT data of both experiments, the only significant interaction involving the between-participants factor Experiment was the interaction with Guessing Accuracy, F (1, 37) = 12.44, MSe = 6374, p b .001. This difference between experiments was also evident in the error rates: Whereas in Experiment 1 incorrect guesses incurred a cost of 0.7%, incorrect guesses resulted in a benefit of 1.3% in Experiment 2, F (1, 37) = 3.71, MSe = .21, p b .07. According to our line of reasoning, the divergence across Experiments 1 and 2 of the effects of Guessing Accuracy per se on the one hand, and the similar interaction of Guessing Accuracy and Task Transition on the other hand, make sense. The effects of Guessing Accuracy per se relate to a certain source of conflict (misled preparation by incorrect guessing and preparation for the relevant task) and reflect

Fig. 3. Experiment 2: Mean reaction time (RT) and error rate (ER) as a function of Task Transition and Guessing Accuracy. Error bars represent standard errors of the means.

Table 2 Experiment 2. Mean RT and ER as a function of Guessing Accuracy, Task Transition, and CTI. Standard errors of the means are presented in parentheses. Correct guess

CTI 200 ms CTI 1000 ms M CTI 200 ms CTI 1000 ms M

Incorrect guess

Task repetition

Task switch

Task repetition

Task switch

971 (29) 887 (28) 929 (38) 6.2 (1.5) 4.2 (0.8) 5.2 (1.5)

1150 (34) 1013 (36) 1081 (41) 10.2 (1.5) 6.3 (1.5) 8.3 (2.0)

1051 (28) 893 (27) 972 (37) 5.5 (0.7) 3.5 (0.5) 4.5 (0.8)

1186 (29) 958 (25) 1072 (36) 7.2 (0.9) 5.5 (0.8) 6.3 (1.1)

M

1090 (53) 938 (53) 7.3 (1.9) 4.9 (1.4)

the adjustments needed to deal with this specific type of conflict, which in large part varied between the two experiments due to the variation of motor requirements in case of incorrect guesses. In contrast, the interaction of Guessing Accuracy and Task Transition relates to the establishment of a certain task set as the target of control, which is challenged by the previous execution of another task in addition to potentially misled preparation due to incorrect guessing. However, there are two points that may be considered as problematic with respect to these experiments. The first point concerns the observation that the interaction of Guessing Accuracy and Task Transition was most evident in task repetitions. Already a quick glance at Figs. 2 and 3 reveals that on a descriptive level, incorrect guessing affected task repetitions more than task switches. Therefore, locating the corresponding mechanism at task repetitions seems to suggest itself. In contrast, our account in terms of two counteracting effects on task switches, with incorrect guessing increasing RT but at the same time reducing switch costs by way of increasing the level of cognitive control, may seem unnecessarily complex. One alternative explanation rests on the assumption that an incorrect task prediction results in a general increase of RT, which in turn might be accompanied by more decay of the task set of the previous task. That is, incorrect guessing may have resulted in a disruption of repetition based facilitation, resulting in the observed increase of RT especially in repetition trials which consequently reduced switch costs. However, two aspects of our data seem to be hard to reconcile with this alternative account. First, if this would have been the case, the reduction of switch costs by an incorrect task prediction should have been less pronounced with a long CTI because a long CTI went along with smaller RTs than a short CTI. As can be seen from Tables 1 and 2, the switch cost reduction by an incorrect task prediction was more pronounced with a long CTI than with a short CTI, while an account in terms of decay of the previous task set would have predicted the opposite pattern. Although the corresponding three-way interaction was not significant, the pattern of results is inconsistent with an account in terms of a mere loss of repetition-based task activation. Second, a comparison of results across the two experiments suggests a dissociation between the main effect of incorrect guessing and the effect of incorrect guesses on switch costs that was due to the variation of response requirements. If, however, the reduction of switch costs by incorrect guessing was merely a byproduct of a disruption of processing brought about by an incorrect guess, we would have expected that differences between the effects of incorrect guessing would result in corresponding differences with regard to the effects that incorrect guessing has on switch costs. Another objection that may be raised against our interpretation of the data is that in case of an incorrectly guessed task switch, actual task repetition may be ‘functional’ task switches because participants had to switch back from the incorrectly guessed task to the preceding task. Somewhat similarly, one might argue that an incorrect task prediction in case of a task repetition might have led participants to inhibit the previous task set, resulting in a selective slowing of task repetitions due to an incorrectly predicted task switch. However, also in this case a stronger reduction of switch costs by an incorrect task prediction should have been observed in conditions with a short CTI because in this

T. Kleinsorge, J. Scheil / Acta Psychologica 159 (2015) 52–60

condition, participants would have been expected to spend much more time inhibiting the old task set compared to conditions with a long CTI. However, as already mentioned, the reduction of switch costs was more pronounced with a long CTI at least on a descriptive level. Nevertheless, it would be desirable to remove the ambiguities that result from possibly different functional roles of incorrect predictions in cases of task switches and repetitions on an empirical basis. Such an attempt should also avoid a second possible shortcoming of the first two experiments. This second point, which may be intimately related to the first one, is given by design and relates to the unequal probabilities of (in)correctly guessing task switches and repetitions, respectively. Because the probability of switches and repetitions amounted to .5 each, but there were three possible types of switch but only one possible type of repetition, correctly guessed repetitions were much more frequent than correctly guessed switches (cf. Appendix A). Regarding the (relative) rareness of correct guesses, the first thing one should note is that correct guesses generally went along with shorter RTs, arguing against accounts suggesting that infrequent events generally slow performance (e.g., Notebaert et al., 2009). However, the reduction of RT by correct guesses was more pronounced with task repetitions compared with task switches, which may in part have been due to correctly guessed switches being even more rare events than correctly guessed repetitions. These observations suggest that guessing-induced expectations may have affected performance on two different levels, one associated with preparation for the guessed task, the other one being due to the relative frequencies of correctly guessed repetitions and switches, respectively (cf. Umbach, Schwager, Frensch, & Gaschler, 2012, for a similar argument). In order to improve our empirical approach on both of these points, we decided to carry out a third experiment in which participants did not guess the identity of the next task but were instead instructed to predict the position (left or right) at which the next precue would be presented. Because the precues were presented equally often to the left and to the right, expected guessing accuracy amounted to .5 for each alternative. Therefore, the correctness of guessing varied fully independent of certain task transitions. In order to prevent participants from choosing each guessing alternative rather randomly without paying attention to their choices, a small number of catch trials was introduced. In these catch trials, participants were asked to report the position of the last precue instead of performing one of the other tasks that were the same as in the previous two experiments. Furthermore, attention to the guessed alternative was reinforced by presenting an empty frame around the guessed precue position. (In pilot testing, it turned out that without these provisions, a large number of participants made their guessing responses in a stereotypic manner without relating this activity to the main task.) While removing the imbalances with respect to the relative frequencies of correct and incorrect guesses in general, and their distribution across task switches and repetitions in particular, asking participants to predict the precue instead of the tasks introduces a significant change in terms of the source of conflict. Nevertheless, because of the strong associations between tasks and their cues when there is a 1:1 cue-totask mapping (cf. Kleinsorge, 2012), we expected that a conflict induced by guessing the wrong precue position would still converge upon the same target of cognitive control, namely the task set needed to perform the precued task.

4. Experiment 3 4.1. Method 4.1.1. Participants 19 right-handed subjects (5 male) with normal or corrected-tonormal vision participated. Their mean age was 23.6 years (range: 18–30).

57

4.1.2. Stimuli, tasks, and apparatus The four tasks and imperative stimuli were the same as in Experiments 1 and 2. In addition, the same four precues were used, but they were presented either to the right or to the left of the imperative stimulus (with a horizontal displacement of about 4.5 cm, measured from the center of the display to the center of the precue). Precue positions were chosen pseudo-randomly, with a probability of 50% each. For guessing the precue position, participants had to press the “y”-key of the standard QWERTZ keyboard for the left and the “x”key for the right position. After the guessing response was made, a square surrounding the guessed cue position was presented until the onset of the imperative stimulus. In case of correctly guessing the precue position, the precue was subsequently presented within this square, otherwise it was presented on the side opposite to the square. Responses to the imperative stimulus were made with the “.”-key for vowels, letters from the first half of the alphabet, even digits and digits smaller than five and the “-“-key for responding to consonants, letters from the second half of the alphabet, odd digits and digits larger than five. Responses to the imperative stimulus had to be executed with the right index finger. In 5% of all guessing trials, the imperative stimulus was replaced by “# #”. In these trials, serving as catch trials, participants had to indicate with their response the position of the precue (left or right, irrespective of the position guessed). 4.1.3. Design and procedure These were the same as in Experiments 1 and 2. 4.2. Results The first block of the first and second phases was considered a practice and therefore excluded from analyses. Trials with RTs exceeding 2500 ms were also discarded, as were trials following an error. As in Experiments 1 and 2, we compared in a first step correctly guessed trials with trials from the first and third phases to detect differences of the CTI effect. For RT data, the interaction between Phase and CTI was significant, F (2, 36) = 4.23, p b .05, MSe = 1853. However, the data pattern differed from the results of Experiment 1 and 2. While the CTI effect of the third phase (218 ms) was significantly smaller compared to the first phase (274 ms, p b .01), the CTI effect of the second (guessing) phase, amounting to 239 ms, differed only marginally (p b .08) from the CTI effect of the first phase and not (p N .28) from that of the third phase. The respective interaction was not significant for ER data (p N .55). Surprisingly, ERs showed the opposite data pattern on a descriptive level, the CTI effect increasing from the first to the third phase (− 0.1%, 0.9%, and 1.3% for the three phases, respectively). The observation that, in contrast to Experiments 1 and 2, the CTI effect was not reduced specifically in the second phase makes sense because, in contrast to guessing an upcoming task, predicting the position of the precue provides no basis for preparing a specific task already in advance of the onset of the precue. For the second phase, 1.1% of all trials were excluded due to RTs exceeding 2500 ms. The number of post error trials amounted to 7.0%. For RT data, all main effects turned out to be significant. A prolongation of the CTI reduced mean RTs from 1211 ms to 966 ms, F (1, 18) = 436.85, p b .0001, MSe = 5241. Task switches went along with higher RTs (1158 ms) compared to task repetitions (1019 ms), F (1, 18) = 94.54, p b .0001, MSe = 7775. Wrongly guessing the precue position resulted in faster reactions (1075 ms) compared to correct guesses (1102 ms), F (1, 18) = 15.74, p b .001, MSe = 1670. Furthermore, the interaction between CTI and Task Transition was significant, F (1, 18) = 39.73, p b .0001, MSe = 5405. Switch costs amounted to 64 ms for the long and 214 ms for the short CTI. The two-way interaction between Task Transition and Guessing Accuracy failed to reach significance, p N .39. Instead, the three-way interaction of all factors turned out to be significant, F (1, 18) = 16.22, p b .001, MSe = 1214

58

T. Kleinsorge, J. Scheil / Acta Psychologica 159 (2015) 52–60

(cf. Fig. 4). When the CTI was long, task repetitions were not affected by Guessing Accuracy (p N .74) whereas task switches were significantly (p b .01) speeded up by wrong guesses. When the CTI was short, task switches were left unaffected by Guessing Accuracy (p N .77) but task repetitions were significantly faster when the correct precue position was guessed (p b .001). For ER data, a significant main effect of CTI occurred, F (1, 18) = 5.07, p b .05, MSe = 7,329, due to higher ERs for short (4.9%) compared to long CTIs (3.9%). In addition, the main effect of Task Transition was significant, F (1, 18) = 11.08, p b .01, MSe = 7,668. Task switches went along with higher ERs (5.1%) compared to task repetitions (3.6%). 4.3. Discussion In this experiment, we observed the predicted interaction of Task Transition and Guessing Accuracy with a CTI of 1000 ms. In contrast to the experiments reported before, this time the reduction of switch costs by incorrect guessing was clearly due to a decrease of RT in the switch trials. In contrast, with a CTI of 200 ms, a completely different picture emerged. In this condition, incorrect predictions had barely any effect on task switches but reduced RT with task repetitions, resulting in increased switch costs in incorrectly guessed trials. The latter pattern was not observed in any other condition of the whole series of experiments reported in this article. We are reluctant to interpret this completely unexpected finding but note that the CTI of 200 ms may have induced a relatively long-lasting inhibition of return by attenuating the processing of the precue within the surrounding square as a consequence of correctly guessing the position of the precue. The latency of inhibition of return is known to increase with increasing task difficulty (cf. Klein, 2000) and may last well over a second with the relatively difficult task of our experiments. Such an attenuation of precue processing is unlikely to occur with a precue duration of 1000 ms. However, this account can only explain the occurrence of a cost of correctly predicting the precue position but not why this cost only accrued in task repetition trials. However, the most important point in our view is that we

Fig. 4. Experiment 3: Mean reaction time (RT) and error rate (ER) as a function of Task Transition, Guessing Accuracy, and CTI. Error bars represent standard errors of the means.

observed a speed-up of task repetitions by incorrect guesses in a condition in which we did not observe a reduction of switch costs, but we observed a reduction of switch costs by incorrect guesses in a condition in which this was clearly attributable to the switch trials only. In Experiment 3, wrong guesses went along with faster responses compared to correct guesses overall. This may seem surprising, especially because the probability of a correct guess amounted to .5 which precludes any account in terms of relative frequency. In our view, this observation adds credibility to our main conjecture, namely that incorrect predictions result in more controlled processing being instigated by an error signal. In this context it should be noted that in a task switching experiment, also repetition trials can be expected to be subject to a considerable degree of cognitive control (e.g., in terms of ‘task shielding’, cf. Dreisbach, 2012). Apart from yielding additional evidence for a transient increase of cognitive control induced by incorrect guessing, the observations of Experiment 3 suggest some generalization with respect to the relation between sources of conflict and targets of control. Whereas in Experiments 1 and 2 additional conflict was induced by guessing the to-be-performed task itself, in Experiment 3 guessing related merely to the position at which information regarding the relevant task was presented. Nevertheless, encoding this information was part of the process of preparing for the upcoming task, thus converging upon the same target of cognitive control. 5. General discussion The present study aimed at combining two sources of conflict that converge upon the same target of cognitive control. The first source of conflict was the one inherent in any task switching situation, namely the conflict between a task set activated by the recent performance of another task and the task set needed to perform the actually relevant task (cf. Vandierendonck et al., 2010). The second source of conflict was induced by requiring participants to guess aspects of the upcoming task. In case of an incorrect guess, a conflict should accrue between the representation of the guessed task (or precue position) and the actually relevant task (or precue position). Our main hypothesis was that adding the second source of conflict to the first one would increase the amount of controlled processing devoted to the establishment of the actually relevant task set as the common target of control for both types of conflict, leading us to predict a reduction of switch costs in case of incorrect guesses. This prediction was borne out by the data: In all three experiments, incorrectly guessing about the next task was associated with reduced switch costs. Experiment 3 provided evidence that the effect of incorrect guessing on task switching performance did not depend on specific preparation for a particular task. As corroborated by different net effects of CTI across the three phases of the experiments, guessing had different effects in terms of task preparation in the first two versus the third experiment. In particular, whereas guessing a particular task was associated with specific preparation for this task in the first two experiments, guessing the position of the precue was not accompanied by a reduced effect of the duration of the CTI in the third experiment. Thus, we assume that the reduction of switch costs by incorrect guessing was generally independent of prior preparation for a task switch. We also assume that the reduction of switch costs by incorrect guessing was not dependent on a disruption of repetition based facilitation because Experiment 3 provided evidence for such an effect despite the fact that guessing the position of the task cue should not affect task switches and repetitions in a systematically different manner. Furthermore, in Experiment 3 there was no indication of a selective slowdown of incorrectly guessed task repetitions. While all three experiments also yielded some inconsistent observations, they all provided converging evidence pertaining to our main point of conflict-induced enhancements of cognitive control within a single trial. Therefore, although much more work has to be done in order to disclose the specific

T. Kleinsorge, J. Scheil / Acta Psychologica 159 (2015) 52–60

conditions under which two conflicts combine in a way as to increase subsequent performance, the present series of experiments reveals a promising route of experimentally teasing those conditions apart. The present study accords well with a recent rise in interest regarding the effects of expectation on performance. Whereas there can be no doubt that correct expectations facilitate performance, an expectation mismatch is generally assumed to impede performance (cf. Gaschler, Schwager, Umbach, Frensch, & Schubert, 2014, for a review). However, the present study suggests that even incorrect expectations may have adaptive value, compared to adopting a passive stance of not expecting anything particular. That is, the present observations suggest that an error of prediction, even if the prediction was based on pure guessing, induces an adaptation that shifts the balance between controlled and more automatic processing in a way capable of overcoming effects of residual task activation. Thus, the mere attempt of exerting some control over an in principle uncontrollable situation seems to be capable of increasing the level of control if the initial attempt to exert control fails. In contrast to adopting a passive stance of ‘waiting what happens’, actively predicting what happens seems to put the cognitive system in a position that (possibly quite automatically) triggers further adaptations when the prediction turns out to be wrong. In this respect, one may ask whether the adoption of a passive stance is what happens normally. In cognitive psychology, there is a strong tradition of conceptualizing behavior as basically stimulus-driven, with expectations or predictions being merely some possibly useful but dispensable add-on. However, this position has become under increasing pressure during recent years both within cognitive psychology (cf. Kunde, Elsner, & Kiesel, 2007; Marien, Aarts, & Custers, 2013) as well as on the basis of neurophysiological evidence suggesting an integral role for predictions in the control of behavior in general (cf. Schiffer, Waszak, & Yeung, 2015, for a recent review). The question arises whether asking participants to engage in certain predictions merely constraints and directs processes which proceed anyway but in a rather unsystematic manner, at least under conditions that are designed to allow for no valid predictions. Such conditions are usually implemented in our experiments but are rarely encountered in the real world.

59

The research reported in this article was supported by grant KL 1205/6-2 of the Deutsche Forschungsgemeinschaft. We thank Rainer Flöring and David Bauer for their technical assistance, and Wout Duthoo for sharing his data with us.

of guessing the correct task amounts to only about 8.3%, because even if a task switch was correctly guessed, there is only a chance of 33.3% to also guess the correct task. Because the guessing rates are only approximately under experimental control due to subjects choosing freely between the tasks, deviations between the expected proportions of correct and wrong guesses and the actual types of guesses presented in Table A1 occur. In Experiment 1, participants guessed the magnitude task in 23.5% of all trials, the parity task in 27.0%. The consonant/vowel task was guessed in 27.1% of all trials, the alphabet task in 22.4%. In Experiment 2, participants guessed the magnitude task in 21.8% of all trials, the parity task in 26.6%, the vowel/consonant task in 28.8% and the alphabet task in 22.7% of all trials. In a second step, we focused on different types of wrong guesses. When switching among four tasks, different kinds of wrong guesses may emerge. Participants may guess a task repetition in trials when the task actually switches. In addition, the opposite can be the case: Participants may guess a task switch but have to perform a task repetition. Finally, a task switch may be guessed, but a different task switch (i.e., a switch to a different task than that one guessed by the participant) may be required. Therefore, additional analyses with the within-subjects factors CTI (200 vs. 1000 ms) and Guessing Accuracy (correctly guessed repetition, correctly guessed switch, guessed repetition in case of a switch, guessed switch in case of a repetition, guessed switch in case of a different switch) were run. Two comparisons may be of special interest in this respect. First, one may assume that actual task repetitions that were guessed to be task switches can be considered as ‘functional task switches’ because participants already replaced the previous task set by the one appropriate for the guessed task. Thus, a comparison of task repetitions associated with incorrect guesses with correctly guessed task repetitions on the one hand and actual task switches that were incorrectly guessed to be task switches to another task on the other hand may be instructive. The latter comparison may provide an estimate of residual task activation despite some preparation for another, actually irrelevant task. As a complement, the former comparison may provide a measure of loss of residual activation by way of preparing for an irrelevant task. The second potentially informative comparison is between task switches that were guessed to be task repetitions and task switches that were correctly guessed to task switches albeit to another, actually wrong task. This comparison may provide a measure of a strengthening of residual task activation due to incorrectly assuming that a task repetition would occur.

Appendix A

Experiment 1

In supplementary analyses, we took in a first step a closer look on the different types of guesses (cf. Table A1 for proportions of different types of guesses for all experiments). Please note that for experiments 1 and 2, the expected proportion of correct guesses is about 33.3%. Task repetitions occurred in 50% of all trials, and participants were informed about this. Therefore, it can be assumed that repetitions were guessed in 50% of all trials, resulting in more or less 25% of correctly guessed task repetitions. For task switches, on the other hand, the probability

For RT data, a significant main effect of Guessing Accuracy occurred, F (4, 72) = 42.05, p b .001, MSe = 5509. With 1007 ms, correctly guessed task repetitions led to fastest RTs that differed significantly (p b .001, Bonferroni adjusted) from all other conditions. Actual task repetitions in which a task switch was guessed resulted in slower reactions (1101 ms) that also differed from all other conditions (p b .05). This observation suggests that preparing for a currently irrelevant task results in a considerable decrease but not a complete

Acknowledgments

Table A1 Mean percentages of correct and wrong guesses as a function of Task Transition and CTI for all three experiments. Experiment

1 2 3

Task repetition

Correct guesses Wrong guesses Correct guesses Wrong guesses Correct guesses Wrong guesses

Σ

Task switch

CTI 200 ms

CTI 1000 ms

CTI 200 ms

CTI 1000 ms

11.5% 12.2% 10.0% 13.3% 12.8% 12.6%

13.0% 13.5% 12.4% 14.4% 12.2% 12.3%

4.6% 20.8% 4.9% 21.0% 12.7% 12.3%

4.2% 20.3% 4.5% 19.6% 12.9% 12.1%

33.3% 66.8% 31.7% 68.3% 50.7% 49.3%

60

T. Kleinsorge, J. Scheil / Acta Psychologica 159 (2015) 52–60

loss of residual task activation. No significant differences occurred between correctly guessed task switches (1158 ms), task switches with a guessed task repetition (1181 ms), and task switches in which a switch to a different task was guessed (1201 ms, all ps N .12). Thus, incorrectly guessing that the task would repeat seems not to be associated with an additional strengthening of the task set carried over from the preceding trial. Besides the main effect, a significant interaction between Guessing Accuracy and CTI occurred, F (4, 72) = 15.93, p b .001, MSe = 3609. This interaction was mainly caused by the fact that for long CTI, no significant differences occurred for the different kinds of switch trials, whereas for the short CTI, RTs for correctly guessed switch trials were significantly faster compared to switches in which a repetition was guessed (p b .01) and switches in which a switch to another task was guessed (p b .001). Therefore, these results corroborate the interpretation of the main effect of Guessing accuracy. The corresponding ER analysis yielded only a significant main effect of Guessing Accuracy, F (4, 72) = 4.32, p b .01, MSe = 10,891. Correctly guessed task repetitions led to smallest ERs (2.2%) that differed significantly (p b .05) from correctly guessed task switches (4.8%). Furthermore, correctly guessed repetitions differed (p b .01) from task switches in which another task switch was guessed (5.0%). For guessed task switches in which a repetition had to be performed on (3.7%) as well as for guessed repetitions in which a switch occurred (4.4%), none of the pair-wise comparisons reached statistical significance. Experiment 2 For RT data, a significant main effect of Guessing Accuracy occurred, F (4, 76) = 35.96, p b .0001, MSe = 5170. Correctly guessed task repetitions led to fastest RTs (931 ms) that differed significantly (ps b .001, Bonferroni adjusted) from all types of actual task switches and marginally significant (p b .08) from task repetitions in which a task switch was guessed (976 ms), which also differed significantly (ps b .001) from all task switch conditions. No difference occurred between correctly guessed task switches (1081 ms), switches in which a repetition was guessed (1070 ms), and switches in which a switch to another task was guessed (1070 ms, all ps N .99). In addition, there was a significant interaction between Guessing Accuracy and CTI, F (4, 76) = 8.90, p b .001, MSe = 4006. For both CTIs, the three different types of task switches did not differ from each other. The interaction was caused by the fact that for long CTIs, correctly and incorrectly guessed task repetitions did not differ from each other, whereas for the short CTI, correctly guessed repetitions went along with faster RTs than repetitions in which a switch was guessed (p b .01). Thus, this

analysis corroborates the corresponding observations of Experiment 1 by providing some evidence for a loss of residual task activation by incorrectly guessing a task switch which, however, is far from complete as actual task repetitions were still performed considerably faster than any kind of task switch irrespective of guessing accuracy. Furthermore, the 43 ms difference between correctly and incorrectly guessed task repetitions was significantly smaller than the corresponding difference of 114 ms observed in Experiment 1, t (37) = 2.40, p b .05. The corresponding analysis of ER data yielded a significant main effect of Guessing Accuracy, F (4, 76) = 4.36, p b .01, MSe = 20,030. Correctly guessed task switches led to significantly higher ERs (8.3%) compared correctly guessed task repetitions (5.2%, p b .05) and actual task repetitions in which a task switch was guessed (4.5%, p b .01). None of the other pairwise comparisons reached statistical significance (all ps N .10). References Botvinick, M.M., Braver, T.S., Barch, D.M., Carter, C.S., & Cohen, J.D. (2001). Conflict monitoring and cognitive control. Psychological Review, 108, 624–652. Dreisbach, G. (2012). Mechanisms of cognitive control the functional role of task rules. Current directions in psychological science, 21, 227–231. Duthoo, W., Abrahamse, E.L., Braem, S., Boehler, C.N., & Notebaert, W. (2014). The heterogeneous world of congruency sequence effects: an update. Frontiers in Psychology, 5 (Article 1001). Duthoo, W., De Baene, W., Wühr, P., & Notebaert, W. (2012). When predictions take control: the effect of task predictions on task switching performance. Frontiers in Psychology, 3 (Article 282). Egner, T. (2014). Creatures of habit (and control): a multi-level learning perspective on the modulation of congruency effects. Frontiers in Psychology, 5 (Article 1247). Gaschler, R., Schwager, S., Umbach, V.J., Frensch, P.A., & Schubert, T. (2014). Expectation mismatch: differences between self-generated and cue-induced expectations. Neuroscience & Biobehavioral Reviews, 46, 139–157. Kim, S., & Cho, Y.S. (2014). Congruency sequence effect without feature integration and contingency learning. Acta Psychologica, 149, 60–68. Klein, R.M. (2000). Inhibition of return. Trends in Cognitive Sciences, 4, 138–147. Kleinsorge, T. (2012). Task switching with a 2: 1 cue-to-task mapping: separating cue disambiguation from task-rule retrieval. Psychological Research, 76, 329–335. Kunde, W., Elsner, K., & Kiesel, A. (2007). No anticipation–no action: the role of anticipation in action and perception. Cognitive Processing, 8, 71–78. Marien, H., Aarts, H., & Custers, R. (2013). Adaptive control of human action: the role of outcome representations and reward signals. Frontiers in Psychology, 4 (Article 602). Notebaert, W., Houtman, F., Van Opstal, F., Gevers, W., Fias, W., & Verguts, T. (2009). Posterror slowing: an orienting account. Cognition, 111, 275–279. Schiffer, A.M., Waszak, F., & Yeung, N. (2015). The role of prediction and outcomes in adaptive cognitive control. Journal of Physiology-Paris, 109, 38–52. Umbach, V.J., Schwager, S., Frensch, P.A., & Gaschler, R. (2012). Does explicit expectation really affect preparation? Frontiers in Psychology, 3 (Article 378). Vandierendonck, A., Liefooghe, B., & Verbruggen, F. (2010). Task switching: interplay of reconfiguration and interference control. Psychological Bulletin, 136, 601–626.