It has been consistently found that increasing the time available to prepare in anticipation of a switch in task reduces RT switch cost (e.g., Goschke 2000; Meiran 1996; Rogers and Monsell 1995). This finding has been interpreted as evidence for a cognitive control process of task-set reconfiguration that can be activated prior to the onset of an imperative stimulus, leading to a reduction in the RT switch cost (Rogers and Monsell). According to De Jong (2000), task-set reconfiguration is an all-or-none process that can be activated before or after stimulus onset depending on task and subject parameters (see Section 1.2 below). While more recent studies have questioned the notion of an ‘all-or-none’ task-set reconfiguration process (e.g., Brown et al. 2006; Lien et al. 2005), failures to engage in anticipatory task-set reconfiguration on some proportion of trials are now recognized to at least partially account for the residual switch cost that remains even at long preparation intervals (Rogers and Monsell).#

In contrast to task-set reconfiguration accounts of RT switch cost, Allport and colleagues (Allport et al. 1994; Allport and Wylie 2000; Wylie and Allport 2000; Waszak et al. 2003) argue that the switch cost results from differences in task-set inertia or performance readiness between switch and repeat trials. Performance readiness is affected by factors such as positive and negative priming from previous task practice, cued activation of relevant/irrelevant stimulus, response and/or task-set attributes and inhibition of incongruent attributes. Meiran and colleagues (Meiran 1996 2000; Meiran et al. 2000; see also Goschke, 2000) offered a conciliatory position showing evidence for both active task-set reconfiguration and passive task-set inertia effects on RT switch cost. Independent manipulation of the opportunity for anticipatory task-set reconfiguration in the cue–stimulus interval (CSI) and the time available for passive dissipation of activation across the overall response-stimulus interval (RSI) both resulted in significant reductions in RT switch cost (Meiran 1996; Meiran et al. 2000; Nicholson et al. 2005).#

Although there is considerable evidence regarding the existence of a process of anticipatory task-set reconfiguration (but see Koch 2005; Ruthruff et al. 2001; Sohn and Carlson 2000), relatively little is known about what this process actually entails. For task-set reconfiguration to be activated prior to stimulus onset, it is necessary for participants to have foreknowledge that the next trial will require a switch in task, as is the case with alternating runs paradigms that use a predictable task sequence (e.g., task AABB) and with task cueing paradigms that validly cue the relevant task prior to stimulus onset. Foreknowledge is only useful if the length of the preparation interval (e.g., the RSI in alternating runs paradigms or the CSI in cued task paradigms) is sufficient to allow activation of anticipatory task-set reconfiguration (Rogers and Monsell, 1995) and the participant actively engages this process prior to stimulus onset (De Jong, 2000). If these conditions are not met, task-set reconfiguration is initiated and/or completed after stimulus onset, resulting in longer switch trial RT and hence increased RT switch cost.#

Anticipatory task-set reconfiguration can also be affected by parameters that promote or impede active engagement in these processes. Goschke (2000) found that interrupting anticipatory task-set reconfiguration by requiring participants to recite irrelevant verbal material during a long RSI (1200 ms) resulted in RT switch cost equivalent to the no preparation or short RSI (150 ms) condition. In contrast, when participants verbalized the task to be completed on the next trial, there was a significant reduction in RT switch cost compared to either of the above conditions and equivalent to that obtained with a long RSI when no verbalization is required. This suggests that retrieval and application of the new task-set from long term into working memory are important parts of anticipatory task-set reconfiguration (Mayr and Kliegl, 2000).#

Changing switch trial salience by reducing the proportion of switch relative to repeat trials has also been shown to increase the engagement of active task-set reconfiguration processes (Monsell and Mizon, 2006). Using cues that manipulate subjective expectancy of equiprobable switch and repeat trials, Dreisbach et al. (2002) found equivalent effects of subjective expectancy on both switch and repeat trials, suggesting that task preparation may occur in anticipation of both switch and repeat trials. Furthermore, task preparation did not interact with task foreknowledge (see also Sohn and Carlson, 2000), suggesting that these manipulations affect independent processes and supporting a preparation readiness account of switch cost. Alternatively, Monsell and Mizon (2006) argue that these effects may be at least partly attributed to cue complexity as complex cues may inadvertently introduce an additional task, resulting in task-set reconfiguration for both switch and repeat trials.#

Inhibitory processes may also affect task-set reconfiguration. Mayr and Keele (2000) found that switching back to a recently abandoned task-set resulted in larger RT switch cost as compared to switching to a third task—a phenomenon they refer to as backward inhibition. Specifically, RT was larger on the third trial of a CBC task sequence as compared to an ABC sequence (see also Arbuthnott and Frank 2000; Koch et al. 2004). This suggests that the previously relevant, but now irrelevant, task-set (task C) was inhibited in the former task sequence, thereby resulting in longer RT when this task-set was reactivated within a short period. Using a go/no-go paradigm, Schuch and Koch (2003) found that backward inhibition did not occur following no-go trials that required task-set preparation but no response execution, suggesting that selection or execution of the response triggers the inhibitory processes. Mayr and Keele (2000) found that the amount of backward inhibition was not reduced with increasing CSI, suggesting that the inhibition process is not affected by the amount of time available for anticipatory task-set reconfiguration. In contrast, backward inhibition was reduced at longer response–cue intervals (RCI) that facilitated greater passive dissipation of task-set interference (see also Koch et al., 2004). Furthermore, Gade, 2005 showed that the opportunity for passive dissipation of interference across the interval between trial n and trial n−2 had a greater effect than the interval between n and n−1 (e.g., in task sequence ABA, the amount of backward inhibition was affected more by the interval between the two task A trials than between task B and task A trials). Dreisbach et al. (2002) reported that ‘semi-specific’ cues (i.e., cues that signal an impending switch trial without identifying which specific task to prepare) resulted in larger RT switch cost than ‘specific’ cues that indicated which task to switch to. Unlike specific cues, semi-specific cues did not produce subjective expectancy effects, suggesting that knowledge that the task would change without specification of which task would be performed did not result in any differential response benefit. Hubner et al. (2003) showed that backward inhibition effects were comparable for an uncued switch in task and a cued switch in task with semi-specific cues.#

Taken together, these findings suggest that task-set reconfiguration may involve not only activation of the currently active task-set but also inhibition of the previously active task-set. Furthermore, it appears as if the inhibition process cannot be initiated independently of activation of the currently active task-set. Mayr and Keele (2000) argue that inhibition may reflect a low-level control process, such as lateral inhibition of competing action schemas. Dreisbach et al. (2002) suggest that active engagement of the cued task-set triggers automatic inhibition of the previously relevant task-set when the response is executed. That is, they argue that inhibition of the irrelevant task-set is not under endogenous control and cannot occur independently of activation of the relevant task-set (see also Hubner et al., 2003).#

Overall, there are still considerable gaps in our knowledge regarding the number and type of cognitive processes that underlie anticipatory task-set reconfiguration. Although there is evidence of a role for an inhibitory process, its timing and degree of dependence on other activation processes remain unclear. Furthermore, the precise set of operations that constitute activation of the relevant task-set in anticipation of a switch in task remains to be defined.#

Event-related brain potentials (ERPs) in task-switching experiments can help elucidate the cognitive processes that lead up to behavioral differences between switch and repeat trials in general, as well as those processes more specifically involved in task-set reconfiguration. Most ERP task-switching studies have focused on differences between ERPs to switch and repeat stimuli occurring after stimulus onset (e.g., Barcelo et al. 2000 2002; Gehring et al. 2003; Hsieh and Yu 2003; Hsieh and Liu 2005; Poulsen et al. 2005; Swainson et al. 2003). ERP differences between switch and repeat trials occurring after stimulus onset may reflect processing differences occurring as a result of differential levels of proactive interference for switch compared to repeat stimuli, differential level of activation of the relevant task-set at stimulus onset, as well as differential stimulus-response interference elicited by the stimulus itself. Therefore, although anticipatory task-set reconfiguration may indirectly affect stimulus-locked ERPs (e.g., Barcelo et al., 2000), it is difficult to isolate this effect from that of other passive interference or stimulus-elicited processes.#

It is possible to isolate ERP effects associated with anticipatory task-set reconfiguration by examining either the interval preceding stimulus onset in an alternating runs paradigm or the interval following cue onset in a cued task-switching paradigm. Using Rogers and Monsell's (1995) predictable task-switching paradigm, Karayanidis et al. (2003) identified an increased positivity for switch as compared to repeat trials in the interval between the response to the previous trial and the onset of the next trial (i.e., during the RSI). This differential positivity for switch as compared to repeat waveforms (labeled D-Pos¹1The term D-Pos is used as a convenience label to refer to the differential switch minus repeat positivity without necessarily implying that the label represents a specific or new ERP component that is not evident in the original waveforms.) was largest parietally and peaked around 400 ms after the previous response. At short preparation intervals (150 or 300 ms), D-Pos began prior to but did not peak until after stimulus onset, whereas with long preparation intervals (600 and 1200 ms), D-Pos peaked and returned to baseline before the stimulus was presented. A similar switch-related positivity has been reported in other studies. Wylie et al. (2003) found that trials preceding a predictable switch trial were associated with a larger late positivity over posterior scalp electrodes compared to trials preceding a repeat trial. Other studies found that ERP waveforms time-locked to cue presentation show a larger positivity for switch compared to repeat cues over parietal sites around 500 ms after cue onset (Miniussi et al. 2005; Rushworth et al. 2002 2005). Using a cued task-switching paradigm, Nicholson et al. (2005) found that the morphology of the parietal switch positivity was modified by manipulation of the CSI (150 to 600 ms) but not the RSI (750 to 1200 ms), indicating that it is affected by opportunity to engage in active task-set reconfiguration but not by the passive passage of time between the previous and the current trial.#

In summary, anticipatory task-set reconfiguration is associated with increased positivity in anticipation of switch relative to repeat trials, particularly over parietal electrodes. At longer preparation intervals, this differential activity for switch trials can be completed prior to stimulus presentation and is associated with reduced RT switch cost. With short preparation intervals, which provide little or no opportunity for anticipatory task-set reconfiguration, these processes occur after stimulus onset and are associated with larger RT switch cost.#

The current study investigates one component process of task-set reconfiguration, the activation of the currently relevant task-set. Previous ERP studies have identified a differential positivity for switch compared to repeat trials in the interval preceding stimulus presentation that is believed to reflect anticipatory task-set reconfiguration processes (Karayanidis et al. 2003; Nicholson et al. 2005). The current study examined whether this anticipatory positivity is specifically associated with activation of the relevant task-set or is affected by other processes that may occur during preparation for a switch in task.#

A cued-trials task-switching paradigm was used. Participants randomly alternated between three tasks defined for the same stimulus set. Three types of cues were used to signal the requirements of the next trial. The first cue type signaled task repetition, the second cue type signaled a task switch and specified which task to switch to (switch-to or specific cues), whereas the third cue type signaled a task switch but did not specify which of the two alternate tasks would be active (switch-away or semi-specific cues). With the latter cues, participants knew that they would not be repeating the same task as on the previous trial but had no information about which task-set would be relevant and had to await stimulus onset before the new task-set could be activated. Therefore, the use of switch-to and switch-away cues was designed to isolate processes associated with activation of the relevant task-set.#

Switch-to trials were identical to the switch trials defined in previous ERP and behavioral studies. Therefore, switch-to trials were expected to produce a large RT switch cost that reduced with increasing CSI and with increasing RSI, as well as a large switch-related positivity time-locked to cue onset and peaking before stimulus onset in the long CSI condition (Karayanidis et al. 2003; Nicholson et al. 2005). Semi-specific or switch-away cues index that the current task-set will not be active upon stimulus onset, but do not specify which of the remaining two task-sets will be active. Since for switch-away cues, activation of the relevant task-set could not be completed until after stimulus onset regardless of preparation interval, it was expected that switch-away cues would have larger RT switch cost than switch-to cues and that switch-away RT cost would not reduce with increasing CSI. However, as passive dissipation of task-set interference should not be affected by the information provided by the two switch cue types, increasing RSI for constant CSI was expected to reduce RT switch cost equally for switch-to and switch-away trials.#

ERP waveforms for switch-to and switch-away trials were compared only at the long CSI (1000 ms) that affords the opportunity for activation of anticipatory task-set reconfiguration and provides temporal dissociation between cue and stimulus related processes. If the cue-locked differential switch positivity that peaks before stimulus onset with long CSIs for switch-to trials indexes activation of the relevant task-set in anticipation of a cued switch in task-set, then since this process cannot be activated until after stimulus onset for switch-away trials, there should be no evidence of a cue-locked differential switch positivity for switch-away trials. Instead, for switch-away trials, this cue-locked differential switch positivity should emerge after stimulus onset just as it does when there is no advance cueing or when the CSI is very short (Nicholson et al., 2005). If a differential switch positivity is still evident within the CSI for switch-away trials, then clearly it cannot indicate activation of the relevant task-set, but presumably some other component of task-set reconfiguration that can be triggered by both switch-to and switch-away cues.#

Mean RT and error rate for each timing condition, trial type and task are shown in Table 1. The main effects of condition, trial type and task (F(2,70)=36.2, p<0.001; F(1,48)=229.3, p<0.001; F(1,35)=22.2, p<0.001, respectively) and the condition×trial type and trial type×task interactions (F(4,140)=6.9, p<0.001; F(2,58)=8.5, p<0.005) were significant. Overall, RT was longer for the digit task compared to letter task, and this effect was larger for both types of switch trials compared to repeat trials (Table 1). RT was also larger for the short CSI (RSI-1200:CSI-200) compared to the two long CSI conditions. The reduction in RT with increasing CSI was significant for all trial types, indicating a general preparation effect. This effect did not differ between repeat and switch-away trials (F<1) but was significantly larger for switch-to as compared to repeat trials (F(2,70)=9.29, p<0.001).#

These effects were further examined by calculating RT switch cost for switch-to and switch-away trials (Fig. 1). Switch cost was larger for the digit than for the letter task (F(1,35)=11.3, p<.005), however task did not interact with switch type or condition. As shown in Fig. 1, RT switch cost was larger for switch-away than for switch-to trials across all three conditions (F(1,35)=139.8, p<0.001). Paired comparisons between conditions showed that, for switch-to cues, RT switch cost did not differ significantly between short (1200 ms) and long (1600 ms) RSI (162 ms and 159 ms, respectively), but it was significantly greater in the short (200) compared to the long (1000) CSI (210 ms and 162 ms, respectively, F(1,35)=13.9, p<0.005). For switch-away cues, RT switch cost did not differ significantly between short and long RSI (269 ms and 252 ms, respectively) nor between short and long CSI (265 ms and 269 ms, respectively).#

The overall error rate was very low, ranging between 1.3 and 4.6% (Table 1). Transformed error scores showed a significant main effect of condition (F(2,70)=6.5, p<.005) reflecting slightly higher error rates in the short CSI (3.8%) compared to the long CSI conditions (3.0–3.2%). Both switch trial types produced more errors (3.9% for switch-away and 4% for switch-to) than repeat trials (2%, F(2,68)=23.5, p<0.001). There were no further significant effects in the error data.#

Behavioral data indicated that, although there was an overall effect of task on RT, task did not interact with RT switch cost type or condition. Therefore, in order to maximize signal to noise ratio, all ERP data were averaged over task (letter/digit).#

The switch-to cues used here were identical to the switch cues used in most previous task-switching studies in that they specified both the need to shift away from the current task-set and identified which task-set would be relevant on the upcoming trial. RT on switch-to trials was much slower as compared to repeat trials and RT switch cost ranged from 159 to 211 ms, which is larger than that found in our earlier studies (Nicholson et al., 2005). A number of factors may have contributed to this inflated switch cost. Firstly, this study included only bivalent incongruent stimuli that have been shown to produce slower RT and greater switch cost (Rogers and Monsell 1995; Woodward et al. 2003). Secondly, the task involved a larger switch trial probability (two-thirds of all trials) compared to our previous studies (Nicholson et al., 2005). Monsell and Mizon (2006) suggest that increasing the proportion of switch trials reduces participants' tendency to engage in anticipatory task-set reconfiguration, thus potentially leading to increased mean RT switch cost (De Jong, 2000). Thirdly, this study involved switching between three instead of two task-sets. This may have resulted in larger memory requirements and more stimulus-response mapping interference.#

As found in many previous studies (Meiran 1996; Nicholson et al. 2005; Rogers and Monsell 1995), increasing the preparation interval resulted in a reduction in switch-to RT cost (i.e., 48 ms decline as CSI increased from 200 to 1000 ms). However, contrary to previous studies (Meiran et al. 2000; Nicholson et al. 2005), increasing opportunity for passive dissipation of task-set interference (i.e., increasing RSI from 1200 to 1600 ms) had no further effect on switch-to RT cost. This may be attributable to the specific RSI values used in the current study. Although Nicholson et al. found that switch cost significantly declined as the RSI increased from 750 to 1200 ms, Meiran et al. found no reduction in RT switch cost as the RSI increased from 1000 to 3000 ms. At the shortest RSI value used in the current study (1200 ms), passive dissipation of task-set interference may have already plateaued. Thus, the further increase to 1600 ms provided no additional benefit.#

A large ‘residual’ switch cost (Rogers and Monsell, 1995) remained on switch-to trials, even with a preparation interval of 1 s (see also Nicholson et al., 2005). This ‘residual’ switch cost may reflect failures to engage in anticipatory task-set reconfiguration on some proportion of switch-to trials, thereby resulting in an overall increase in mean switch-to RT (De Jong, 2000). Upon stimulus onset, the spatial location of the stimulus provided a redundant cue as to which task was relevant. Therefore, although participants were encouraged to use switch-to cues to prepare in anticipation of stimulus onset, if they failed to do this on some trials, they could still respond successfully by initiating task-set reconfiguration after stimulus onset. Nicholson et al. (2006) found that eliminating redundant task cues and removing the cue prior to stimulus onset encouraged anticipatory task-set reconfiguration and reduced variability in strategy implementation both between participants and between trials for the same participant.#

Overall, RT switch cost was significantly larger for switch-away (∼260 ms) compared to switch-to (∼180 ms) trials. In contrast to switch-to trials, increasing the preparation interval from 200 to 1000 ms did not have any differential effect on switch-away as compared to repeat trial RT. This finding is compatible with the earlier findings by Dreisbach et al. (2002) and Hubner et al. (2003) that information about an impending switch trial without specific knowledge about which task to switch to does not provide any differential behavioral advantage and supports the suggestion by Dreisbach et al. that irrelevant task-set inhibition cannot occur independently of relevant task-set activation. Interestingly, RT cost was larger for switch-away than switch-to cues even at the short (200 ms) preparation interval. Nicholson et al. (2005) found that RT switch cost was smaller for 150 ms CSI compared to a no cue condition. These findings suggest that active task-set reconfiguration can at least be initiated prior to stimulus onset even with a very short CSI.#

Cue-locked ERP waveforms for switch-to trials showed a significant differential positivity that began approximately 100 ms after cue onset, peaked around 350–400 ms and was largest parietally. At long preparation intervals, this differential switch-related positivity extended across most of the CSI but resolved prior to stimulus onset. At the short preparation interval (200 ms), the differential positivity peaked after stimulus onset and had a shorter duration. For all conditions, switch-to trials showed a large stimulus-locked differential negativity. This emerged earlier for long CSI conditions (around 150 ms) than for the short CSI condition (around 300 ms) but peaked around 500 ms in all conditions. These findings are consistent with those of previous studies (Karayanidis et al. 2003; Miniussi et al. 2005; Nicholson et al. 2005; Rushworth et al. 2002 2005) showing differential processing in anticipation of a switch trial and differential processing of switch and repeat stimuli, thereby supporting the contribution of both active task-set reconfiguration and stimulus-elicited interference processes on RT switch cost.#

ERP differences between switch-to and switch-away trials were examined at the long preparation interval that provided opportunity for anticipatory task-set reconfiguration processes to be initiated prior to stimulus onset and temporally separated cue- and stimulus-locked processes. Switch-away difference waveforms showed a cue-locked differential positivity emerging approximately 150 ms after cue onset similar to that for switch-to difference waveforms (Fig. 3B, left). However, while for switch-to trials, this positivity remained significant centroparietally across the entire CSI, for switch-away trials, it reduced in amplitude over 350–450 ms and returned to baseline by 600 ms. Specific point-by-point comparisons between switch-to and switch-away waveforms showed that the later portion of the positivity was significantly smaller centroparietally over 400–450 ms and parietally over 650–750 ms after cue onset for switch-away trials. So, a differential switch-related positivity emerged within the CSI for both switch-to and switch-away waveforms but had a smaller amplitude and shorter duration in the latter cue type. Additionally, switch-away difference waveforms showed the emergence of differential positivity after stimulus onset (Fig. 3B, right). This post-stimulus differential positivity for switch-away trials was broadly distributed emerging around 150 ms and extending frontally beyond 300 ms. It was succeeded by a centroparietal differential switch negativity that peaked around 500 ms after stimulus onset, similar to that obtained for switch-to trials.#

Previous work (e.g., Nicholson et al., 2005) has suggested that the differential positivity to specific switch cues (i.e., cues that both signal an impending switch trial and specify which task will be active on the next trial) reflects task-set reconfiguration. The finding that, at long preparation intervals, this positivity is restricted within the CSI suggests that at least some aspects of task-set reconfiguration can be completed in anticipation of stimulus onset (see current switch-to data, but also Karayanidis et al. 2003; Nicholson et al. 2005; Rushworth et al. 2002 2005). However, the fact that a residual RT switch cost remains at even the longest preparation intervals combined with the evidence for differential post-stimulus processing of switch and repeat trials suggests that this anticipatory component of task-set reconfiguration cannot fully account for RT switch cost. Stimulus and/or response interference, as well as passive dissipation of the previously activated task-set may account for the post-stimulus differential switch negativity and contribute to the RT switch cost. Additionally, at short preparation intervals and in no cue conditions, the switch-related positivity emerges or peaks after stimulus onset (see current switch-to data for short CSI, but also Karayanidis et al. 2003; Nicholson et al. 2005), suggesting that task-set reconfiguration is a crucial process that is completed either before or after stimulus onset, depending on task parameters.#

Within this framework, the current findings for switch-away trials suggest that the process of task-set reconfiguration consists of multiple subcomponents that can be temporally dissociated. One process appears to be reflected in the emergence of a cue-locked differential positivity for both switch-to and switch-away trials. This finding indicates that forewarning of an impending switch trial initiates differential processing between switch and repeat trials, even in the absence of information about which specific task will be active on the subsequent trial. This process cannot relate to activation of the task-set that will be relevant for the next stimulus because this information is not yet available for switch-away trials. A second process is reflected in the post-stimulus differential positivity for switch-away compared to repeat trials. This is believed to reflect the active engagement of the relevant task-set, which for switch-away trials is only possible after the stimulus has been presented and its position has indicated which task to perform.#

Therefore, it is suggested that the late component of the cue-locked differential positivity for switch-to trials and the stimulus-locked differential positivity for switch-away trials both reflect activation of the relevant task-set. This interpretation is supported by the finding that switch-away trials have a larger RT switch cost than switch-to trials, even at the short 200 ms CSI (i.e., even with a CSI of only 200 ms, participants were able to begin preparation for the new task-set following cue presentation on switch-to trials, resulting in reduced RT switch cost compared to switch-away trials). It is also supported by earlier findings that, at short preparation intervals (e.g., see short CSI condition in Fig. 2A) and in no cue conditions (Nicholson et al., 2005), the differential switch positivity emerges and/or peaks after stimulus onset.#

The functional significance of the early cue-locked positivity for both switch-to and switch-away trials is more obscure and a number of possible interpretations will be examined. Given that the only common information provided by switch-to and switch-away cues is that the following trial will not be a repeat trial, it is possible that the common early differential positivity may reflect suppression of the previously active but now irrelevant task-set (e.g., having just completed a letter task, the letter task can be safely inhibited). As switch-away trials are associated with larger RT switch cost than switch-to trials and show no reduction in RT switch cost with increasing CSI, it would appear that, if such a suppression process exists, it has no effect on speed of responding. This, however, is not necessarily true. It is possible that any behavioral effect of this suppression or inhibition process would have reached a plateau by 200 ms and would only be evident when comparing performance on trials with a completely non-informative cue vs. switch-away cues. This interpretation appears largely compatible with Barceló et al.'s (2002) frontal P3a component that is elicited to feedback cues indicating a shift in task-set within the Wisconsin Card Sorting Task (WCST) and that can be differentiated from a later parietal component that Barceló et al. suggest is a P3B, which reflects retrieval of task-set into working memory. Like our switch-away cues, Barceló's 3D shift feedback cues indicate that the previous task-set is no longer relevant but do not specify which task-set will be relevant on the subsequent trial.³3Although our switch-away cues and Barceló's 3D shift feedback cues both indicate that the previously active task-set is no longer relevant, they convey different information about how to proceed. In the WCST, participants select one of the alternative task-sets, implement it upon stimulus onset and wait for the next feedback cue to determine whether it was correct. In the present task, the relevant task-set is defined upon stimulus onset and, as discussed below, there was no evidence that participants either prepared both task-sets or prepared one task-set based on subjective sequence expectancies. Despite these task differences, in both studies, the late parietal positivity to shift cues behaved quite consistently; Barceló et al. found that P3b amplitude reduced with increasing certainty about which task-set would be active upon stimulus onset (e.g., 3D shift vs. 2D shift feedback cues), and the present data showed a larger late differential switch positivity for switch-to than for switch-away cues. However, this interpretation still contradicts the conclusion drawn on the basis of recent behavioral data that inhibition is not an independent process, but a by-product of activation of the relevant task-set (e.g., Dreisbach et al. 2002; Hubner et al. 2003) or is triggered upon response selection (Schuch and Koch, 2003). Clearly, there are differences in the type and timing of information provided by ERP and behavioral data and further research is necessary to reconcile these discrepancies.#

An alternative possibility is that the early differential positivity for both switch-away and switch-to relative to repeat cues reflects differing amounts of information provided by the cue.⁴4We thank Dr. Leuthold for this suggestion. Specifically, repeat cues may be said to provide no new information as the same task-set will be implemented again, switch-away cues provide information about an impending task change, whereas switch-to cues also indicate which task-set needs to be implemented. Given the sensitivity of the P300 component to stimulus information value (e.g., Rugg and Coles, 1995), it is possible that the early portion of the differential positivity for switch-to and switch-away cues represents modulation of P300 by cue information value. Within this framework, given that the early differential positivity did not differ in amplitude between switch-to and switch-away cues, it would then appear to reflect general processes associated with preparing for a switch in task-set, whereas the prolonged differential positivity for switch-to cues would reflect processes more specifically associated with activation of the new task-set. Further work is needed to define what is encompassed by the general preparation processes that are evident for both switch-to and switch-away cues and to explain why this general preparation does not appear to contribute to behavioral performance (see earlier discussion).#

Another possibility is that switch-away cues elicit the activation of both possible task-sets and their maintenance in working memory. This could account for the overall larger RT cost on switch-away compared to switch-to trials and for the lack of a switch cost reduction with increasing CSI for switch-away trials as activation and maintenance of two task-sets take longer to be completed. However, if this were the case, then switch-away trials would be expected to show a larger and more prolonged differential positivity in the CSI than switch-to trials, reflecting greater activation of task-set reconfiguration processes. This was clearly not the case in the current data. Furthermore, this interpretation cannot also account for the additional post-stimulus positivity that occurred exclusively for switch-away trials.#

A related possibility is that, on switch-away trials, participants activated one of the two possible task-sets basing their choice on subjective expectancies derived using local probabilities (e.g., if the digit task has not been presented recently, a participant may activate that task-set based on their assumption that there is an increased probability that it will be presented on the next trial). In this case, on roughly half of the trials they would have successfully reconfigured the correct task-set, whereas on the other half they will have to reconfigure again after stimulus onset. This explanation can account for the overall larger RT switch cost for switch-away than switch-to trials, for the emergence of an early differential positivity for both types of switch trials and for the brief post-stimulus differential positivity on switch-away trials only. However, it is hard to reconcile with the reduced amplitude and duration of the cue-locked differential positivity for switch-away compared to switch-to trials as well as with the absence of any reduction in RT switch cost with increasing CSI for switch-away trials.#

The differential early positivity for switch-to and switch-away trials could, alternatively, reflect differential processing of the cue on repeat compared to switch trials. That is, it may not represent a process that is common to the two types of switch trials, but a process that occurs for repeat trials only. This differential positivity may then represent processes that contribute to a repetition benefit (Dreisbach et al. 2002; Logan and Bundesen 2003; Mayr and Kliegl 2003). This appears to be unlikely in the current data. The outline of the circle and the six wedges were continuously displayed, and the cue involved highlighting two of the six wedges that were associated with particular tasks (Fig. 4A). The cue itself was thus identical on all trials, only its spatial position across the circle differed. Additionally, Nicholson et al. (2006) show that, in well practiced participants, alternating between explicitly mapped and easy to process cues produces no behavioral effects of a switch in cue, while ERP effects of a cue switch are restricted to early endogenous components. On the other hand, based on the argument by Dreisbach et al. (2002) that active preparation occurs for the least expected trial type and the fact that, overall, there were fewer repeat than switch trials in the current paradigm, it could be argued that task-set reconfiguration occurred here for repeat rather than switch trials. However, in that instance, one would expect that the switch-related differential positivity would occur for repeat trials and that no switch cost would be observed, which was clearly not the case.#

The electrophysiological data are compatible with the construct of task-set reconfiguration as an active control process that is necessary for successful switching between tasks. These findings suggest that the process of task-set reconfiguration consists of multiple subcomponents, however it is the activation of the currently relevant task-set prior to stimulus onset that facilitates the reduction in RT switch cost previously observed with increasing preparation interval. With long preparation intervals and complete information about an upcoming switch trial, task-set reconfiguration can be completed before stimulus onset. However, if the preparation interval is insufficient or the information provided is inadequate, then all or part of this process will be completed after stimulus onset.#

Thirty-six undergraduate students (mean age 23 years±6.8 years, range 18 to 40 years; 29 female) participated for course credit in an introductory psychology course. Participants had no prior exposure to task-switching paradigms and provided written informed consent. The study was approved by the Newcastle University Human Research Ethics Committee.#

For each run of trials, a circle (230 pixels diameter) divided into 6 equal wedges was continuously displayed in the middle of a computer monitor at approximately 90 cm viewing distance. On each trial, a stimulus was displayed in the center of one of the six wedges (Fig. 4A). Pairs of adjacent wedges were grouped by thicker lines extending slightly beyond the perimeter of the circle, thus demarcating three sections. Each of the three sections was assigned to one of three tasks. The assignment of task to section was counterbalanced across participants.#

In the letter task, participants responded whether the letter presented was a vowel (A, E, I, U) or a consonant (G, K, M, R). In the digit task, participants responded whether the digit presented was odd (3, 5, 7, 9) or even (2, 4, 6, 8). In the color task, participants responded whether the color of the stimulus was hot (red, pink, orange, burgundy) or cold (ocean blue, emerald green, sky blue, turquoise). Participants used their left and right index fingers to respond (Fig. 4B), and the hand assigned to each response was counterbalanced across participants.#

Stimuli consisted of a pair of characters in Times New Roman font (60 by 60 pixels). Characters could be a letter, a digit or a non-alphanumeric character (#, ?, *, %) that was not mapped to a response (neutral). Stimuli were presented either in gray (neutral) or in one of eight colors listed above. Each stimulus therefore consisted of three dimensions: character 1, character 2 and color. One dimension was selected from the currently relevant task-set (i.e., a digit for the digit task). The second dimension was selected from one of the two alternative task-sets (i.e., a letter or a color). This second dimension was always incongruently mapped to the first dimension, whereas the third dimension was always neutral. For example, assume that the digit task was relevant on the current trial and that the number ‘3’ was presented, thus requiring a left hand response. The second character could be either a letter or a neutral stimulus. If it was a letter, then it would have to be a consonant, so as to be mapped to an incongruent response (i.e., right hand). In this case, the third stimulus dimension (color) would be neutral (i.e., gray). For example, the stimulus ‘3G’ would be presented in gray and the correct response would be left. Alternatively, if the second character was neutral (e.g., #), then the stimulus would be presented in one of the cold colors, so that again it would be mapped to an incongruent response (right hand). For example, the stimulus ‘3#’ would be presented in green. The combination of stimulus dimensions and the position of task-relevant and task-irrelevant characters (e.g., 4U, U4) were varied pseudorandomly across trials. Stimuli were selected pseudorandomly from their character set with the only restriction that the same stimulus could not appear on two successive trials.#

Each trial began with a cue that highlighted the border surrounding two of the six wedges (Fig. 4C). Three different trial types were defined by the location of the cue and were pseudorandomly selected with equal probability. The same trial type was never repeated more than 3 times. On repeat trials (1/3 of trials), the cue highlighted the section of the circle assigned to the same task as on the previous trial. Fig. 4C (repeat sequence) shows that, having completed a letter task trial, the next cue signals that the following trial will also be a letter task trial. On half of the repeat trials, the stimulus appeared in the same wedge as on the previous trial, while on the other half it appeared in the adjacent wedge within that task section.⁵5As noted by an anonymous reviewer, stimulus location repetition is thus possible for task repetitions, whereas it cannot occur for task-switch trials, possibly providing some extra positive priming on task repeat trials. Unfortunately, this was unavoidable as the alternative (i.e., changing the location on every repeat trial) would have produced an even greater confound. That is, participants would have known the exact location of the upcoming stimulus on repeat trials only, possibly facilitating faster RT compared to switch trials where the stimulus could appear in one of two alternative locations. The other two types of cues indicated that the next trial would involve a switch in task, but provided different degrees of information regarding what the next task would be. Switch-to cues (1/3 of trials) highlighted a section assigned to one of the other two tasks, thereby validly cueing which task-set would be active on the upcoming trial. Fig. 4C (switch-to sequence) shows that, having performed a letter task trial, the next cue indicates that the following trial will involve a switch-to the digit task. Switch-away cues (1/3 of trials) transgressed task sections and highlighted wedges belonging to the two tasks that were irrelevant on the previous trial, thereby indicating that the current task would not be repeated, but not indicating which task-set would be active on the upcoming trial. As shown in Fig. 4C (switch-away sequence), the cue highlighted two segments, one assigned to the digit task and one assigned to the color task. Therefore, having completed a letter trial, the highlight indicated that the upcoming trial would require a switch-away from the letter task-set. The active task-set (digit or color) is then determined by the position of the stimulus itself and thus the new task-set cannot be activated until after stimulus onset. The cue and the stimulus remained on the screen until a response was generated or 5000 ms had elapsed.#

Three timing conditions were used to vary the CSI and RSI (Fig. 4D). RSI was manipulated across two levels: short (1200 ms) and long (1600 ms). At the shorter RSI, CSI varied across two levels: short (200 ms) and long (1000 ms). Condition labels represent the duration of the RSI and CSI (e.g., RSI-1200:CSI-200=RSI of 1200 ms and CSI of 200 ms).#

All participants attended 2 sessions scheduled 2 to 14 days apart. The first session provided substantial task training and practice. Training began with two runs (48 trials/run) on each task alone. This was followed by two runs (72 trials/run) of repeat and switch-to trials and two runs of repeat and switch-away trials. Session 1 finished with 2 runs (228 trials/run) of combined repeat, switch-to and switch-away trials. Throughout training, stimulus-response mapping (Fig. 4B) and task location (Fig. 4A) were displayed continuously at the bottom of the computer monitor and the CSI was a minimum of 600 ms.#

The second session included further practice followed by the behavioral and ERP testing session. Second day practice included one run (48 trials/run) on each task alone, two runs (72 trials/run) with repeat and switch-to, and repeat and switch-away trials, respectively, and two runs (228 trials/run) of combined repeat, switch-to and switch-away trials (RSI-1200:CSI-1000 and RSI-1200:CSI-200, respectively). Participants thus completed a total of 1776 practice trials across the 2 training and practice sessions.#

The behavioral and ERP testing session consisted of the 3 timing conditions presented in blocks of 3 runs (228 trials/run). Each condition was presented in a separate block, and the order of block presentation was counterbalanced across participants using a Latin square design. The first 12 trials of every run were consider ‘warm-up’ trials and were discarded from analysis. For all participants, stimulus-response mapping and assignment of tasks to specific sections within the circle remained constant across training and testing sessions. Participants were instructed to respond as quickly as possible, without making too many errors. At the start of each block, participants were informed of the specific RSI and CSI to be used and were encouraged to use the CSI to prepare for the next trial based on the information provided by the cue. Incorrect responses resulted in auditory feedback, and the onset of the next cue was delayed by 1000 ms. A short break was provided at the end of each run, and behavioral feedback (overall average RT and percentage of trials correct) was displayed. Participants were encouraged to monitor and improve their performance.#

The main analysis of both behavioral and ERP measures was conducted on the data from digit and letter task trials alone. Data from color task trials were excluded from the main analyses for a number of reasons. Firstly, this increased comparability with our earlier work that had included only the letter and digit tasks (Karayanidis et al. 2003; Nicholson et al. 2005). Secondly, selective attention to color attributes has been shown to result in differences in behavioral and ERP measures of attention as compared to attention to other attributes such as location, orientation, size and pattern (e.g., Karayanidis and Michie 1997; Michie et al. 1999), suggesting possible differences in attentional processing of color and other attributes. Finally, initial inspection of the data indicated that the pattern of behavioral findings was different for the color task than for the other two tasks. Mean RT was considerably faster for the color task (771 ms) than the other two tasks (883 ms; F(1,35)=51.58, p<0.001) and showed significantly less switch cost (switch-to cost: 122 vs. 177 ms; F(1,35)= 18.36, p<0.001; switch-away cost: 220 vs. 262 ms F(1,35)= 7.5, p<0.01). Further inspection of ERP waveforms (Fig. 5) showed that the repeat trial waveform for the color task was noticeably different to those for letter and digit tasks, resulting in spurious task differences in the switch-related difference waveforms. These differences are interesting insofar as they suggest a difference in task-switching processes dependent on task attributes and/or task difficulty that need to be further explored in their own right. However, as these effects are not central to the aims of this paper, color task trials were excluded from behavioral and ERP data analysis.#

Trials associated with an incorrect response, trials immediately following an incorrect response and trials on which the response occurred outside a window of 200 to 2000 ms after stimulus onset were excluded from analysis. Where necessary, critical values were adjusted using the Greenhouse–Geisser correction to avoid violating the assumption of sphericity (Vasey and Thayer, 1987). Standard error values were calculated using the method for within-subjects designs described by Loftus and Masson (1994).#