An alternative view of preparation

Each of the studies discussed above has assumed that any reduction of the switch cost by the preparation interval results from processes which reduce the disadvantage of switch trials, compared with nonswitch trials. This might result

The carryover effect m ay be m inim ised by activating the task set for the forthcom ing task and/or inhibiting the task set for the previous task.

either from active engagement of the forthcoming task during the preparation interval (e.g. Meiran, 1996) or passive disengagement of the previous task (e.g. Meiran et al., 2000). However, a very different view has recently been put forward by Dreisbach, Haider, and Kluwe (2002; see Sohn & Carlson, 2000 for a similar account). According to Dreisbach et al. (2002), preparation leads to an ^expectation advantage^ for the forthcoming task, which is potentially of equal benefit for switch and nonswitch trials. However, these authors claim that subjects in task switching experiments implicitly expect task repetition trials. Thus, the manipulation of the preparation interval may have less effect on nonswitch trials, which were (implicitly) expected anyway, compared with switch trials, which are unexpected. In other words, whatever it is that makes switch trials less efficient than nonswitch trials is not affected by the preparation interval. Rather, the preparation interval is able to lead to a general improvement in performance, which is less evident on nonswitch trials since subjects prepare for them by default.

To support this hypothesis, Dreisbach et al. (2002) report a series of experiments where each trial was preceded by a cue which announced the forthcoming task probabilistically. There were four tasks, each associated with a different colour. The task cue consisted of four coloured squares. If all four squares were blue, for example, this predicted the 'blue' task with 100% probability. If three squares were blue and one square was red, this predicted the blue task with 75% probability and the red task with 25% probability. By cueing the task in this manner, Dreisbach et al. (2002) were able to compare the performance of each task when it was predicted with 25%, 50%, 75% or 100% probabüity. They found that performance improved with increasing probability (i.e. there was an 'expectation advantage'), but this improvement was equal for switch and nonswitch trials. Thus, task preparation speeded responses without influencing

the switch cost. Dreisbach et al. (2002) therefore conclude that switch costs arise from an automatic carryover effect from the previous task that cannot be overcome by advance reconfiguration.

The results from these experiments certainly seem to indicate that there is at least one factor, related to task expectancy, which improves performance without affecting the cost of task switching. This is consistent with the hypothesis put forward by Dreisbach et al. (2002), but it does not establish that the switch cost is

always impenetrable by preparation. This strong conclusion rests on the assumption that the probabilistic cueing method employed by Dreisbach et al. (2002) is an unbiased methodology for assessing the effect of the preparation interval on switch and nonswitch trials, whereas other task switching methodologies are biased by subjects' implicit expectation of nonswitch trials. However, in the absence of direct evidence for this assumption, it seems premature to conclude that preparation can never reduce switch costs by leading to an advance reconfiguration of task set^.

One relevant finding for the evaluation of Dreisbach et al.'s (2002) hypothesis is that preparation, while reducing switch costs, typically does not seem to reduce the amount of Stroop-like interference, or 'cross-talk', between tasks (e.g. Allport et al., 1994; De Jong et al., submitted). Dreisbach et al. (2002) would be able to explain this finding if they assume that task expectancy, as well as having equal effects on switch and nonswitch trials, also has equal effects on incongruent and

^ A piece o f evidence consistent with this assum ption is the greater ability o f preparation to reduce switch costs in random than predictable task sequences (T om ay & M ilan, 2001). It seem s m ore plausible that subjects im plicitly expect a task repetition in unpredictable sequences, since preparing for a switch before the task cue is as likely as not to be counterproductive. It is less plausible that subjects w ould im plicitly expect a task repetition in the preparation interval before a sw itch trial when the task sequence is regular.

neutral trials. For alternative accounts which are able to explain this finding, see Rubinstein et al. (2001) and Meiran (2000a). However, a reduction in Stroop-like interference with preparation has been found by other researchers (Goschke, 2000; Sudevan & Taylor, 1987), which is problematic for theories that rule out any effect of preparation on the amount of cross-talk between tasks.

In summary, many unresolved questions remain concerning the effects of preparation on task switching. In particular, it is still not clear whether there is some inexorable limit to the ability to prepare for a switch, or whether it is possible in some circumstances to eliminate the switch cost by preparation. If it is possible to eliminate the switch cost, it is not clear which circumstances promote this possibility. Another outstanding question concerns what subjects actually achieve through preparation (e.g. full reconfiguration of task set, 'expectancy' for the forthcoming task, 'goal shifting', etc.) It does not seem possible at present to arbitrate between task carryover and extra process accounts of switch costs on the basis of the evidence derived from studies of preparation.

It is perhaps not surprising that the nature of preparation for task switching is iU understood, when one considers the number of factors which have been implicated in this process. The role of preparation appears to be affected by the use of random versus regular task sequences (Tomay & Müân, 2001), whether the preparation interval is random or blocked (Rogers & Monsell, 1995), the nature of task cueing (e.g. verbal task cues versus spatial cues, cf. Arbuthnott & Woodward, 2002), whether task cues are probabilistic (Dreisbach et al., 2002), and whether the stimuli and responses are univalent or bivalent (Meiran, 2000b). Few studies have independently manipulated cue-target and response-cue intervals, in order to separate the effects of decay of the previous task set versus

active preparation for the forthcoming task (cf. Meiran et al., 2000). In addition, the number of tasks between which the subject must switch has differed between studies. Whereas most studies have employed two tasks (e.g. Rogers & Monsell, 1995), others have used three (e.g. Arbuthnott & Woodward, 2002) or four (Dreisbach et al., 2002). Finally, there have been differences between studies in the nature of the tasks that subjects perform. Some studies have used simple spatial S-R tasks (e.g. Meiran, 1996), whereas others have used more complex classification tasks, involving simple perceptual classifications (e.g. Rubinstein et al., 2001), simple semantic classifications of single characters (e.g. odd vs. even, or vowel vs. consonant; Rogers & Monsell, 1995; Tomay & Milân, 2001) or more complex semantic classifications of words (e.g. animacy; Mayr & Kliegl, 2000). Systematic investigation of each of these factors, and their interrelations, may result in a better understanding of the complex processes involved in preparation.