Task-Switching
The human environment is extremely complex. At any one time, there can be many hundreds of stimuli competing for our attention. Yet, regardless of the complexity, humans are able to filter through the plethora of incoming information to act on stimuli that are relevant to a current behavioural goal; all goal-irrelevant stimuli can be successfully ignored.
It is common during our daily lives that we must switch attention between competing stimuli to achieve a particular goal. A good example of this is driving: traffic lights, other cars, road signs, and pedestrians contribute to a dangerous environment that must be navigated with extreme precision to avoid disaster. The processes underlying the ability to efficiently switch attention are deceptively complex, and form the basis of my research.
The task-switching paradigm
In the laboratory, the mechanisms enabling fluent shifts of attention are often researched utilizing the so-called explicitly cued task-switching paradigm. Typically, participants must perform simple cognitive judgments on number stimuli (e.g. Odd/Even or Higher/Lower than 5 on numbers 1-9). Participants know which task to perform on the number on the basis of an instructional cue presented before the stimulus.
The above figure shows a schematic view of a typical task-switching experiment. Participants must perform their number judgment as quickly and as accurately as possible. It is a consistent finding within this paradigm that reaction times (RT) to task-switches are slower and more error prone than RT to task repetitions. This delay in response has been called the "Switch Cost", and has become a topic of great debate within the literature. It is an appealing effect, as the switch cost was thought to reflect the extra time needed for the cognitive system to 'reconfigure' itself to be able to perform the new task. During task-repetitions, such reconfiguration is not required as the correct task is already 'installed'. Consistent with this view is the finding that the switch cost is reduced given longer preparation times, measured by the distance between the cue presentation and stimulus display (cue-stimulus interval, CSI). Therefore, it was once thought that the switch cost may reflect the temporal dynamics of cognitive control.
However, there are many problems with the reconfiguration metaphor. Most problematic might be that the reduction of switch cost at extended CSIs is not as 'pervasive' as once believed. For example, the work of Erik Altmann and Iring Koch has suggested that the CSI*sequence interaction is only evident if CSI is manipulated within-subjects. As findings suggest that longer preparation times affect repeat-trial RTs as well as switch trial RTs at longer CSIs (between subjects), the preparation being done is not switch specific. This favours an activation account of switch cost, i.e. that there is no specific mechanism for switching, rather performance is aided by a switch-neutral mechanism of activating the relevant task code so it is active above all competitors. This mechanism is switch-neutral in as much as task codes on a repetition trial must be more active than its competitors just as task codes on switch trials must be. The difference is that task-repetitions benefit from repetition priming. The advantage of this account is its parsimony; no new mechanism needs to be postulated just to explain one experimental effect, rather it can be explained in terms of existing memory and attentional processes. One mechanism that might allow a task code to become dominant over its competitors is inhibition, for which there is growing evidence for in the task-switching literature.
Problems with the task-switching paradigm
Another challenge to the reconfiguration view of switch costs is the problem that task-switches are confounded with cue-switches. This can clearly be seen from the above figure: whenever the task repeats, so too does the cue; when the task switches, the cue switches. Using two cues per task allows researchers to overcome this problem, and thus allows us to separate the effects on performance of switching cues Vs switching tasks. Two cues per task allows three types of sequence: Both Repeat (task repeats, cue repeats - traditional 'repetition' trial), Task-Switch (task switches, cue switches - traditional 'switch' trial), and unique to this paradigm, Cue-Switches (task repeats, but cue switches, e.g. "Odd/Even - Parity"). The cost of switching cues can now be measured by subtracting both-repeat RTs from cue-switch RTs, and an unconfounded measure of task-switching by subtracting cue-switch RTs from task-switch RTs. It is a consistent finding that merely switching cues induces a large RT cost, called the cue-switch cost. It is universally accepted that cue-switch costs are generated by repeated cue encoding on both-repeat trials. Thus, any process that allows translating the task cue into a durable task representation in working memory is primed when the cue repeats; the cue-switch cost therefore becomes an important phenomenon allowing investigations into what processes serve forming these task representations in WM. Ongoing work for my PhD is finding interesting effects that suggest that this process can be affected by the complexity of the cue-task relationship, but more on this soon hopefully (Grange & Houghton, submitted)!
What is even more problematic for the reconfiguration view is that RTs for task-switches are sometimes not significantly greater than cue-switch RTs, thus suggesting much of the traditional switch cost is due to a repeated cue-encoding benefit on task-repetition trials. However, this debate is generally coming to the conclusion that under a lot of situations significant task-switch costs can be found. Indeed, my work shows task-switch costs are still evident under some conditions despite no (or reduced) cue-switch costs, suggesting the task-switch costs can not be explained exclusively by cue-encoding processes. However this does not automatically mean they are generated by reconfiguration processes. For an overview of a lot of the research already done into the costs of switching cues and tasks including formal mathematical modeling, see Darryl Schneider's home page.