Shifting Intentional Set: Exploring the Dynamic Control of Tasks

We describe a series of seven exploratory experiments, designed to investigate the effects of shifting different components of task set, both in speeded response tasks and in monitoring rapidly presented word sequences. Subjects made preinstructed attention shifts between different stimulus dimensions, between different semantic categories or attributes, between different cognitive operations, and between different response modes. Under many task-shift conditions (but not all), we found large and systematic costs in the speed and/or accuracy of performance, compared to conditions in which the same set of cognitive and perceptual motor operations was performed repeatedly over successive stimuli.Our intention was to make use of these behavioral costs to explore the time course of underlying, voluntary control processes, and - if possible - to identify separable and/or shared executive resources, responsible for different features of task control. For example, is the difficulty of shifting set affected by the difficulty of the individual tasks, or by the degree of ''automaticity'' or ''control'' required by one or both of them? Does the requirement to shift concurrently more than one task attribute increase the RT cost of shifting? Does a stimulus-cued shift of task show smaller (or even zero) RT costs? The answers to several of these questions are strikingly counterintuitive, and we report a number of behavioral phenomena not previously described.As our exploration proceeded, however, it became increasingly clear that the RT costs of shifting or reorienting task set do not directly reflect the time needed to complete a hypothetical shift-of-set operation, prior to executing the shifted task. Instead, we demonstrate that the shift costs represent a form of proactive interference from features of the task set implemented in preceding tasks. This task set inertia persists over periods lasting very much longer than the RT costs themselves.In semantic monitoring of word lists under rapid sequential visual presentation (RSVP) conditions, a voluntary shift of criterion causes a large and abrupt drop in target identification accuracy, which then progressively recovers, eventually to reach preshift levels of accuracy. However, we also found that the process of recovery from this shift cost, in RSVP monitoring, is not a function of the real-time delay of a target item after the shift cue. Neither the RT shift cost, therefore, nor the RSVP criterion shift cost in monitoring accuracy, directly reflect real-time control operations. These findings raise some fundamental questions for our understanding of voluntary executive control.