Resolving dual-task interference: an fMRI study

Abstract

The human cognitive system is severely limited in the amount of information it can process simultaneously. When two tasks are presented within a short stimulus-onset-asynchrony (SOA), reaction time of each task, especially task 2, is dramatically delayed. Previous studies have shown that such delay is accompanied by increased activation in the right inferior frontal gyrus (GFi). In this study, we address the role of right GFi in resolving dual-task interference at two different stages: allocation of perceptual attention and response selection. We scan 12 subjects using functional MRI while they conduct two tasks—shape discrimination in task 1 and color discrimination in task 2—and vary the SOA between tasks as 100 or 1500 ms. The targets are located at the center or at the periphery. When both are at the center, they compete primarily for response selection. When both are at the periphery, they additionally compete for the allocation of perceptual attention. Results show that the right GFi and frontal operculum regions are significantly more active in the short SOA than the long SOA condition, but only when subjects attend to the periphery in both tasks. We conclude that the right lateral frontal regions are important for resolving dual-task interference at the perceptual attention stage.

Introduction

One of the most fascinating aspects of human cognition is our ineffectiveness in conducting two tasks at the same time. Despite frequent wishes to divide attention between two tasks—such as turning left in traffic while carrying out a coherent conversation—most of us choose to stop performing one task until the other is complete. But what prevents one from responding to the traffic light at the same time as choosing the next sentence to say? Research in cognitive psychology has provided extensive evidence that human attention is limited in at least two respects: perceptual attention and response selection (Jiang and Kanwisher, 2003a).

Perceptual attention is used primarily to select targets and filter out distractors when multiple objects are presented. It can be simultaneously directed to approximately four visual objects. For example, in multiple-object tracking, subjects first see 10 dots, a few of them blinked initially, and then all dots move randomly on the screen. The subjects' ability to track the previously blinked dots is severely impaired if more than four dots are to be tracked (Pylyshyn and Storm, 1988).

Response selection also poses a cognitive limitation. Pashler, 1984, Pashler, 1994 proposes that there is a central cognitive bottleneck involved whenever a stimulus has to be mapped onto a response from an arbitrary rule, such as deciding to step on the brake when the traffic light turns red. Mapping a stimulus to a response—a decision process—precedes the execution of motor responses. Thus, while we can press a key and say a word at the same time (Pashler, 1993), we cannot simultaneously decide which key to press on the basis of a shape and decide which word to say on the basis of a tone.

Response selection is not only different from motor execution, but also separate from the allocation of perceptual attention. This is revealed in that the central bottleneck cannot be divided between two response selection processes, but perceptual attention can be divided between a few perceptual objects. Thus, when people encounter a competition for response selection, they hold the second task in a queue; but when they encounter a competition for perceptual attention, they divide a proportion of perceptual attention to each object. Therefore, perceptual attention can be simultaneously divided between up to four perceptual objects Luck and Vogel, 1997, Pylyshyn and Storm, 1988, but response selection can only be sequentially allocated to one task at a time Pashler, 1989, Pashler, 1991.

In this study, we use functional MRI to investigate how the brain resolves dual-task interference at the stage of perceptual attention and the stage of response selection. Competition at a third stage—the motor response stage—has been addressed by a separate study by Herath et al. (2001) and will be discussed later. We use the “overlapping task” design adopted from behavioral studies on the central bottleneck Pashler, 1984, Welford, 1952. In this design, two tasks are presented, separated by a variable stimulus-onset-asynchrony (SOA). When the SOA is short (e.g., 100 ms), the two tasks overlap in time and compete for perceptual attention and response selection. In this case, subjects have to perform task 1 AND task 2. As the SOA increases (e.g., to 1500 ms), the two tasks no longer overlap, so subjects only need to perform task 1 OR task 2. The difference between a short SOA condition (“AND”) and a long SOA condition (“OR”) reflects how the brain copes with overlapping stages of processing between two different tasks.

Several previous neuroimaging studies have studied the effect of dual-task processing (e.g., Adcock et al., 2000, Bunge et al., 2000, D'Esposito et al., 1995, Klingberg, 1998, Koechlin et al., 1999). These studies compare blocks of concurrent processing of two tasks (e.g., sentence comprehension and mental rotation) with blocks of single tasks alone. They observe increased activation in the dual-task condition in the dorsolateral prefrontal cortex and other regions. This suggests that additional central executive control is needed to resolve the dual-task interference. However, two design properties make these studies nonideal for determining what goes on during overlapping processes. First, the dual-task condition uses a mixed-block design, which entails task-set switching between the two tasks Allport and Wylie, 2000, Monsell and Driver, 2000. When a difference is found between a mixed-block and a single-task block, it may be attributed to either dual-task processing or task-set switching. Second, although the two tasks overlap in time in some studies, other studies have deliberately avoided overlapping processing. For example, in Klingberg's (1998) study, the two tasks in the dual-task block are scheduled apart by 1 s to avoid a competition in motor output, so the dual-task blocks actually involve the processing of “task 1 OR task 2”. In the studies of D'Esposito et al., Adcock et al., and Bunge et al., although the two tasks are presented inside the same time window, each trial of one or both tasks is extended for several seconds. This may allow subjects to sequentially perform the two tasks (cf. Pashler, 1994). Thus, these studies have primarily examined the effect of maintaining two task sets and switching between them (“OR” vs. single tasks; see also Dove et al., 2000, Kimberg et al., 2000, Konishi et al., 1998, MacDonald et al., 2000, Rushworth et al., 2002), and have not isolated the effect of overlapping task processing per se (“AND” vs. “OR”).

Four other studies have investigated the effect of overlapping tasks. Herath et al. (2001) present two tasks separated by either a short SOA (200–300 ms) or a long SOA (1150–1350 ms). One task is a simple visual RT task in which subjects press a key whenever an LCD is illuminated. The other task is a simple tactile RT task in which subjects press another key whenever they receive tactile stimulation. The choice of two different input modalities minimizes competition for perceptual processing (Pashler, 1998), and the choice of two simple RT tasks minimizes competition for response selection (Pashler, 1994). Herath et al. find increased activation in the right inferior frontal gyrus (GFi) in the short SOA compared with the long SOA condition. They suggest that the activation results from a competition for the same motor effector.

Szameitat et al. (2002) and Schubert and Szameitat (2003) test brain regions involved in tasks that overlap at a central stage. They use two choice-RT tasks that compete for the central bottleneck: a visual–motor and an auditory–motor task. Subjects are tested in a short SOA condition and blocks of single tasks. The short SOA condition shows increased activation in the right GFi compared with single tasks, suggesting that this region is involved not only in competition for the same motor effector (Herath et al., 2001), but also in competition for the central bottleneck. The interpretation of the studies of Szameitat et al. and Schubert and Szameitat is limited, however, by the omission of a long SOA condition. As noted earlier, the difference between a short SOA condition and single task blocks includes two components: resolving dual-task interference and switching task sets. It is unclear whether the right GFi activation is produced by the need to switch task sets in the short SOA condition, by competition for the central bottleneck, or both. In fact, when a long SOA condition is added in a recent fMRI study (Jiang et al., in press), the difference between the short SOA and the long SOA condition in the right GFi and other brain regions largely disappears, while the difference between the long SOA condition and single task blocks remains (Jiang et al., in press). This suggests that the right GFi may be more involved in task-set switching than in overlapping processing of response selection.

In this study, we further investigate the role of the right GFi in resolving dual-task interference at two different stages: perceptual attention and response selection. We hypothesize that GFi is activated whenever two processes proceed simultaneously, but not when they occur sequentially. We further hypothesize that simultaneous processing occurs when two perceptual processes or when two motor processes are separated by a short SOA, but not when two response selection processes are separated by a short SOA. As a result, the GFi will increase its activation when two tasks compete for the allocation of perceptual attention or for motor processing (Herath et al., 2001), but not when they compete for response selection. This study focuses on perceptual attention and response selection.

We test subjects in three task combinations: when subjects attend to the center target in both tasks, when they attend to the center target in task 1 and to the peripheral target in task 2, and when they attend to the peripheral target in both tasks.

On each display, subjects are presented with one central target, one peripheral target, and seven peripheral distractors. The items are geometric shapes in task 1 (square or circle as the targets and triangles as the distractors) and colors in task 2 (red or green as the targets and mixed colors as the distractors). Subjects are told to attend either to the central or to the peripheral targets and make responses to the target's shape in task 1 and to the target's color in task 2. Because the perception of a central target is easy, the competition involved in the center–center condition is primarily between the response selections of the two tasks. In the center–periphery condition, subjects have to shift attention from the center to the periphery. Finally, in the periphery–periphery condition, the two tasks compete not only for response selection but also for spatial attention to the periphery. Thus, while response selection is the main competition in the center–center condition, perceptual attention is an additional source for competition in the periphery–periphery condition. Will the neural substrates involved in resolving these two kinds of interference be different?