#essay

Attentional Limitations in Doing Two Tasks at Once The Search for Exceptions Mei-Ching Lien,1 Eric Ruthruff,2 and James C. Johnston3

1 Oregon State University,

2 University of New Mexico, and

3 NASA Ames Research Center

ABSTRACT—People generally have difficulty doing two

tasks at once. To explain this fact, theorists have proposed

that central processing—the thought-like stages following

perceptual encoding and preceding response processing—

takes place for only one task at a time. Because this bot-

tleneck imposes severe limits on human cognitive process-

es, research has attempted to find exceptions. There is now

solid evidence that, at least in the laboratory, the entire bo-

ttleneck can be completely bypassed under favorable com-

binations of circumstances. While these findings provide a

ray of hope for enabling parallel multitasking in real-

world scenarios, it will not be easy to take advantage of the

combination of conditions that appear to be necessary.

KEYWORDS—dual-task interference; central bottleneck

The question of whether humans can perform multiple tasks in

parallel has long been practically important for special popu-

lations—such as aircraft pilots and air traffic controllers—and

has been the subject of considerable psychological research. In

recent years, advancing technology has presented a wider seg-

ment of the population with multitasking challenges, most no-

tably talking on cell phones while driving. In years to come,

further challenges are sure to appear, such as using computer-

ized navigational aids while driving. In addition, complex

computerized systems—in everything from nuclear power plants

to spaceships—will become increasingly capable of far more

multitasking than human operators can easily keep up with.

Although many people believe they can parallel multitask,

laboratory studies have, with remarkably few exceptions, found

otherwise. Dual-task interference has been found with a wide

range of tasks, including very easy ones. These findings led to the

theory that central mental processing takes place for only one

task at a time (Welford, 1952). This ‘‘central bottleneck theory’’

has important implications both theoretically and practically.

Theoretically, the central bottleneck poses a mystery: Why

should the human brain, which contains hundreds of subregions

capable of working in parallel, act like a serial processor (i.e., a

single-processor von Neumann computer)? Practically, the

theory predicts that people will have trouble with real-world

situations requiring simultaneous performance of multiple tasks

(e.g., driving and talking).

In spite of the apparent generality of the central bottleneck, a

few studies have reported successful multitasking. It is impor-

tant to verify such reported exceptions. If true, they might pro-

vide a basis for promoting multitasking in the real world—

modifying either the technology (through design) or the user

(through training). For instance, could one eliminate interfer-

ence between talking on cell phones and driving by modifying

either phones or cars? The present paper discusses recent ad-

vances in the search for such exceptions to the central bottle-

neck and their implications for real-world scenarios.

ASSESSING DUAL-TASK INTERFERENCE

When assessing dual-task behavior, which measure of perfor-

mance should one emphasize: accuracy or response time? A

single-channel bottleneck will frequently cause response de-

lays, but it need not produce any consequences that would count

as an error. If you ask someone a question while they are typing a

message, they will not necessarily answer the question wrongly

or make a typographical error. However, they likely will pause,

causing a measurable delay in typing the message.

To measure response-time delays in dual-task situations,

laboratory studies have relied heavily on the psychological re-

fractory period (PRP) paradigm. This paradigm requires par-

ticipants to respond as quickly as possible to two tasks, Task 1

and Task 2, with a variable time between the stimulus onsets—

known as the stimulus onset asynchrony (or SOA). At long SOAs,

in which simultaneous work on both tasks is not required, one

Address correspondence to Mei-Ching Lien, Department of Psychol- ogy, Oregon State University, Corvallis, OR 97331; e-mail: mei. [email protected].

C U R R E N T D I R E C T I O N S I N P S Y C H O L O G I C A L S C I E N C E

Volume 15—Number 2 89Copyright r 2006 Association for Psychological Science

can measure the baseline response time to Task 1 (RT1) and Task

2 (RT2). Using response times obtained at the long SOA as a

baseline, one can then measure RT2 slowing at short SOAs,

which require the stimuli for both tasks to be processed simul-

taneously. The ubiquitous result is a pronounced lengthening of

RT2 at short SOAs (known as the PRP effect).

There is considerable evidence that PRP effects are due in

large part to a central-processing bottleneck (e.g., Pashler, 1992).

The key assumption, illustrated in Figure 1, is that Task-2 central

processing is delayed until Task-1 central processing has fin-

ished. To facilitate intuitive reasoning about the hypothesized

bottleneck, Pashler used the analogy of a bank teller who can

handle only one customer at a time. If two customers arrive in

close succession, the second will experience a ‘‘bottleneck delay.’’

THE SEARCH FOR EXCEPTIONS TO THE CENTRAL

BOTTLENECK

To gain insight into why a bottleneck occurs, it is useful to de-

termine when it does not occur. Accordingly, researchers have

searched for special conditions, such as high similarity between

stimulus and response, having practiced the tasks to a high

degree, and the use of special response subsystems (e.g., eye

movements), that might allow bottleneck bypassing.

Although this search seems straightforward, a problem is

lurking. Tricks to promote bottleneck bypassing do so, in one

way or another, by making the tasks easier, which inevitably

shortens stage durations. But shortening the Task-1 central-

processing stage can dramatically reduce, or even eliminate, the

PRP effect without actually bypassing the bottleneck. In the

extreme scenario shown in Figure 2, the bottleneck limitation is

still present but is ‘‘latent’’—that is, it has no observable effect

on performance (Ruthruff, Johnston, Van Selst, Whitsell, &

Remington, 2003). Van Selst, Ruthruff, and Johnston (1999)

estimated that a latent bottleneck can occur with mean RT1s of

200 to 400 milliseconds—just the range of RT1s found in recent

efforts to demonstrate bottleneck bypassing.

Failure to appreciate the possibility that a bottleneck may be

present but latent has led some researchers to equate the ab-

sence of observable bottleneck delays with the absence of the

underlying bottleneck limitation. Careless reasoning may be

invited by the common intermingling of two different senses of

the word ‘‘bottleneck’’: the bottleneck time delay and the un-

derlying bottleneck limitation. As a physical analogy of a bot-

tleneck, consider traffic crossing a one-lane bridge. Spacing out

traffic could entirely eliminate bottleneck delays, but would not

eliminate the underlying bottleneck limitation: Still only one car

can pass at a time.

OVERCOMING ROADBLOCKS TO DIAGNOSING

BOTTLENECK BYPASS

So far, it might appear that we have reached an impasse: Trying to

bypass the bottleneck by making tasks easier also makes it

(a) Short SOA

(b) Long SOA

Response Processing

Central Processing

Perceptual Encoding

RT1

RT2SOA

Task 2

Response Processing

Central Processing

Perceptual Encoding

RT1

RT2SOA

Central Processing

Perceptual Encoding

Response Processing

Task 1

Central Processing

Perceptual Encoding

Response Processing

Task 2

Task 1

Fig. 1. The central bottleneck model. Task-2 central processing (e.g., response selection, memory retrieval, etc.) does not begin until Task-1 central processing is completed, resulting in a period of cognitive slack (dotted line in diagram a) when there is a short stimulus-onset asynchrony (SOA), but not when there is a long SOA (diagram b). The central stages are shaded. RT1 5 response time for Task 1; RT2 5 response time for Task 2.

90 Volume 15—Number 2

Attentional Limitations in Doing Two Tasks at Once

harder to tell whether the bottleneck was in fact bypassed.

Fortunately, recent research has shown several ways around this

impasse. First, there are a number of supplementary empirical

tests (beyond PRP size) for the presence of a bottleneck. For

instance, trials that produce longer RT1s should also produce

longer RT2 delays, leading to a positive correlation of RT1 and

RT2. Also, whereas serial central processing (Task 1 then Task

2) should produce serial responding in the same order, parallel

central processing should often produce response reversals

(Task 2 then Task 1). These different tools often produce

agreement about whether a bottleneck is present or absent.

Recent studies have hit on a different trick. Logically, by-

passing the central bottleneck should not require making both

tasks easy. If either Task 1 or Task 2 does not require the central-

processing mechanism, that should be enough. Making Task 1

easy shortens RT1, which can eliminate interference regardless

of whether or not the bottleneck is bypassed. Making Task 2 easy,

however, shortens RT2 without inviting a latent bottleneck; as

long as RT1 is relatively long, a bottleneck would still produce

PRP effects. Thus, a promising strategy is to look for bottleneck

bypassing with an easy Task 2 but not an easy Task 1.

APPLYING BOTTLENECK DIAGNOSTICS

Ideomotor Compatibility

Greenwald and Shulman (1973) investigated what they called

ideomotor-compatible tasks, for which the ‘‘stimulus resembles

sensory feedback from the response’’ (p. 70). An example would

be responding to an auditory word by speaking the same word

(e.g., say ‘‘high’’ when you hear ‘‘high’’) or moving a joystick in

the direction of an arrow. The hope is that such responses can

reuse the mental codes used to represent the stimuli, eliminating

the need for central processes. Greenwald and Shulman reported

eliminating the PRP effect and therefore concluded that ideo-

motor-compatible tasks bypass the central bottleneck.

This plausible conclusion was widely accepted for decades,

but it has recently been challenged (Lien, McCann, Ruthruff, &

Proctor, 2005; Lien, Proctor, & Allen, 2002). We (Lien et al.,

2005) studied different combinations of ideomotor-compatible

tasks and other tasks. PRP delays were substantial when only

Task 2 was ideomotor compatible but declined when Task 1 was

also ideomotor compatible. This decline, however, was similar to

the decline in RT1, just as a bottleneck model would predict.

Note that Greenwald and Shulman (1973) did not explicitly

consider the latent-bottleneck hypothesis. The data showed

several other indications of a bottleneck, including a strong

RT1–RT2 correlation. Furthermore, we (Lien et al., 2005)

showed that computer simulations of a processing bottleneck

could reproduce all the critical data trends.

Although ideomotor-compatible tasks apparently do not gen-

erally bypass the bottleneck, certain special cases might.

Johnston and Delgado (1993) carried out PRP experiments for

which Task 1 was judging whether a tone was high or low pitched

and Task 2 was a special analog tracking task—requiring par-

ticipants to keep a circle (whose position was controlled with a

joystick) over a moving stimulus cross. In such a case, virtually no

PRP interference was found, supporting the absence of a central

bottleneck. The finding of frequent response reversals (Task 2

response before Task 1 response) confirmed this conclusion.

Johnston and Delgado proposed that the tracking task had ‘‘pre-

authorized’’ joystick responses to occur as needed without the

usual central approval. Interestingly, the bottleneck returned

when the joystick task was Task 1: Responses to a tone (Task 2)

following the cross movement (Task 1) showed large PRP delays.

Practice

Does extensive practice on a task allow it to become ‘‘automa-

tized’’ and thus bypass the central bottleneck? Several studies

from the 1970s seemed to support this hypothesis. Spelke, Hirst,

and Neisser (1976) found that, after 6 weeks of practice, par-

ticipants had no difficulty reading stories while accurately

transcribing spoken words. In addition, practice was found to

dramatically improve search for a target on a screen (Schneider

& Shiffrin, 1977). These studies do not, however, prove that

practice eliminates dual-task interference. Spelke et al. (1976)

focused on accuracy measures, which, as noted earlier, can be

insensitive to dual-task interference. The visual-search findings

suggest that certain perceptual processes can operate in parallel

after practice, but they do not establish that central processes

capable of commanding actions can occur in parallel.

A more rigorous assessment of practice effects is possible with

the PRP paradigm. Early PRP studies found that practice not

only failed to eliminate the bottleneck but barely even reduced

its duration. Van Selst et al. (1999) showed that this curious

result was an artifact of requiring manual responses in both

tasks. With separate response modalities (manual and vocal),

practice does reduce bottleneck delays. Other aspects of the data

suggested that a residual bottleneck was still present, albeit

reduced in size because mean RT1 was so short.

In some cases, practice appears to eliminate the central bot-

tleneck entirely (e.g., Hazeltine, Teague, & Ivry, 2002). Ruth-

Response Processing

Central Processing

Perceptual Encoding

Perceptual Encoding

Central Processing

Response Processing

RT1

RT2SOA

Task 2

Task 1

Fig. 2. A latent central bottleneck. When response time for Task 1 (RT1) is very short, Task-1 central processing might finish before Task-2 central processing is ready to begin. If so, a central bottleneck would not delay Task 2, even at a short stimulus onset asynchrony (SOA). (RT2 5 response time for Task 2.)

Volume 15—Number 2 91

Mei-Ching Lien, Eric Ruthruff, and James C. Johnston

ruff, Van Selst, Johnston, and Remington (in press) had partic-

ipants practice one task alone for 8 sessions and then perform it,

along with another task, in a PRP design. They reported several

converging indications of bottleneck bypass, at least for a mi-

nority of participants. Interestingly, though, bypassing occurred

primarily when participants had practiced the easier of the two

tasks (a tone judgment) and this task served as Task 2.

Special Response Systems

Almost all of what we know about the central bottleneck comes

from tasks with manual or vocal responses. Since the hands and

voice are ‘‘general-purpose instruments,’’ perhaps those re-

sponses are normally controlled by central processes (subject to

the bottleneck), whereas narrow-purpose response systems (e.g.,

the eyes) are not.

Pashler, Carrier, and Hoffman (1993) found that focusing the

eyes on an object as Task 2 in a PRP study bypassed the bot-

tleneck. The converging lines of evidence included small PRP

effects despite long RT1s, weak RT1–RT2 correlations, and

response reversals. Importantly, these results were found with-

out any attempt to make Task 1 easy (i.e., using a Task 1 that

produced a bottleneck in previous research).

Eye movements might bypass the bottleneck because they are

a quasi-reflexive action, possibly supported by special neural

circuitry (bypassing general-purpose central resources). An-

other possibility, however, is that looking at a stimulus of interest

is a highly-practiced action and it is the high practice levels that

allowed bottleneck bypassing.

IMPLICATIONS FOR DUAL-TASK THEORY

The studies discussed above suggest three main conclusions

with implications for dual-task theory. First, complete bottle-

neck bypassing, albeit rare, is in fact possible under favorable

conditions. It has now been observed with an eye-movement

task, a tracking task, and with a highly practiced tone judgment.

Why did these particular tasks bypass the bottleneck when

others did not? The key might be ‘‘preapproval’’ of the required

response. Eye movements, for instance, are at low risk to conflict

with other actions and might have ‘‘blanket’’ preapproval; typi-

cally, the eyes can track objects of interest without any conscious

command to do so. It is plausible that analog tracking would

encourage the same mental set. If this hypothesis is correct,

training regimens that encourage preapproval of important tasks

may improve dual-task performance.

Second, complete bottleneck bypassing is possible even when

only one task is easy. Bypassing appears to be more likely with an

easy Task 2 than with an easy Task 1. Why? Although automa-

tized tasks do not need central resources, they might ‘‘greedily’’

use those resources anyway, if available (see Lien et al., 2005;

Ruthruff et al., in press). Consider two bank customers: A, who

must use the human teller; and B, who can use either the teller or

the ATM machine. If customer A appears first and occupies the

teller, customer B can use the ATM and avoid a bottleneck delay.

But if customer B appears first, he or she may ‘‘greedily’’ use the

available live teller, delaying customer A who must wait for the

teller.

Third, complete bottleneck bypassing is rare. Hundreds of

studies have reported the presence of a processing bottleneck,

whereas only a handful of studies have reported the absence of a

bottleneck. Even with highly-compatible or highly-practiced

tasks, processing bottlenecks are often reported. So, for any new

situation, the default assumption is that a processing bottleneck

will be encountered.

IMPLICATIONS FOR REAL-WORLD SCENARIOS

Given that bottleneck delays are the rule in the laboratory, with

simplified tasks and simplified paradigms, they may be the rule

in the real world as well. In a driving simulation, Levy, Pashler,

and Boer (2006) found that even a task as simple and highly

practiced as braking was slowed markedly in dual-task condi-

tions. Tasks like driving and talking on a cell phone might be

impossible to fully automatize with practice because they con-

tain many different subtasks and because too much of the

processing is nonrecurring—one can hardly expect drivers to

have previously encountered every possible scenario. Therefore,

reports of substantial interference between these activities

(Strayer & Johnston, 2001) will likely generalize widely. Al-

though reported exceptions to the central bottleneck provide a

ray of hope, it remains to be seen whether any combination of

device design and human training can make multitasking a re-

liable real-world phenomenon.

DIRECTIONS FOR FURTHER RESEARCH

The present review discussed several well-documented excep-

tions to the central bottleneck. Further research is needed to

determine precisely which conditions are critical for producing

these exceptions and whether they can be successfully imple-

mented in real-world scenarios. In particular, more research is

needed to determine which forms of practice provide the most

reliable path to bypassing the bottleneck. There is also a clear

need for research into whether practice enables real-world tasks,

which are more complicated than typical laboratory tasks, to

bypass the bottleneck.

As the search for exceptions continues, it is important to learn

from past mistakes and concentrate research efforts on condi-

tions capable of diagnosing bottleneck bypassing. The studies

reviewed above suggest trying to bypass the central bottleneck

by making Task 2 easy, but not Task 1. This approach has been

shown to facilitate bottleneck bypass while sidestepping the

latent-bottleneck problem. It is quite fortunate that bypassing is

(apparently) most likely under the very conditions (easy Task 2)

92 Volume 15—Number 2

Attentional Limitations in Doing Two Tasks at Once

in which it is easiest to detect. This may be a case where the keys

are, in fact, right under the lamppost.

Recommended Reading Logan, G.D., & Gordon, R.D. (2001). Executive control of visual at-

tention in dual-task situations. Psychological Review, 108, 393– 434.

Meyer, D.E., & Kieras, D.E. (1997). A computational theory of execu-

tive cognitive processes and multiple-task performance: Part 2.

Accounts of psychological refractory-period phenomena. Psy- chological Review, 104, 749–791.

Acknowledgments—This work was funded by the College of

Liberal Arts at Oregon State University and NASA Grant NCC 2-

1325.

REFERENCES

Greenwald, A.G., & Shulman, H.G. (1973). On doing two things at once:

II. Elimination of the psychological refractory period effect.

Journal of Experimental Psychology, 101, 70–76.

Hazeltine, E., Teague, D., & Ivry, R.B. (2002). Simultaneous dual-task

performance reveals parallel response selection after practice.

Journal of Experimental Psychology: Human Perception and Per- formance, 28, 527–545.

Johnston, J.C., & Delgado, D.F. (1993, November). Bypassing the single- channel bottleneck in dual-task performance. Paper presented at the annual meeting of the Psychonomic Society, Washington, DC.

Levy, J., Pashler, H., & Boer, E. (2006). Central interference in driving:

Is there any stopping the psychological refractory period? Psy- chological Science, 17, 228–235.

Lien, M.-C., McCann, R.S., Ruthruff, E., & Proctor, R.W. (2005). Dual-

task performance with ideomotor-compatible tasks: Is the central

processing bottleneck intact, bypassed, or shifted in locus? Jour- nal of Experimental Psychology: Human Perception and Perfor- mance, 31, 122–144.

Lien, M.-C., Proctor, R.W., & Allen, P.A. (2002). Ideomotor compati-

bility in the psychological refractory period effect: 29 years of

oversimplification. Journal of Experimental Psychology: Human Perception and Performance, 28, 396–409.

Pashler, H. (1992). Attentional limitations in doing two tasks at

the same time. Current Directions in Psychological Science, 1, 44–48.

Pashler, H., Carrier, M., & Hoffman, J. (1993). Saccadic eye movements

and dual-task interference. Quarterly Journal of Experimental Psychology, 46, 51–82.

Ruthruff, E., Johnston, J.C., Van Selst, M.V., Whitsell, S., & Remington,

R. (2003). Vanishing dual-task interference after practice: Has the

bottleneck been eliminated or is it merely latent? Journal of Ex- perimental Psychology: Human Perception and Performance, 29, 280–289.

Ruthruff, E., Van Selst, M., Johnston, J.C., & Remington, R.W. (in

press). How does practice reduce dual-task interference: Inte-

gration, automatization, or simply stage-shortening? Psychologi- cal Research.

Schneider, W., & Shiffrin, R.M. (1977). Controlled and automatic hu-

man information processing: 1. Detection, search, and attention.

Psychological Review, 84, 1–66.

Spelke, E.S., Hirst, W., & Neisser, U. (1976). Skills of divided attention.

Cognition, 4, 215–230.

Strayer, D.L., & Johnston, W.A. (2001). Driven to distraction: Dual-task

studies of simulated driving and conversing on a cellular tele-

phone. Psychological Science, 12, 462–466.

Van Selst, M., Ruthruff, E., & Johnston, J.C. (1999). Can practice

eliminate the psychological refractory period effect? Journal of Experimental Psychology: Human Perception and Performance, 25, 1268–1283.

Welford, A.T. (1952). The ‘‘psychological refractory period’’ and the

timing of high-speed performance: A review and a theory. British Journal of Psychology, 43, 2–19.

Volume 15—Number 2 93

Mei-Ching Lien, Eric Ruthruff, and James C. Johnston