Toward a Unified Theory of the Multitasking Continuum:
From Concurrent Performance to Task Switching,
Interruption, and Resumption

Dario D. Salvucci
1

1
Department of Computer Science
Drexel University
3141 Chestnut St.
Philadelphia, PA 19104, USA
salvucci@cs.drexel.edu
Niels A. Taatgen
2,3

2
Department of Psychology
Carnegie Mellon University
5000 Forbes Ave.
Pittsburgh, PA 15213, USA
taatgen@cmu.edu
Jelmer P. Borst
3

3
Dept. of Artificial Intelligence
University of Groningen
Postbus 407
9700 AK Groningen, the Netherlands
jpborst@ai.rug.nl

ABSTRACT
Multitasking in user behavior can be represented along a
continuum in terms of the time spent on one task before
switching to another. In this paper, we present a theory of
behavior along the multitasking continuum, from concurrent
tasks with rapid switching to sequential tasks with longer
time between switching. Our theory unifies several
theoretical effects — the ACT-R cognitive architecture, the
threaded cognition theory of concurrent multitasking, and the
memory-for-goals theory of interruption and resumption —
to better understand and predict multitasking behavior. We
outline the theory and discuss how it accounts for numerous
phenomena in the recent empirical literature.
Author Keywords
Multitasking, attention, interruption, cognitive architecture.
ACM Classification Keywords
H.5.2 [Information Interfaces and Presentation] User
Interfaces – theory and methods; H.1.2 [Models and
Principles] User/Machine Systems – human factors, human
information processing.
INTRODUCTION
The modern world is a multitasking world — in all kinds of
environments and scenarios, people spend a great deal of
time engaged in multiple tasks at the same time. For
example, a recent study [16] found that employees of an
information-technology company spent an average of only
3 minutes per task before switching to another task. In
addition, user interfaces have rapidly spread from standard
desktop settings into real-world multitasking environments
due to the proliferation of mobile computing devices (e.g.,
the use of cell phones while walking or driving). All
together, these trends necessitate a better understanding of
the capacity and the limitations of human multitasking
performance: Only with such knowledge can we design and
build new user interfaces that complement, rather than
distract from or hamper, our daily lives.
The Multitasking Continuum
One rough but useful way of characterizing multitasking
behavior is in terms of the time spent on one task before
switching to another. This span, which we call the
multitasking continuum, is shown in Figure 1 along with
sample task domains at their approximate position along the
continuum. For example, on the left-hand side of the
continuum, talking while driving involves frequent switching
between tasks — with switches perhaps every second, if not
more often in normal conversation. On the right-hand side,
cooking while reading a book may involve fairly long spans
between task switches; for instance, one might start boiling
pasta, read for 10 minutes until the pasta is cooked, then
strain and prepare the pasta for a meal. Multitasking
behavior thus spans time scales at several orders of
magnitude, namely, as termed by Newell [31], the levels
corresponding to the Cognitive Band (10
-1
–10
1
s) and
Rational Band (10
2
–10
4
s) of human behavior.
The tasks on the left-hand side of the continuum could be
characterized as concurrent multitasking, in which the tasks
are, in essence, performed at the same time. There has been
a long and detailed research literature on concurrent
multitasking dating back at least to the 1930s [e.g., 42].
Some of the earlier work, which continues today, examines
concurrent performance of simple stimulus-response tasks
(e.g., in the dual-choice paradigm [32]). At the same time,
research has explored concurrent performance in a wide
variety of real-world tasks, from piloting [23] to driving [38]
to radar operation [43]. This empirical work has been
accompanied by theoretical and computational models of
concurrent multitasking [e.g., 24, 34, 36] that have aimed to
explain empirical phenomena with unifying ideas and
frameworks.


Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise,
or republish, to post on servers or to redistribute to lists, requires prior
specific permission and/or a fee.
CHI 2009, April 4–9, 2009, Boston, MA, USA.
Copyright 2009 ACM 978-1-60558-246-7/09/04…$5.00
CHI 2009 ~ Understanding UI 2 April 8th, 2009 ~ Boston, MA, USA
1819

At the same time, the tasks on the right-hand side of the
continuum could be characterized as sequential multitasking,
in which a longer time (say, minutes to hours) might be spent
on one task before switching to another. As for concurrent
multitasking, there has been a rich history of literature on
what we are calling sequential multitasking, broadly in the
areas of task switching, interruption, and resumption. Basic
psychological research [see 30] has primarily focused on the
“switch cost” (in time) encountered when switching between
tasks (using tasks of short duration but enforcing sequential
multitasking through the experimental paradigm itself).
More complex studies have examined analogous measures
for real-world human-computer interaction tasks [e.g., 1, 9,
12, 13, 22]. Again, like for concurrent multitasking,
researchers have developed conceptual and computational
models [e.g., 2, 14, 28] that attempt to make explicit the
sources of these switch costs and how they might be
mitigated in different scenarios and environments.
Considering that concurrent and sequential multitasking
represent different ranges on the same continuum, there has
been surprisingly little cross-fertilization between the
research in the two areas. In part, this separation between the
two areas has evolved for a good reason: Each area has
focused on distinct characteristics of behavior that are
interesting and warrant detailed study in their own right.
Nevertheless, because of their overlap on the multitasking
continuum, we strive for a unified theory of human
multitasking that is able to account for both concurrent and
sequential multitasking — that is, provide an account of
human behavior at all points along the multitasking
continuum.
Toward a Unified Theory of Multitasking
Our goal in this paper is to outline a unified theory of human
multitasking that spans both concurrent and sequential
multitasking. The theory incorporates three core
components: the ACT-R cognitive architecture [5, 6], which
provides a theory and computational framework for human
processing resources and their limitations; threaded
cognition theory [36], which provides an account of
concurrent performance for two or more arbitrary tasks; and
memory-for-goals theory [2], which provides an account of
task interruption and resumption based on activation and
recall of task goals in declarative memory. We use these
components as the building blocks of our theory, showing
how each contributes to the larger view of multitasking and
how the unification of these components helps to account for
a wide range of behavior across the multitasking continuum.
BEGINNINGS: CONCURRENT MULTITASKING
We begin by examining concurrent multitasking, which we
define as the execution of two or more tasks at the same
time. Our account of concurrent multitasking employs the
two component theories of ACT-R and threaded cognition.
We now briefly describe this theory, and then outline how
the theory accounts for basic phenomena related to
concurrent multitasking for both simple laboratory and
complex real-world tasks.
A Theory of Concurrent Multitasking
The ACT-R architecture [5] holds as a central assumption
that the best way to understand cognition at the functional
level is to consider it as a set of relatively independent but
interacting modules (or resources [44]). Some of the
modules deal with perception and motor activities, like
vision, audition, manual control, and speech. More
characteristic for ACT-R, though, is a set of central cognitive
modules. The declarative memory module serves as a
memory for factual knowledge, which also includes episodic
knowledge, and task instructions. The goal module
represents the current goal of the system, and can keep track
of progress and other state information. The problem
representation module (sometimes called the imaginal
module) holds partial representations needed by the task —
for example, an intermediate expression during algebraic
equation solving. Finally, the procedural module connects
all other modules together, using knowledge in the form of
condition-action production rules to control the flow of
information among modules. These production rules match
the state of information represented in other modules and
map these onto actions that are to be executed by the various
modules. For example, the visual module might read “1+2,”
after which the procedural module queries the declarative
memory module for the 1+2 addition fact. When the
declarative module produces the answer, the procedural
module would send this answer to the manual module, which
would then type “3” on the keyboard.

Figure 1: The Multitasking Continuum.
CHI 2009 ~ Understanding UI 2 April 8th, 2009 ~ Boston, MA, USA
1820