TAMODIA 2004 | PAPERS 15-16 November | Prague, Czech Republic
17
Towards A Composite Modelling Approach for
Multitasking
Peter J Wild, Peter Johnson and Hilary Johnson
HCI Group, Department of Computer Science
University of Bath, Bath
BA2 7AY, UK. Tel: +44 1225 386811
Email: {pwild, p.johnson, h.johnson}@cs.bath.ac.uk

ABSTRACT
Much information and knowledge work (with and
without information technology) can be characterised as
multitasking and interrupt driven. A whole host of
characterisations and buzzwords imply an increase in the
number of roles, tasks/activities, IT artefacts,
interruptions and exceptions that people have to deal
with. This provides a challenge for Task Analysis
approaches as they have historically focussed around
single tasks and users.
A preliminary version of a composite modelling approach
(the Composite Multitasking Model) is presented that
draws from approaches that model task, events,
interruptions, exceptions and the temporal aspects of
tasks. As well consideration of how information about
multiple tasks is elicited, we apply the approach to the
modelling of data from our own studies.
Keywords
Tasks, multitasking, exceptions interruptions, events,
goals.
INTRODUCTION
Advances in Information Technology (IT) provide greater
enabling support for carrying out individual tasks in a
parallel or interleaved fashion. Once we would have to
wait around for a desktop computer to complete printing
a document and its limited memory meant we could only
open one application at a time. Now mainstream
computing has enough power to provide greater
flexibility about how we arrange our work. The
convergence of telecommunications and IT has brought
separate tasks and artefacts such as document
preparation, fax, phone, into one electronic workspace
(e.g., desk- / lap-top computer, personal digital assistant
or mobile phone). Through the use of automation, tasks
that were conceived to be performed separately are
integrated into supervisory task allocation schemes.
These technological shifts both encourage the
performance of traditionally specialist tasks by
generalists (e.g., document production). Resulting in the
potential for people to undertake a greater amount of, and
a wider variety of work, and the possibility of greater
interleaving of goals and sub goals [1, 2, 5]. As well as
technological shifts, there have been shifts towards
economies where services and knowledge work form a
larger part of the work carried out. In turn how, when
and by whom tasks are carried, out are less constrained
by the physical world. Furthermore, as the volume of
tasks increases, the management of tasks itself, becomes
a recurring and important task. Overall these shifts
amount to what some refer to as interrupt-driven [11]
multitasking [1], or polychronic behaviour [16].
However, technological and artefactual task support is
not generally designed to take account of phenomena
such as multiple tasks, multiple instances of the same
task, interruptions, exceptions, tasks with multiple
participants, and multiple ways of ordering the sequence
of subtasks and opportunism in dynamic environments.
We cannot assume that because we have designed a task
well from a single user - single task perspective, that it
will necessarily offer effective support for such
phenomena. At best, IT artefacts may be inefficient, at
worst inflexible, frustrating, error causing, and ultimately
unusable.
As a step towards design to support we need to be able to
model people multitasking. When modelling people
multitasking, we face the “classical” difficulty of
modelling the task and its context, but also the interaction
and potential clashes between tasks and contexts. In turn
this presents a challenge for relevant modelling
approaches - whether of task / activity, domain, tools,
event or interruption. Each modelling approach, by being
predominantly focussed around one issue, misses other
important facets relevant to the performance of multiple
tasks. Task modelling tends to model one person, one
task; domain modellers only seem to assume that a
person works within one domain; work on events,
interruptions and temporal modelling is aware of the
polychronic nature of work (i.e., lots of things happening
at once) but have little representation of tasks. We also
need to go beyond yet another set of primitives for
modelling to something that attempts to tie together both
the needed primitives, the theoretical reasoning behind
the approaches and cross referencing between the
approaches.
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.
TAMODIA’04, Prague, Czeck Republic.
Copyright ©2004 ACM 1-59593-000-0/04/0011…$5.00

TAMODIA 2004 | PAPERS 15-16 November | Prague, Czech Republic
18
Starting To Unpack Multiple Tasks
Multitasking, in common parlance has come to mean
doing many things at once. When understanding the
multitasking it is useful to scope the multiple “what” and
the overall duration of the “when.” Traditionally in
cognitive psychology multitasking along with task
switching has relied on “tasks” that are easily specified,
learnt and performed in laboratory contexts and
undertaken in a few seconds [e.g., 12]. We refer to this
type of multitasking as micro-multitasking. Whilst a
powerful paradigm for understanding the constraints on
cognitive processing, in HCI contexts tasks have
attendant ambiguities about their outcome, multiple
participants and roles, mixed resources, as well as being
broader in scope and longer in duration. Accepting the
strong suggestion in the literature that performing even a
single simple task involves parallel and interleaved
cognitive processes (i.e., micro-multitasking); we shift
our attention to the forms of multitasking discussed in the
paper’s introduction. With regards to the “once” element
we refer to forms of parallel and interleaved tasks that are
undertaken over longer time scales: minutes, hours, days
and months. These involve tasks that are open ended,
that converge or diverge in relation to the dynamics of the
environment. Or require collaboration and cooperation
between multiple agents. This is the form of multitasking
we refer to in this paper. This is in contrast to the fully
specifiable task traditional in laboratory studies of micro-
multitasking or present in earlier HCI studies.
The Need for a Composite Modelling Approach
Previous reports of our efforts in this area have suggested
extension to a specific task analysis method. Since then
we have come to the view that what is need is a
composite approach. Tasks remain an important thread
running through our approach, but the approach also
draws on concepts from models of events [4], exceptions,
interruptions [10] and the temporal aspects of work [8,
16].
We have three reasons for a composite approach. Firstly,
no one approach can be seen as complete. Secondly it is
rare in HCI that approaches are combined, and
synthesized [c.f., 6]. Yet the challenge of modelling tasks
has increased because of the shifts in technology and
work practice alluded to in the introduction. Finally,
design attempts have often not been based on models; in
part this is due to the lack of unified approach to
modelling multitasking. We put forward our nascent
approach as a first step towards this.
TASK ANALYSIS AND MULTITASKING
Task analysis methods have slowly enriched the set of
modelling primitives available, for example, sub role,
agent, task type [13, 14]. And in some cases deepened
the underlying theory [7, 18]. But in this context we have
a number of concerns about Task Analysis (TA) methods.
The first is that it is rare to see a task analysis focus on a
set of different tasks. Task models are predominantly put
forward as generic models of a particular single task-
class. But being able to account for the core task of a role
is different form understanding how the whole milieu of
tasks that a person performs interrelate and interact.
Another concern is that they are often normative,
abstracting away from the complications and
idiosyncrasies of real world task performance, such as
events, interruptions and exceptions. Whilst individual
work has looked into these issues [4, 10] they have not
been integrated into other approaches.
Our next concern is with their ability to express between-
task relationships. Many task analysis approaches
represent temporal relationships that are internal to
specific task for example serialization, choice,
parallelism, interleaving [7, 14]. The key observation
that can be made of temporal issues in task modelling is
that the temporal representations have only been applied
to model temporal relations within a task, not between
tasks. While they can capture and represent some aspects
of switching between tasks, they miss other aspects, such
as what higher level goals “force” the switch and what
contextual cues enable or act as a stimulus for the switch.
Furthermore, contemporary within-task task temporal
representations neglect the temporal contexts as
experience as experienced by people. Failing to
recognise basic distinctions such as a task is performed at
9.30am and a task is usually performed at 9.30am [16].
SEEDS OF THE APPROACH: WHAT SHOULD BE
MODELLED?
A number of studies of multitasking have been
undertaken by ourselves [17, 18] and others [2, 5].
Across our own studies [17, 18] the following
phenomena were replicated:
• linearization and switching [1, 2], where people
perform a seeming continuous stream of activity across
a number of task by switching between a set of tasks.
• temporal gaps / lags / natural breaks [4], where the task
structure has periods where no activity can be
undertaken by the person. For example when awaiting
information or permission.
• interruptions [10, 11], where the task receives an
interruption, whether from the agent performing the
task, or other participant in the task environment.
• Goals and roles are a traditional part task modelling
[7]. We and others [2, 5] have observed users with
multiple tasks, goals and roles.
New phenomena generated by our own studies:
• Layered goal and task execution choices. We can infer
higher level goals than usually modeled by TA
methods. These higher level goals can affect how
specific task instance is executed. For example,
whether to use email, phone or face to face
conversation to discuss tasks with a person. Or in how
we handle events.
• Groupings [18], where tasks or subtask are grouped
according to contextual phenomena such aslocation,
participant or deadline.
• Single events and multiple statuses, rather thanan event
having one status (i.e., meaning) ascribed to it, multiple