You are on page 1of 14

31

Human Cognitive Architecture


John Sweller
University of New South Wales, Sydney, Australia

CONTENTS
Introduction .....................................................................................................................................................................370
Human Cognitive Architecture .......................................................................................................................................370
The Information Store Principle and Human Long-Term Memory .....................................................................371
Origins, Evidence, and Implications............................................................................................................371
The Borrowing Principle and Transferring Knowledge .......................................................................................371
Origins, Evidence, and Implications............................................................................................................372
Randomness as Genesis Principle and Creating Knowledge ...............................................................................372
Origins, Evidence, and Implications............................................................................................................372
The Narrow Limits of Change and Human Working Memory ............................................................................ 373
Origins, Evidence, and Implications............................................................................................................373
The Environment Organizing and Linking Principle............................................................................................373
Origins, Evidence, and Implications............................................................................................................373
Cognitive Load Theory ...................................................................................................................................................374
Cognitive Load Theory Effects and Technology-Based Instruction.....................................................................374
Worked Example Effect ...............................................................................................................................374
Split-Attention Effect ...................................................................................................................................375
Modality Effect ............................................................................................................................................375
Redundancy Effect .......................................................................................................................................376
Expertise Reversal Effect.............................................................................................................................377
Guidance Fading Effect ...............................................................................................................................377
Imagination Effect........................................................................................................................................378
Element Interactivity Effect .........................................................................................................................378
Isolated Interacting Elements Effect............................................................................................................379
Discussion........................................................................................................................................................................379
Acknowledgments ...........................................................................................................................................................380
References .......................................................................................................................................................................380

369
John Sweller

ABSTRACT institutions. Observing and interacting with the world


would provide the best educational environment.
Cognitive load theory integrates the origins of human Under such conditions, all instruction would have to
cognition in evolutionary theory with the structures be as realistic as possible and would only be required
and functions of human cognitive architecture to pro- for economic or safety reasons rather than for educa-
vide effective instructional design principles. Many of tional reasons. In contrast, it is suggested in this chap-
those principles are directly relevant to instructional ter that, because of the characteristics of human cog-
technology. This chapter outlines the evolutionary nition, in many conditions learning is best facilitated
bases for human cognitive architecture, examines by instruction that does not accurately model reality.
those aspects of human cognition that are directly rel- Cognitive load theory can be used to determine
evant to instruction, and discusses the various instruc- some of the characteristics of effective instruction.
tional principles generated by cognitive load theory This integrated theory is intended to provide a system-
with specic reference to their applicability to instruc- atically organized hierarchy leading from evolutionary/
tional technology. biological reasons for the characteristics of human
cognitive architecture to the instructional conse-
quences that ow from that architecture. Only some
KEYWORDS of the instructional procedures generated by cognitive
load theory are relevant to educational technology, and
Cognitive load theory: An instructional design theory those procedures are emphasized in this chapter. We
based on our knowledge of human cognitive archi- begin by discussing human cognitive architecture
tecture. within an evolutionary and biological framework.
Human cognitive architecture: The manner in which
structures and functions required for human cog-
nitive processes are organized. HUMAN COGNITIVE ARCHITECTURE
Long-term memory: The store holding all knowledge
acquired during the processes of learning. Human cognition has evolved to assimilate, process,
Natural information processing system: The proce- and use information (or knowledge, used synony-
dures by which natural systems such as human mously in this chapter when dealing with human cog-
cognition and evolution by natural selection pro- nition) to direct human action. It constitutes an exam-
cess information. ple of a natural information processing system that is
Working memory: The structure that processes infor- a class of information processing systems that can be
mation coming from either the environment or found in nature. As a natural information processing
long-term memory and that transfers learned infor- system, human cognition is hardly likely to be unique
mation for storage in long-term memory. and, indeed, evolution by natural selection can itself
be classed as a natural information processing system
(Sweller, 2003, 2004; Sweller and Sweller, 2006). The
INTRODUCTION characteristics of such systems will vary depending on
their functions, but all natural information processing
The extent to which any instruction is effective systems share an identical basic structure or frame-
depends heavily on whether it takes the characteristics work. That basic structure in turn can be used to deter-
of human cognition into account. To determine the mine how humans deal with information and what
conditions that maximize learning, we need to closely types of instructional procedures, including technol-
study human cognition. Once we have established the ogy-dependent instruction, are likely to be effective.
mechanisms of human cognition, including why those The essential characteristics of a natural informa-
mechanisms have their particular characteristics, we tion processing system include: (1) a very large store
are in a position to design learning environments in of information that allows the system to function in
accord with human cognitive architecture. Ideal learn- the varied environments faced by natural information
ing environments in accord with human cognitive processing systems; (2) procedures for perpetuating
architecture are not always in accord with realistic the store of information by transferring information
learning environments that mimic the real world. from one entity to another; (3) procedures for changing
Indeed, if human cognition was organized in a manner the store by creating new information to deal with a
that always favored learning in realistic over articial changing environment; (4) procedures to ensure that
environments, there would never be a need for special- changes to the store do not destroy its effectiveness;
ized instructional procedures or even for educational and (5) procedures to relate information to the external

370
Human Cognitive Architecture

world. These core characteristics as applied to human them. Such skills are only useful in a mathematical
cognition are discussed here in terms of ve principles context. Second, the work on noviceexpert differ-
(Sweller and Sweller, 2006). ences explains why it takes so long to become an
expert in a substantial eld. It takes about 10 years of
The Information Store Principle concentrated work for a person to become a chess
and Human Long-Term Memory grand master (Simon and Gilmartin, 1973). During that
time, the person is not learning complicated, general
Human long-term memory provides the human cogni- problem-solving strategies. There is no evidence that
tive system with the large store of information indis- learnable or teachable general problem-solving strate-
pensable to a natural information system. A species gies exist; rather, the chess player is learning to rec-
genome has a similar function in biological evolution. ognize the huge number of board congurations
Our knowledge of the function of long-term memory required for chess expertise, just as all of us are
has altered over time, and, arguably, specifying the role required to learn the huge number of problem situa-
of long-term memory constitutes the primary nding of tions required to acquire problem-solving skill in a
the cognitive science revolution. We no longer see long- particular area. Third, data on noviceexpert differ-
term memory as a repository of isolated, unrelated facts ences demonstrate the central importance of knowl-
that are occasionally stored and retrieved; instead, it is edge or information held in long-term memory to skill
the central structure of human cognitive architecture. in any area (Chi et al., 1982). To be skillful, we must
have a huge stock of knowledge available. Based on
Origins, Evidence, and Implications this conceptualization, long-term memory is critical to
skilled performance. Knowledge held in long-term
De Groots (1946/1965) work on expertise in the game memory allows us to function in the variety of contexts
of chess can be seen as the initial work transforming in which we nd ourselves in much the same way as
our perception of the role of long-term memory. De a genome allows a species to function biologically.
Groot found that, if chess masters are shown a board Accordingly, a major function of instruction, including
conguration taken from a real game for about 5 sec- technology-based instruction, is to ensure that appro-
onds, they can reproduce it much more accurately than priate knowledge is held in long-term memory.
weekend players. Chase and Simon (1973) reproduced
this nding but also found that masters and weekend The Borrowing Principle
players did not differ in their ability to reproduce ran- and Transferring Knowledge
dom board congurations. This difference between
experts and novices in memory for real congurations Once discovered, information held by natural informa-
and situations has been replicated in a variety of elds tion processing systems must be perpetuated, and a
(Egan and Schwartz, 1979; Jeffries et al., 1981; primary function of education is to ensure that knowl-
Sweller and Cooper, 1985). Furthermore, it is the only edge is not lost because, as will be discussed below,
reliable difference that has been obtained differentiat- there are structural reasons why the act of discovery
ing novices and experts in problem-solving skill and is inordinately difcult. In genetics, the store of DNA-
is the only difference required to fully explain why an based information is perpetuated over long periods of
individual is an expert in solving particular classes of time by sexual and asexual reproduction. Asexual
problems. A chess master has learned to recognize reproduction includes precise copying from one gen-
many thousands of board congurations. When faced eration to the next and appears to have no cognitive
with a conguration, he or she recognizes it and knows equivalent. In contrast, sexual reproduction is a con-
the best move to make given that conguration. We all structive procedure designed to ensure that offspring
have that skill in our own areas of expertise and, differ from their parents. This process has an inevitable
indeed, the ability to recognize situations and the random component, with the precise combination of
appropriate actions that they require constitutes our genetic material from male and female ancestors being
skill base in a given area. intrinsically unpredictable.
Several implications ow from these ndings. Psychological mechanisms are required equally to
First, problem-solving skill is domain specic. A bril- preserve information held in long-term memory by its
liant mathematician has acquired mathematical prob- transmission from individual to individual. Accordingly,
lem-solving skills that are unlikely to transfer to, for we have evolved to both efciently transmit and receive
example, nancial or personal relationship skills. The information from other humans in either auditory or
skills allow a mathematician to recognize mathemati- visual form. We imitate what others do, listen to what
cal problem states and the best moves associated with they say, and read what they write. That skill permits

371
John Sweller

knowledge held in an individuals long-term memory to quate. The procedure for changing the store of infor-
be perpetuated indenitely. Nevertheless, as is the case mation held by natural information processing sys-
with sexual reproduction, the process is constructive in tems is standard. All creative changes are random,
nature. We combine new information with information but only effective changes are retained, and ineffec-
previously stored in long-term memory (Bartlett, 1932). tive changes are jettisoned. In evolution by natural
As is also the case with sexual reproduction, the process selection, this mechanism is well accepted. Random
has a random component in that we cannot predict pre- mutation is the initial source of all genetic variation
cisely how the two sources of information will be com- with only successful mutations resulting in reproduc-
bined. This randomness has implications for the narrow tion and continuation. In psychology, and especially
limits of change principle discussed below. instructional psychology, the suggestion requires fur-
ther explanation.
Origins, Evidence, and Implications Random generation followed by effectiveness test-
ing is unavoidable in functioning, natural information-
A major function of instruction is to organize efcient processing systems. Consider a person solving a novel
procedures that will permit knowledge to be trans- problem, an activity intended to create a solution that
ferred to the long-term memories of learners. Cogni- is new for that person. When deciding on a problem-
tive load theory has generated a variety of instructional solving move or a series of moves, the two basic cat-
procedures relevant to technology-based instruction egories of move generators are random generation and
(see below). All are dependent on the borrowing prin- generation based on previous knowledge. These two
ciple. The success of cognitive load theory in gener- categories or a combination of these two categories are
ating instructional procedures is, at least in part, due the only source of move generation. Previously
to its emphasis on learning from presented informa- acquired knowledge held in long-term memory can act
tion. That success provides some of the evidence for as a central executive determining moves, but, of
the borrowing principle. course, that knowledge is not always available. There
In earlier eras, the suggestion that a function of is no general central executive that can determine
education is to transfer knowledge would have been moves in the absence of relevant knowledge. Random
considered self-evident. More recently, an emphasis generation, either mentally or in the real world, is left
on discovery and constructivist procedures encouraged as the only other alternative.
learners to discover knowledge rather than to have
instructors transmit knowledge. The impossibility of Origins, Evidence, and Implications
discovering even a tiny fraction of the huge amount of
information required in the modern world was partially The origin of this principle is largely logical rather than
obscured by the elds ignorance of the massive size empirical. If knowledge to determine a problem-solv-
of long-term memory. That ignorance is being rectied ing move is unavailable, random generation followed
and increasing numbers of investigators are reacting by tests for effectiveness is the only other option. We
against the previous orthodoxy (Kirschner et al., 2006; can see the impact of this logic in computer models of
Klahr and Nigam, 2004; Mayer, 2004). problem solving. Such models (Sweller, 1988) ran-
Our increasing knowledge of the importance of domly generate moves if information is lacking.
imitation in human learning provides additional evi- Does random generation mean that whether or not
dence for the borrowing principle. The discovery of a novel problem is solved is entirely due to chance?
mirror neurons that re in the same manner when we Random generation can only function properly when
take an action, observe someone else make the same coupled with effectiveness testing, and the combina-
action, or even listen to a sentence describing the tion of random generation and effectiveness testing
action provides neuropsychological evidence for the provides the knowledge- or information-creating pro-
importance of imitation as a learning mechanism (Tet- cess of natural information-processing systems. By
tamanti et al., 2005). only accepting effective alterations to an information
store and rejecting ineffective alterations, the effective-
Randomness as Genesis Principle ness of the store can increase incrementally over time.
and Creating Knowledge By transferring the information in a store to other
entities that can continue the process of random gen-
The need to transfer knowledge should not obscure eration followed by effectiveness testing indenitely,
the fact that knowledge must rst be created to have very complex, sophisticated stores can by built by nat-
something to transfer. In addition, circumstances ural information-processing systems. Both evolution
change and current knowledge may no longer be ade- by natural selection and human cognition provide

372
Human Cognitive Architecture

examples. Nevertheless, it must be recognized that extremely limited in capacity (Miller, 1956). Although
random generation followed by effectiveness testing is the precise nature of this limitation has provided a
a very slow way of generating knowledge (Cooper and source of research and discussion (Cowan, 2005), few
Sweller, 1987; Sweller and Cooper, 1985) and should would dispute the existence of the limitation. The con-
only be used when the alternative procedure of knowl- sequences of this limitation have profound conse-
edge transfer is unavailable. Knowledge transfer is quences for instructional design. Instructional designs
vastly more effective. Accordingly, technology-based that ignore this limitation are likely to be ineffective.
instruction requiring problem solving (random gener-
ation followed by effectiveness testing) can be The Environment Organizing
expected to be less effective than instruction providing and Linking Principle
demonstrations (knowledge transfer via the borrowing
principle). This principle pertains to how we use information to
function in our environment. In contrast to the limita-
The Narrow Limits of Change tions of working memory, when dealing with familiar
and Human Working Memory information that is already organized in long-term
memory, there is no functional reason for working
There are limits to knowledge creation. In the initial memory to be limited. Accordingly, huge amounts of
instance, knowledge creation occurs using the random- organized information can be transferred from long-
ness as genesis principle, which requires random gen- term memory to working memory without overloading
eration followed by effectiveness testing. Knowledge working memory (Ericsson and Kintsch, 1995). That
may be altered using the borrowing principle but that information can then be used to allow us to function
also involves random generation when new informa- in our complex environment; for example, if readers
tion is combined with old information. Mechanisms attempt to reproduce from memory the immensely
are required to ensure that any alterations to the store complex set of squiggles that constitute the last sen-
of information in long-term memory maximize the tence, most will be able to do so effortlessly. We are
probability that a particular change will be effective able to do so because of the organized, schematic
and minimize the probability that a change will destroy information held in long-term memory that can be
the functionality of the store. These conditions are met transferred and used in working memory. That infor-
by the use of a series of small, incremental changes, mation allows us to read, process, organize, and relate
each tested for effectiveness, rather than a single, very text to the external world. Similarly, whereas changes
large change. to a genome must be small and incremental, huge
Assume that the effectiveness is being tested of amounts of previously organized genetic information
adding a permutation of 4 elements to the store. can be used simultaneously to produce the complex
Assume further that no knowledge is available to elim- proteins required for biological survival in a complex
inate some of the possible permutations or select others environment.
for priority testing. Under these circumstances, there
are 4! = 24 possible permutations. In contrast, assume Origins, Evidence, and Implications
that rather than 4 elements being under consideration
there are 10. The number of possible permutations is This principle has multiple origins. Millers (1956)
10! = 3,628,800. If only a limited number or perhaps evidence that multiple elements could be chunked
only one permutation is actually effective, dealing with together to act as a single element in working memory,
more than a very small number of elements that must schema theorists demonstrating that the manner in
be combined randomly and tested for effectiveness is which we process information depends on previous
futile because the presence of any more elements rap- knowledge (Bartlett, 1932), and, more recently, Erics-
idly leads to a combinatorial explosion that no natural son and Kintschs (1995) concept of long-term work-
information processing system can possibly handle ing memory all contribute to the environment organiz-
unless it has prior knowledge. ing and linking principle. Such lines of research dem-
onstrate how we use our large store of information to
Origins, Evidence, and Implications impose order and meaning on our environment. The
environment organizing and linking principle provides
This factor alone explains some of the major charac- the ultimate justication for human cognition. The pur-
teristics of human cognitive architecture. When deal- pose of the previous four principles is to permit cog-
ing with novel information for which there is no or nition to occur so we can function mentally in our
limited prior knowledge, human working memory is environment.

373
John Sweller

COGNITIVE LOAD THEORY Based on the cognitive architecture outlined above,


extraneous cognitive load can be reduced and germane
These characteristics of human cognitive architecture cognitive load increased by assisting learners to transfer
have direct implications for instructional design and information to long-term memory. Novel information
hence the design and purpose of technology-based coming from the senses is not organized and imposes
instruction. The preferred characteristics of technol- a heavy load on working memory (the narrow limits of
ogy-based instruction can be determined by using cog- change principle). Information coming from long-term
nitive load theory (Clark et al., 2006; Paas et al., 2003, memory is organized and imposes a minimal load on
2004; Sweller, 2005a; van Merrinboer and Sweller, working memory (the environment organising and link-
2005), as well as closely related theories such as ing principle). In other words, once learned, information
Mayers theory of multimedia learning (Mayer, 2005) no longer imposes a working memory load. How should
or van Merrinboers (1997) 4C/ID model, which use we assist learners to transfer novel information to long-
this architecture to determine general instructional term memory? Wherever possible, that information
design principles. As might be expected, all theories should be in an organized form so learners do not have
that use this architecture are compatible and make to expend working memory resources in imposing an
similar or identical predications. organizational structure. Almost all information that
Cognitive load theory species two sources of learners must acquire has previously been laboriously
cognitive (or working memory) load that determine organized over many generations using the randomness
the effectiveness of instruction. Extraneous cognitive as genesis principle outlined above. Nothing is gained
load is due to inappropriate instructional designs and by requiring learners in an instructional setting to
so must be reduced. If working memory resources attempt to use random generation followed by tests of
are being fully expended, a reduction in extraneous effectiveness rather than the borrowing principle.
cognitive load is required to permit an increase in Contrary to much current educational dogma (Kir-
germane cognitive load, a form of cognitive load that schner et al., 2006; Mayer, 2004) and as indicated
can result in useful alterations to long-term memory. above, our cognitive architecture strongly facilitates
These two sources of cognitive load determine the learning through knowledge transfer via the borrowing
effectiveness of instruction, but a third source of cog- principle rather than knowledge generation via the ran-
nitive load, intrinsic cognitive load, cannot be manip- domness as genesis principle. Not only should knowl-
ulated without compromising understanding. Intrin- edge be presented to learners rather than have them
sic cognitive load (Sweller, 1994) can be thought of engage in the impossible task of attempting to generate
as the intrinsic complexity of the material being stud- it themselves, but it should also be presented in a
ied. For learners at a given level of expertise, that manner that reduces extraneous cognitive load and
complexity can be reduced (Pollock et al., 2002) but maximizes germane cognitive load. Cognitive load
only by reducing learners understanding of the sub- theory, making use of the borrowing principle, has
ject matter. Intrinsic cognitive load can, of course, generated a range of techniques intended to achieve
also be reduced by learning (i.e., by testing learners this purpose. Most of those techniques are relevant to
with more expertise). technology-based instruction and will be discussed
These sources of cognitive load are additive and next. Each technique has been studied as an experi-
cannot exceed the available capacity of working mem- mental effect using randomized, controlled experi-
ory. If intrinsic cognitive load is low, it may be possible ments in which an instructional technique generated
for germane cognitive load to be high even with inap- by cognitive load theory is compared to an alternative,
propriate instructional techniques causing a high extra- usually more traditional technique. These effects pro-
neous cognitive load. The low intrinsic cognitive load vide the instructional recommendations generated by
is likely to leave sufcient working memory resources cognitive load theory.
for students to learn even with a poor instructional
design. In contrast, if intrinsic cognitive load is high
due to high complexity material, unless the extraneous
Cognitive Load Theory Effects
cognitive load is low there may be insufcient working and Technology-Based Instruction
memory capacity to permit a level of germane cogni-
Worked Example Effect
tive load that can result in learning. With a high intrin-
sic cognitive load, it is essential to keep the extraneous The worked example effect (Sweller and Cooper, 1985)
cognitive load low to permit a sufcient level of ger- occurs when novice learners studying worked solutions
mane cognitive load. In other words, instructional to problems perform better on a problem-solving test than
design becomes critical with complex material. learners who have been given the equivalent problems

374
Human Cognitive Architecture

to solve during training. This effect ows directly from ciated statements are likely to be unintelligible without
the cognitive architecture described above. When solv- reference to the diagram. To understand the material,
ing an unfamiliar problem, problem solvers have no learners must mentally integrate the two sources of
choice but to use a random generation followed by information. Mental integration requires working mem-
effectiveness testing procedure at those choice points ory resources to be used to search for appropriate ref-
where they have insufcient knowledge to direct erences between the multiple sources of information.
choice. The working memory load associated with That search process is indistinguishable from the ran-
using an effective problem-solving procedure will dom generation followed by effectiveness testing of
interfere with learning (Cooper and Sweller, 1987; problem solving. All searches not based on knowledge
Paas and van Merrinboer, 1994) and so constitutes an require the randomness as genesis principle. If a geom-
extraneous cognitive load. In contrast, using the bor- etry solution includes the statement angle ABC, then
rowing rather than the randomness as genesis principle learners must randomly choose an angle and test
should enhance learning. Showing learners how to whether it is angle ABC, with the process continuing
solve problems via worked examples by providing an until the angle is found. Little can be expected to be
organized structure rather than leaving learners to learned by this use of working memory resources. In
devise their own organization using their limited capac- contrast, borrowing other peoples knowledge can
ity working memory should minimize extraneous cog- reduce extraneous cognitive load. If the instructions are
nitive load. Findings from a large number of studies physically integrated so angle ABC is clearly asso-
carried out by many researchers in the 1980s and 1990s ciated with the appropriate angle, obviating the need to
provided strong support for these hypotheses. search for the angle, then extraneous cognitive load
There is every reason to suppose that the worked should be reduced and learning enhanced. That result
example effect should apply directly to technology- also has been obtained on numerous occasions (Ayres
based instruction. Consider a computer-based simula- and Sweller, 2005; Sweller et al., 1990). It must be
tion. A simulation should demonstrate a process or emphasized that the split-attention effect only occurs
procedure. It should not require learners to solve novel when multiple sources of information must be inte-
problems even if problem solving is part of the process grated before they can be understood. Multiple sources
being simulated. The aim of a simulation demonstrat- of information that can be understood in isolation
ing problem solving should be identical to the aim of should not be physically integrated. Different instruc-
any other instruction in problem solving, and that aim tional procedures are required when multiple sources
is to assist learners in acquiring knowledge in long- of information can be understood in isolation and are
term memory concerning the problem-solving proce- discussed in the Redundancy Effect section.
dures relevant to that particular problem. Once that Technology-based instruction that ignores the
knowledge is acquired, learners will recognize the split-attention effect is likely to be less than effective;
problem as belonging to a category requiring particular for example, a simulation demonstrating the functions
moves for solution (Chi et al., 1982). Searching for of a mechanical device in which the function of a part
novel solutions using random generation followed by of the device can only be understood in relation to the
effectiveness testing as part of a simulation is no more function of another part of the device runs the risk of
likely to achieve this aim than searching for novel split attention. Whenever possible, the simulation
solutions in any other instructional context. Using the should be structured to clearly indicate the relation
borrowing principle, a good simulation can eliminate between the two parts. In addition, any written text
randomness by precisely indicating the structure of should be formatted in a manner that reduces or elim-
what needs to be learned. (Practicing familiar problem inates a search for referents. Learners should not be
solutions has a different function and is discussed left to work out relations between aspects of a simu-
below in the section on the expertise reversal effect.) lation. Working memory resources can be better
employed.
Split-Attention Effect
Modality Effect
Assume that the material being presented to learners
consists of two or more sources of information. Assume This effect occurs under the same conditions as the
further, that the sources of information are unintelligi- split-attention effect in that both effects occur under
ble in isolation and can only be understood in conjunc- conditions with multiple sources of information that
tion with each other. A geometry worked example pro- cannot be understood in isolation and so must be inte-
vides an instance. The diagram tells novice learners grated (Low and Sweller, 2005). The effect relies on
little if anything of the problem solution, and the asso- a particular characteristic of working memory. In the

375
John Sweller

previous discussion of human cognition, working tion. In contrast, the redundancy effect applies to mul-
memory was treated as a unitary concept; in fact, it is tiple sources of information that are intelligible in iso-
best thought of as consisting of multiple channels or lation. That difference in the logical relation between
processors (Baddeley, 1992). A visual processor for sources of information results in quite different instruc-
dealing with two- or three-dimensional objects and an tional consequences. The difference is important
auditory processor for dealing with language are par- because frequently, on the surface, conditions that lead
tially independent. As a consequence, the simultaneous to the split-attention or redundancy effects can look
use of both processors can expand the effective size identical. It is only by considering the relation between
of working memory, under some circumstances. One the multiple sources of information that the appropriate
of those circumstances consists of the conditions lead- instructional recommendations can be provided.
ing to the split-attention effectnamely, multiple In the current context, redundant information is
sources of information that must be integrated before dened as any information that is not relevant to
they can be understood. Under those conditions, the learning. Most commonly, redundant information
use of both the auditory and visual processor can consists of the same information presented in differ-
expand the effective size of working memory in an ent forms or media such as presenting the same verbal
instructionally favorable manner. information in spoken and written form, but it can
Consider a geometry example again as discussed also consist of any unnecessary, additional informa-
above. To understand a geometry worked example, the tion such as decorative pictures, background sound,
learner must simultaneously consider both the diagram or cartoons.
and the text because neither conveys the required In instructional contexts, redundancy can fre-
meaning in isolation. By presenting the verbal infor- quently be found when textual information repeats
mation in spoken rather than written form, working information found in a diagram. A diagram indicating
memory can be effectively expanded because informa- the ow of blood in the heart, lungs, and body along
tion is shifted from the overloaded visual processor to with a statement that blood ows from the left ven-
be shared by both the visual (for the diagram) and tricle to the aorta provides an example of redundancy
auditory (for the text) processors. According to cogni- (Chandler and Sweller, 1991). Although a geometry
tive load theory, such an expansion of effective work- diagram and its associated statements would appear,
ing memory should facilitate learning. Tindall-Ford et on the surface, to have the same properties and hence
al. (1997) demonstrated the modality effect when are governed by the same instructional principles as a
learners presented with instructions in a visual, split- blood ow diagram and its associated statements, they
attention format learned less than learners presented are structurally very different and require very differ-
with the same material but with all of the verbal infor- ent formatting. A geometry diagram tells us little of a
mation presented in spoken rather than written form. problem solution and requires the statements for an
The modality effect is directly applicable to tech- intelligible solution to be communicated to learners.
nology-based instruction. Whereas an aspect of tech- Those statements must be integrated with the diagram
nology-based instruction requires verbal input to be so a search for referents is reduced. In contrast, the
intelligible, extraneous cognitive load can be reduced blood ow diagram can be fully intelligible in its own
by using spoken rather than written text. The use of right and provide a full explanation. The appropriate
spoken text is particularly important during an anima- instructional procedure is not to integrate the state-
tion. The expansion of effective working memory due ments with the diagram but rather to eliminate the least
to the use of both the auditory and visual processors effective source of information, which, in this case, is
can permit an animation to be viewed while simulta- the set of statements. Many experimental examples
neously attending to speech that explains otherwise demonstrate that the elimination of redundancy facil-
unintelligible aspects of the animation. The use of writ- itates learning (Chandler and Sweller, 1991).
ten rather than spoken text runs the risk of overloading Although the redundancy effect can be considered
the visual processor. In contrast, appropriate use of both counter-intuitive, from a cognitive load theory per-
visual and auditory information can maximize the spective the reason why redundant information has
potentially powerful effects of the borrowing principle. negative consequences is straightforward. Attending to
unnecessary information and attempting to integrate it
Redundancy Effect with essential information requires working memory
resources that consequently are unavailable for learn-
The previous two effects discussed, the split-attention ing. Redundant information imposes an extraneous
and modality effects, apply to multiple sources of cognitive load. It is an ineffective use of the borrowing
information, each of which is unintelligible in isola- principle.

376
Human Cognitive Architecture

There are many forms of redundancy other than the The effect occurs when an instructional procedure
diagram and text redundancy described above (for a that is relatively effective for novices compared to a
detailed summary, see Sweller, 2005b). All forms of control procedure rst loses its advantage as levels of
redundancy potentially apply to technology-based expertise increase and then begins to be worse than the
instruction; for example, a spoken commentary associ- control procedure with further increases in expertise
ated with a simulation should only be included if the (Kalyuga et al., 2003). As an instance, studying worked
visual material is unintelligible without the commentary. examples is better than solving the equivalent problems
If visual material is intelligible in isolation, a spoken for novices, but with increased expertise solving prob-
commentary will increase cognitive load as learners lems becomes better than studying examples (Kalyuga
attempt to integrate the auditory and visual information. et al., 2001). As other instances, for novices, integrated
When devising a simulation, care must be taken to format or dual-modality instruction is better than split-
ensure that all information presented is essential and not attention format instruction. With increasing expertise,
simply an alternative way of presenting the same infor- rather than integrating, for example, diagrams and text
mation. Equally damaging is additional information that or using dual-modality instruction, it is better to elim-
bears little or no relation to the required information. It inate the text entirely (Kalyuga et al., 1998, 2000).
is a trap to assume that additional information will, at The expertise reversal effect is a complex effect
worst, be neutral in its effect and could be benecial. that relies on redundancy. As expertise increases, pre-
Increasing technical sophistication is permitting viously essential information becomes redundant and
increasingly realistic simulations. By denition, the so imposes an extraneous cognitive load. Studying
real world is realistic, but we developed instructional worked examples may be essential for novices to
systems precisely because many aspects of the real reduce cognitive load, but such activity becomes
world provide inadequate instruction. That inadequacy redundant with increasing expertise and is better
frequently is caused by redundancy. A realistic simu- replaced by practice at solving problems which, at
lation that provides a depiction of a mechanical system higher levels of expertise, no longer imposes a cogni-
may be indistinguishable from the real mechanical tive load. Similarly, explanatory text may be essential
system, but that system may be almost useless as an for novices and so should be physically integrated with
instructional tool. When learning how the blood ows diagrams or presented in spoken form to reduce extra-
in the heart, lungs, and body, most of the structures, neous cognitive load. With increasing expertise, the
functions, and characteristics of the body are irrele- text becomes redundant and should be eliminated.
vant, which is why it took so long to discover the The expertise reversal effect suggests that the detail
processes of the circulatory system using the random- provided in technology-based instruction should be
ness as genesis principle. The realistic but irrelevant determined by the knowledge base of the learners.
features served to conceal the critical features. Realis- Details that are essential for novices may be redundant
tic features and processes should not be included in a for more expert learners. Thus, technology-based
simulation if the sole reason for inclusion is realism. instruction must be constructed so its specications
To avoid redundancy and maximize the effect of the change with changes in expertise. Furthermore, if
borrowing principle, there should be a clear instruc- instructional materials are to change with changes in
tional reason for including any information. expertise, a method is required to rapidly determine
levels of expertise. Kalyuga and Sweller (2004) pro-
Expertise Reversal Effect vided a rapid test of knowledge based on the cognitive
architecture described above. During instruction,
Two related points must be made concerning the dis- learners were presented a partially completed problem
cussion of the previous effects. First, all of the above and asked to indicate the next step required for solu-
effects assume that learners are novices. It is novices tion. The extent to which a learner knows the next step
who most frequently require instruction via the bor- to solution depends on the knowledge base held in
rowing principle. Second, the previous explanation of long-term memory. That information can be used to
the redundancy effect implicitly assumed that redun- determine subsequent instruction. It should similarly
dancy is purely a function of the materials being used; be possible to use this rapid assessment technique to
in fact, it is equally a function of levels of expertise. determine the nature of any subsequent instruction.
Information that is redundant for a more expert learner
may be critically necessary for a less expert learner. A Guidance Fading Effect
novice may need to borrow information from someone
else, an expert may not. The expertise reversal effect The guidance fading effect (Renkl and Atkinson, 2003;
was born from these considerations. Renkl et al., 2004) is closely related to the worked

377
John Sweller

example and expertise reversal effects and is also a and guidance fading effects, for novices information
compound effect. It occurs when novices are initially is best borrowed from another persons long-term
presented with worked examples, but with increasing memory, but as levels of expertise increase that infor-
expertise those worked examples are replaced by com- mation can be borrowed from ones own long-term
pletion problems (van Merrinboer et al., 2002) in memory for purposes of practice, in this case mental
which a partial solution is provided and the learner is practice.
required to complete the problem. With further The imagination effect provides information on
increases in expertise, the completion problems should what mental activity learners should be engaged in
be replaced by full problems. Again, this sequence is when dealing with a simulation. Initially, they should
predicated on the assumption that what constitutes an simply study or interact with the simulation materials
extraneous cognitive load depends not just on the because due to working memory limitations they are
nature of the instruction but on an interaction between unlikely to have sufcient knowledge to be able to
the instructional procedures and learner characteristics effectively imagine the procedures and concepts. With
in the form of levels of expertise. At low levels of increasing expertise, they should attempt to imagine
expertise, the learner must make heavy use of other the information covered by the instruction because that
peoples knowledge via the borrowing principle. With procedure seems to be the most rapid technique for
increasing expertise, the same information can be bor- transferring information to long-term memory and so
rowed from the learners own long-term memory and increasing levels of expertise.
used for practice purposes.
Technology-based instruction should take account Element Interactivity Effect
of the guidance fading effect by initially providing
substantial guidance which should be gradually faded The expertise reversal effect discussed above places
as expertise increases. As an example, initially, learn- an emphasis on the cognitive load implications of indi-
ers should be shown exactly what they need to do with vidual differences in expertise. Differences in the
minimal action required on their part. With increases structure of the information being considered are
in expertise, determined by rapid assessment tech- equally important. None of the above effects is obtain-
niques, learner activity should be increased and guid- able using low-complexity material (Sweller, 1994)
ance decreased. Ultimately, it should be possible to that has a low intrinsic cognitive load. Recall that total
remove all guidance with the learner simply practicing cognitive load is an addition of extraneous, intrinsic,
the skill. and germane cognitive load. The above effects are
primarily determined by an excessive extraneous cog-
Imagination Effect nitive load that reduces germane cognitive load
because working memory capacity is exceeded. If
The imagination effect occurs when learners who are intrinsic cognitive load is low, a high extraneous cog-
asked to imagine a procedure or concept learn more nitive load may not matter a great deal. There may be
than learners who are asked to study the same proce- sufcient working memory capacity available to
dure or concept (Cooper et al., 2001; Leahy and enable germane cognitive load and its attendant rapid
Sweller, 2005). Imagination instructions ask learners learning to occur.
to turn away from the material and attempt to imagine What determines levels of intrinsic cognitive load?
the relevant procedures or concepts. Imagining The only relevant factor within a cognitive load theory
involves running material through working memory framework is element interactivity, which is deter-
which should assist in the transfer of the information mined by the number of interacting elements that must
to long-term memory. The technique is highly effective be considered simultaneously to understand the mate-
but only when used by learners with sufcient experi- rial. Some information is low in element interactivity
ence in the domain to be able to process all of the in that the elements can be learned one element at a
necessary information in working memory without time without considering any other elements. Learning
assistance from the instructional material. For novices, technical terminology provides an example. One can
attempts to run a procedure through working memory learn the name of a component without learning the
are likely to fail, so instructions to study material, names of any other components, so working memory
which involves considering it while looking at it, are load may be very low. Cognitive load effects are not
superior to imagination instructions. The switch from likely to be relevant when intrinsic cognitive load is
studying being superior to imagination being superior low. In contrast, learning how components interact in
provides another example of the expertise reversal a machine has high element interactivity because it
effect. Again, as was the case for the expertise reversal may be impossible to understand the function of one

378
Human Cognitive Architecture

component without simultaneously considering the DISCUSSION


function of all of the components. High element inter-
activity results in a high intrinsic cognitive load, leav- Cognitive load theory provides an integrated system
ing little working memory capacity available for learn- dealing with the evolutionary origins of human cogni-
ing. Under these circumstances, levels of extraneous tion leading to an explanation of cognitive structures
cognitive load become critical, and the cognitive load and processes. In turn, those structures and processes
effects discussed above become relevant; thus, cogni- can be used to generate instructional principles. The
tive load effects become critical when technology- generation of applications in the form of instructional
based instruction deals with complex, high element principles provides some warrant for the validity of the
interactivity material. original cognitive architecture. This unied system can
provide a base for instructional design including the
structure and function of technology-based instruction.
Isolated Interacting Elements Effect
There are two aspects of technology-based educa-
When element interactivity is very high, it may be tion in general and of the work reported in this chapter
impossible for learners to understand the material that require emphasis. First, the use of technology in
because it may be impossible for them to simulta- education should not be based merely on the availabil-
neously process all of the interacting elements in work- ity of technology. There is a long history of new tech-
ing memory. How should such material be presented? nological applications such as radio, lms, television,
From a cognitive load theory perspective, the only way mainframe computers attached to terminals, stand-
seems to be to initially present the material as individ- alone microcomputers and now the Web being hailed
ual elements ignoring their interactions. This proce- as potentially revolutionizing education. Frequently,
dure will permit the elements to be learned but without these technological advances have had minimal, long-
understanding. Once the individual elements have been term educational impact. While technology changes,
learned, their interactions can be emphasized. It is only human cognitive architecture does not. The introduc-
at that point that the material will be understood tion of educational technology without reference to its
because it cannot be understood by simply considering cognitive consequences is unlikely to be effective. The
individual elements. Empirical work has demonstrated ve natural information processing principles outlined
that teaching individual elements rst, at the expense above provide an initial guide. According to those
of understanding, followed by teaching the interactions principles, learning consists of changes to the long-
between elements results in more effective learning term store, the most efcient method of bringing about
than attempting to have learners understand very high those changes is by borrowing knowledge from knowl-
element interactivity material right from the beginning edgeable educators, and that knowledge must be struc-
of instruction (Pollock et al., 2002). tured in a manner that reduces working memory load.
The isolated, interacting elements effect may have I am not aware of any evidence that simply using
considerable relevance to technology-based instruc- technology will necessarily conform to any of the ve
tion such as instructional simulations. Material is dif- principles. Instruction that ignores human cognitive
cult to understand because it is high in element architecture is likely to be ineffective whether or not
interactivity. Presenting that material in a realistic it uses technology. In contrast, technology-based
fashion during a simulation may be condemning instruction that is explicitly structured to conform to
learners to attempting to understand information that what we know of human cognition has a much better
vastly exceeds their working memory capacity chance of being effective.
because a realistic simulation may involve a huge All of the cognitive load effects discussed above
number of interacting elements. A less realistic sim- were based on the assumption that the aim of instruc-
ulation with fewer interacting elements may be more tion is the acquisition of knowledge in long-term mem-
readily understood and learned. Although a full ory and that the best way of achieving this aim is to
understanding of the material is impossible by this make use of the borrowing principle by providing
technique, it is equally impossible to process a large direct instructional guidance organized to reduce extra-
number of interacting elements simultaneously. It neous working memory load.
may be better to provide simulations that result in The second point that requires emphasis concerns
limited understanding initially, followed by more how we determine whether technology-based educa-
complete versions that permit full understanding. In tion is effective. The instructional effects described
this manner, learners may be spared the need to above were not only based on our knowledge of human
attempt to understand material that is quite impossible cognitive architecture but also were tested using con-
for them to deal with. trolled, randomized experimental designs. Experiments

379
John Sweller

in which participants are randomly allocated to two or Baddeley, A. (1992). Working memory. Science, 255,
more experimental groups with different instructional 556559.*
Bartlett, F. (1932). Remembering: A Study in Experimental and
procedures but identical test procedures are essential Social Psychology. Cambridge, U.K.: Cambridge University
when determining effective instructional procedures in Press.
this eld. Verication procedures in which instructors Chandler, P. and Sweller, J. (1991). Cognitive load theory and
and learners are presented instruction using a new tech- the format of instruction. Cogn. Instruct., 8, 293332.
nology and simply asked if they think it is effective or Chase, W. G. and Simon, H. A. (1973). Perception in chess.
Cogn. Psychol., 4, 5581.
if they enjoyed it have very limited utility. Chi, M., Glaser, R., and Rees, E. (1982). Expertise in problem
While cognitive load theory indicates many condi- solving. In Advances in the Psychology of Human Intelli-
tions under which instruction may or may not be effec- gence, edited by R. Sternberg, pp. 775. Hillsdale, NJ:
tive, those conditions are closely dependent on the Lawrence Erlbaum Associates.
cognitive architecture discussed and, specically, on Clark, R., Nguyen, F., and Sweller, J. (2006). Efciency in
Learning: Evidence-Based Guidelines to Manage Cognitive
the characteristics of working and long-term memory. Load. San Francisco, CA: Pfeiffer.*
The theory is not intended as a general theory of cog- Cooper, G. and Sweller, J. (1987). The effects of schema acqui-
nition or of instruction, and so many important vari- sition and rule automation on mathematical problem-solving
ables are not considered by the theory; for example, transfer. J. Educ. Psychol., 79, 347362.
despite some discussion about including motivational Cooper, G., Tindall-Ford, S., Chandler, P., and Sweller, J. (2001).
Learning by imagining. J. Exp. Psychol. Appl., 7, 6882.
variables in cognitive load theory and considering this Cowan, N. (2005). Working Memory Capacity. Oxford, U.K.:
issue in future research, at present it has not been Psychology Press.*
incorporated into the theory. That incorporation will, De Groot, A. (1946/1965). Thought and Choice in Chess. The
in all probability, only occur if relations between cog- Hague: Mouton.
nitive architecture and motivation can be established. Egan, D. E. and Schwartz B. J. (1979). Chunking in recall of
symbolic drawings. Memory Cogn., 7, 149158.
In conclusion, advances in instructional technol- Ericsson, K. A. and Kintsch, W. (1995). Long-term working
ogy have provided us with the ability to use instruc- memory. Psychol. Rev., 102, 211245.*
tional procedures that until recently were difcult or Jeffries, R., Turner, A., Polson, P., and Atwood, M. (1981).
impossible to implement. Frequently, those procedures Processes involved in designing software. In Cognitive Skills
have been recommended purely because they are now and Their Acquisition, edited by J. R. Anderson, pp.
255283. Hillsdale, NJ: Lawrence Erlbaum Associates.
possible rather than because there was evidence for Kalyuga, S. and Sweller, J. (2004). Measuring knowledge to
their cognitive effectiveness or even desirability. Cog- optimize cognitive load factors during instruction. J. Educ.
nitive load theory is intended to provide a guide to the Psychol., 96, 558568.
types of techniques that are likely to work. It has been Kalyuga, S., Chandler, P., and Sweller, J. (1998). Levels of
tested using controlled, randomized experiments over expertise and instructional design. Hum. Factors, 40, 117.
Kalyuga, S., Chandler, P., and Sweller, J. (2000). Incorporating
many years by many investigators in a wide variety of learner experience into the design of multimedia instruction.
technology-based environments. There is every reason J. Educ. Psychol., 92, 126136.
to suppose that the use of our knowledge of human Kalyuga, S., Chandler, P., Tuovinen, J. E., and Sweller, J.
cognition will continue to generate technology-based (2001). When problem solving is superior to studying
instructional procedures. worked examples. J. Educ. Psychol., 93, 579588.
Kalyuga, S., Ayres, P., Chandler, P., and Sweller, J. (2003).
Expertise reversal effect. Educ. Psychol., 38, 2331.
Kirschner, P., Sweller, J., and Clark, R. (2006). Why minimal
ACKNOWLEDGMENTS guidance during instruction does not work: an analysis of
the failure of constructivist, discovery, problem-based, expe-
The work reported here was supported under Ofce of riential and inquiry-based teaching. Educ. Psychol., 41,
7586.
Naval Research Award Number #N00014-04-1-0209, Klahr, D. and Nigam, M. (2004). The equivalence of learning
as administered by the Ofce of Naval Research. The paths in early science instruction: effects of direct instruction
ndings and opinions expressed in this report do not and discovery learning. Psychol. Sci., 15, 661667.
necessarily reect the positions or policies of the Leahy, W. and Sweller, J. (2005). Interactions among the imag-
Ofce of Naval Research. ination, expertise reversal and element interactivity effects.
J. Exp. Psychol. Appl., 11, 266276.*
Low, R. and Sweller, J. (2005). The modality principle. In
REFERENCES Cambridge Handbook of Multimedia Learning, edited by R.
E. Mayer, pp. 147158. Cambridge, U.K.: Cambridge Uni-
Ayres, P. and Sweller, J. (2005). The split-attention principle. versity Press.
In Cambridge Handbook of Multimedia Learning, edited by Mayer, R. E. (2004). Should there be a three-strikes rule against
R. E. Mayer, pp. 135136. Cambridge, U.K.: Cambridge pure discovery learning? The case for guided methods of
University Press.* instruction. Am. Psychol., 59, 1419.

380
Human Cognitive Architecture

Mayer, R. E. (2005). Cognitive theory of multimedia learning. Sweller, J. (2005a). Implications of cognitive load theory for
In Cambridge Handbook of Multimedia Learning, edited by multimedia learning. In Cambridge Handbook of Multime-
R. E. Mayer, pp. 3148. Cambridge, U.K.: Cambridge Uni- dia Learning, edited by R. E. Mayer, pp. 1930. Cambridge,
versity Press.* U.K.: Cambridge University Press.
Miller, G. A. (1956). The magical number seven, plus or minus Sweller, J. (2005b). The redundancy principle. In Cambridge
two: some limits on our capacity for processing information. Handbook of Multimedia Learning, edited by R. E. Mayer,
Psychol. Rev., 63, 8197. pp. 159167. Cambridge, U.K.: Cambridge University
Paas, F. and van Merrinboer, J. J. G. (1994). Variability of Press.
worked examples and transfer of geometrical problem solv- Sweller, J. and Cooper, G. A. (1985). The use of worked exam-
ing skills: a cognitive-load approach. J. Educ. Psychol., 86, ples as a substitute for problem solving in learning algebra.
122133. Cogn. Instruct., 2, 5989.
Paas, F., Renkl, A., and Sweller, J. (2003). Cognitive load theory Sweller, J. and Sweller, S. (2006). Natural information process-
and instructional design: recent developments. Educ. Psy- ing systems. Evol. Psychol., 4, 434458.
chol., 38, 14.* Sweller, J., Chandler, P., Tierney, P., and Cooper, M. (1990).
Paas, F., Renkl, A., and Sweller, J. (2004). Cognitive load the- Cognitive load and selective attention as factors in the struc-
ory: instructional implications of the interaction between turing of technical material. J. Exp. Psychol. Gen., 119,
information structures and cognitive architecture. Instruct. 176192.
Sci., 32, 18. Tettamanti, M., Buccino, G., Saccuman, M., Gallese, V., Danna,
Pollock, E., Chandler, P., and Sweller, J. (2002). Assimilating M., Scifo, P., Fazio, F., Rizzolatti, G., Cappa, S., and Perani,
complex information. Learn. Instruct., 12, 6186. D. (2005). Listening to action-related sentences activates
Renkl, A. and Atkinson, R. (2003). Structuring the transition fronto-parietal motor circuits. J. Cogn. Neurosci., 17,
from example study to problem solving in cognitive skill 273281.
acquisition: a cognitive load perspective. Educ. Psychol., 38, Tindall-Ford, S., Chandler, P., and Sweller, J. (1997). When two
1522. sensory modes are better than one. J. Exp. Psychol. Appl.,
Renkl, A., Atkinson, R., and Groe, C. (2004). How fading 3, 257287.
worked solution steps works: a cognitive load perspective. Van Merrinboer, J. J. G. (1997). Training Complex Cognitive
Instruct. Sci., 32, 5982. Skills. Englewood Cliffs, NJ: Educational Technology Pub-
Simon, H. A. and Gilmartin, K. (1973). A simulation of memory lications.*
for chess positions. Cogn. Psychol., 5, 2946. Van Merrinboer, J. J. G. and Sweller, J. (2005). Cognitive load
Sweller, J. (1988). Cognitive load during problem solving: theory and complex learning: recent developments and
effects on learning. Cogn. Sci., 12, 257285.* future directions. Educ. Psychol. Rev., 17, 147177.*
Sweller, J. (1994). Cognitive load theory, learning difculty and Van Merrinboer, J. J. G., Schuurman, J., de Croock, M., and
instructional design. Learn. Instruct., 4, 295312.* Paas, F. (2002). Redirecting learners attention during train-
Sweller, J. (2003). Evolution of human cognitive architecture. ing: effects on cognitive load, transfer test performance and
In The Psychology of Learning and Motivation, Vol. 43, training efciency. Learn. Instruct., 12, 1137.
edited by B. Ross, pp. 215266. San Diego, CA: Academic
Press.*
Sweller, J. (2004). Instructional design consequences of an anal-
ogy between evolution by natural selection and human cog-
nitive architecture. Instruct. Sci., 32, 931. * Indicates a core reference.

381