Cognitive Load and Discovery Learning

Mr. Juhani E. Tuovinen

Charles Sturt University

School of Education

P O Box 588, Wagga Wagga, NSW, 2678

Ph 02 6933 2461

Fax 02 6933 2888

Email jtuovinen@csu.edu.au

ABSTRACT

Innovative forms of instructional provision such as discovery learning

are discussed in the context of cognitive load theory. Recent

educational experiments with school and university students are

described where methods based on cognitive load theory were used to

measure the educational effectiveness of instruction in computer

education, contrasting discovery learning and other forms of

instruction. The implications of the results of these experiments

suggest considerable deficiencies in discovery learning with data

favouring new, more effective forms of instruction based on cognitive

load theory. These new learning approaches reduce cognitive load by

eliminating the extraneous working memory load caused by the use of

some problem solving strategies during learning, or the elimination of

split-attention and redundancy effects for material that imposes a high

working memory load.

Cognitive Load Theory

Cognitive load theory (Sweller, 1988; 1994) derives instructional

design principles from aspects of our cognitive architecture. The

theory assumes a very limited working memory (Miller, 1956), an

effectively unlimited long-term memory (Simon and Gilmartin, 1973)

holding large numbers of schemas (Chi, Glaser & Rees, 1982) that can

vary in their degree of automaticity (Kotovsky, Hayes & Simon, 1985).

This architecture interacts with instructional material in various

ways.

First, different learners will process the material in different ways.

If the elements of material that require processing are incorporated in

an automated schema, working memory load (or cognitive load) will be

low. Schemas allow many elements to be treated as a single element in

working memory and automatic processing limits working memory demands

compared to controlled, conscious processing (Schneider & Shiffrin,

1977; Shiffrin & Schneider, 1977). As a consequence, if a learner has

acquired appropriate automated schemas, cognitive load will be low and

substantial working memory resources are likely to be free. In

contrast, if the elements of material that require processing must each

be considered as a discreet element in working memory because no schema

is available, cognitive load will be high. Working memory may be

entirely occupied in processing large numbers of individual elements.

Second, the characteristics of the instructional material is important.

Some material can be learned element by element without relating one

element to another. Learning basic word pairs in foreign language

provides an example. Each vocabulary item can be learned without

reference to any other item. Such material is low in element

interactivity and low in intrinsic cognitive load. It imposes minimal

demands on working memory. Alternatively, situations where a number of

elements must be considered simultaneously for the successful execution

of a task, are called high element interactivity tasks. Learning the

order of words in English provides an example. Word order cannot be

learned without considering several words simultaneously. Under these

circumstances, intrinsic cognitive load is high because of high element

interactivity. These situations occur often in mathematics, computer

programming, design development, etc., where no individual component

can be considered in isolation, since any action on a given component

will have complex and far-reaching effects on the task outcome.

Third, the characteristics of the learner and the material to be

learned, interact. Material which imposes a heavy cognitive load for

some people because they must deal with large numbers of interacting

elements may impose less of a cognitive load for other people because

they have acquired automated schemas that incorporate the individual

elements. An expert in elementary algebra will treat the equation, (a +

b)/c = d, as a single, automated schema requiring limited working

memory resources. A novice who has just commenced learning algebra may

need to treat each symbol and relations between symbols as individual,

interacting elements, resulting in a working memory overload.

This theory has proved beneficial for the improvement of the planning,

organisation and implementation of learning in many fields. It is

argued that in the process of dealing with information, working memory

has only a limited processing capacity available to deal with distinct

items at any given time, and that the capacity of working memory is

often overloaded due to inappropriate presentation of material or

inappropriate learner activities, leading to a reduction in learning

and the capacity to solve problems. Thus new material is learned most

effectively and efficiently if the unnecessary cognitive load is

reduced to a minimum.

The cognitive load associated with any task consists of two parts.

There is the intrinsic or natural cognitive load, i.e. the inherent

aspects of the mental task that must be understood for the learner to

be able to carry out the task. Intrinsic cognitive load is determined

by levels of element interactivity. However, in addition, there is

usually a range of extraneous matters associated with the way the

instructional material is taught, that may add to the inherent nucleus

of the intrinsic cognitive load (Sweller, 1988, 1994). This category of

cognitive load is classed as extraneous cognitive load.

In an effort to reduce the extraneous cognitive load associated with

the use of problem solving search strategies, several researchers have

tested the effects of using worked examples. Worked examples require

the reader to attend to problem states and their associated moves,

rather than searching for the right moves involved in conventional

problem solving. In situations where an extraneous cognitive load due

to problem solving search existed, worked examples were found to

effectively reduce that load and enhance learning (e.g. Carroll, 1994;

Cooper & Sweller, 1987; Paas, 1992; Paas & van Merrienboer, 1994;

Sweller & Cooper, 1985; Trafton and Reiser, 1993; Zhu & Simon, 1987).

This presentation deals with some comparisons of approaches to learning

which have a bearing on discovery learning.

Discovery Learning

Discovery learning principles have been widely promoted since the

1960's at various levels of education. Let us begin with three

statements by Jerome Bruner, the father of discovery learning.

"Consider now what benefits might be derived from the experience of

learning through discoveries that one makes oneself. I shall discuss

these under four headings: (1) the increase in intellectual potency,

(2) the shift from extrinsic to intrinsic rewards, (3) the learning of

the heuristics of discovering, and (4) the aid to conserving memory."

(Bruner, 1979, p. 83)

"By recognizing the legitimacy of intuition as an intellectual

operation, schools could spare their students the painful relearning

that is required of them later when , for example, they "really get

into" physics and are required not to prove a given solution but to

find a solution." (Bruner, 1971, p. 92)

"You cannot consider education without taking into account how culture

gets passed on. It seems to me highly unlikely that given the

centrality of culture in man's adaptation to his environment - the fact

that culture serves him in the same way as changes in morphology served

earlier in the evolutionary scale - that, biologically speaking, one

would expect each organism to rediscover the totality of its culture -

this would seem most unlikely. Moreover, it seems equally unlikely,

given the nature of man's dependency as a creature, that this long

period of dependency characteristic of our species was designed

entirely for the most inefficient technique possible for regaining what

has been gathered over a long period of time, i.e. discovery." (Bruner,

1966, p.101).

In these three statements Bruner makes three major claims about

discovery and learning. In the first statement he argues that learning

by discovery is beneficial. In the second statement he asserts that it

is important to learn to discover for oneself, that is, he sees

discovery as an end or a goal of learning, rather than the means, as

suggested by the first statement. In the final statement he cautions

that learning by discovery is inefficient, and should not be expected

to be the main means of education.

Discovery Learning and Cognitive Load

Work in the Cognitive Load Theory framework has resulted in two major

findings that help to clarify some aspects of the questions regarding

the usefulness of discovery learning.

Goal-free problem solving

When students studying kinematics in physics were given goal-free

problems, instead of the conventional problems, which required them to

find one specific answer, their learning improved (Sweller & Levine,

1982). In this situation the students had available a small number of

kinematics equations describing the action of bodies under constant

acceleration. Conventionally after a lesson introducing the equations

they were given problems of the type:

"If a stone begins to move from rest under the action of constant

acceleration of 3 m/s2, find the final velocity after 4 seconds."

The alternative form of goal-free problem used was:

"If a stone begins to move from rest under the action of constant

acceleration of 3 m/s2, find all about it after it has moved for 4

seconds."

In this situation the students were discovering or exploring all the

possible alternative variables describing the object's motion and their

values. The goal-free problem solving produced better learning than the

conventional problem solving.

Why was this format of practice better? On reflection it became

apparent that in the conventional goal-directed practice the students

were employing a problem solving strategy called 'means-ends analysis'.

During the means-ends analysis the students had to attend to the

possible actions for problem solving, the given information, and the

final goal, and how they might reach the final goal, i.e. the

intermediate goals. In the goal-free condition they only had to

process in their limited working memories the possible actions, the

given information and a simpler goal - to find an equation to derive

any new variable value. Thus instead of overloading their working

memories with an extraneous load as described above, they were able to

deal with a smaller number of variables more effectively. Thus instead

of concentrating on the problem solving to the exclusion of the

learning the schema for the subject matter to be learned, they were

able to develop better schemas, due to a smaller cognitive load in the

goal-free problem solving.

However, in a subsequent experiment involving geometry learning the

goal-free problem superiority effect over conventional problem solving

vanished. It became apparent that in the physics situation there was

only a small number of possible equations that could be used, and only

a small number of variables and their values that could be derived.

This is sometimes termed the size of search space in problem solving.

When the search space became larger, the greater variety of options

available produced an intolerable load on the working memory and the

processing advantage disappeared.

Instead the heavy use of worked examples was found to be a more

effective way to teach geometry (and many other things as noted above)

than conventional problem practice.

Exploration learning vs worked examples practice

In a more recent experiment the worked examples practice approach was

directly contrasted with exploration (discovery) practice for learning

to use a computer program, a database (Tuovinen & Sweller,

unpublished). The university students were given common lessons in the

development of database files and their manipulation. Then they were

asked to practice the operation before sitting for a test of the

database operations a week later. In the practice stage they either

freely explored the database operations or read through multiple worked

examples before working through practice tasks.

In this work there was no difference in the learning by the two groups

on low element interactivity material. However, with high element

interactivity content, i.e. the construction of database field

formulas, interesting results were obtained. If the students had

previously experienced some database work, there was no statistically

significant difference between the forms of learning. However, for the

students with no previous database exposure before this unit of work,

the worked examples practice was significantly more effective. In fact

the mean test scores for the 4 groups were:

DATABASE EXPERIENCE

NO PREVIOUS EXPERIENCE PREVIOUS EXPERIENCE

WORKED EX 29.6 30.9

EXPLORATION 15.1 35.9

This indicates that the exploration practice was only useful if the

student already possessed the schema required to be used in exploration

or discovery. On the other hand if the students did not possess the

schema, they were significantly disadvantaged by the exploration or

discovery practice format in comparison with the worked examples

practice.

General Discussion

What does this say about discovery or exploration learning in general?

Firstly, it appears that if the material has low element interactivity,

most common practice methods are equivalent. Secondly, it would appear

that if the schema for the area is known, the exploration or discovery

method may be as beneficial as the worked examples approach. However,

if the schema are poorly known the exploration or discovery approach is

more time consuming and the learning is less effective than the worked

examples approach for high element interactivity material.

Similarly if the search space is small, i.e. the number of elements to

be manipulated is low, the exploration of discovery method is more

effective than conventional problem solving practice, due to the

reduction of the extraneous cognitive load. However, even when the

search space is much larger, the worked examples approach is more

effective than the conventional problem solving.

Thus the worked examples approach produced at least as good or better

learning than the exploration or discovery approach.

This work has mainly addressed Bruner's first contention, with respect

to the benefits derived from learning by discovery in specific

contexts, and supported his third cautionary comment regarding the

limitations of the learning by discovery. Thus we have made a beginning

in the long process of finding out, as Cronbach put it in 1966, of

generalisations with respect to the benefits and limitations of

inductive (discovery) learning approaches for:

¥ particular subject matter

¥ inductive experiences of particular type

¥ provided in particular amounts

¥ producing particular patterns of responses

¥ in pupils at a particular level of development (Cronbach, 1966, p. 77).

References

Bruner, J. S. (1966). Some elements of discovery. In L. S. Shulman & E.

R. Keislar (Eds.), Learning by discovery: A critical appraisal (pp.

101-113). Chicago: Rand McNally.

Bruner, J. S. (1971) Toward a disciplined intuition. In A. Gil (Ed.)

The relevance of education (pp.82-107). London: George Allen & Unwin.

Bruner, J. S. (1979) The quest for clarity. In J. S. Bruner (Ed.) On

knowing. Essays for the left hand (pp. 75-96). Cambridge, MA: Harvard

Univ. Press.

Carroll, W. (1994) Using worked examples as an instructional support in

the algebra classroom. Journal of Educational Psychology, 86(3),

360-367.

Chi, M. T. H., Glaser, R. & Rees, E. (1982) Expertise in problem

solving. In R. J. Sternberg (Ed.) Advances in the psychology of human

intelligence, (pp. 7-75) Hillsdale, New Jersey: Lawrence Erlbaum Assc.

Cooper, G., & Sweller, J. (1987). Effects of schema acquisition and

rule automation on mathematical problem-solving transfer. Journal of

Educational Psychology, 79(4), 347-362.

Cronbach, L. J. (1966). The logic of experiments on discovery. In L. S.

Shulman & E. R. Keislar (Eds.), Learning by discovery: A critical

appraisal (pp. 76-92). Chicago: Rand McNally.

Kotovsky, K., Hayes, J. R., & Simon, H. A. (1985) Why are some

problems hard? Evidence from the Tower of Hanoi. Cognitive Psychology,

17, 248-294.

Miller, G. A. (1956). The magical number, seven, plus or minus two:

some limits on our capacity for processing information. Psychological

Review, 63, 81-97.

Owen, E., & Sweller, J. (1985). What do students learn while solving

mathematics problems? Journal of Educational Psychology, 77(3),

272-284.

Paas, F. G. W. C. (1992). Training strategies for attaining transfer of

problem-solving skill in statistics: a cognitive-load approach. Journal

of Educational Psychology, 84(4), 429-434.

Paas, F. G. W. C., & Van Merri‘nboer, J. J. G. (1994). Variability of

worked examples and transfer of geometrical problem-solving skills: A

cognitive-load approach. Journal of Educational Psychology, 86(1),

122-133.

Schneider, W. & Shiffrin, R.(1977) Controlled and automatic information

processing: I. Detection, search and attention. Psychological Review,

84, 1-66.

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

information processing:II. Perceptual learning, automatic attending,

and a general theory. Psychological Review, 84, 127-190.

Simon, H. A. & Gilmartin, K.(1973) A simulation of memory chess

positions. Cognitive Psychology, 5, 29-46.

Sweller, J., & Cooper, G. A. (1985). The use of worked examples as a

substitute for problem solving in learning algebra. Cognition and

Instruction, 2(1), 59-89.

Sweller, J., & Levine, M. (1982). Effects of goal specificity on

means-ends analysis and learning. Journal of Experimental Psychology:

Learning, Memory and Cognition., 8, 463-474.

Sweller, J.(1988) Cognitive load during problem solving: effects on

learning. Cognitive Science, 12, 257-285.

Sweller, J. (1994) Cognitive load theory, learning difficulty, and

instructional design. Learning and Instruction, 4, 295-312.

Trafton, J. G., & Reiser, B. J. (1993). The contributions of studying

examples and solving problems to skill acquisition. In Proceedings of

the Fifteenth Annual Conference of the Cognitive Science Society, (pp.

1017-1022). Institute of Cognitive Science, University of

Colorado-Boulder: Lawrence Erlbaum Assoc., Pub.

Tuovinen, J. E. & Sweller, J. (unpublished) Cognitive load associated

with discovery learning techniques.

Zhu, X., & Simon, H. A. (1987). Learning mathematics from examples and

by doing. Cognition and Instruction, 4, 137-166.