Cognitive Learning Research in Online Multimedia Education®
Dr. Juhani E. Tuovinen
Senior Research Fellow in Interactive Multimedia
Monash University, CeLTS, Churchill, Vic. 3842
Ph 03 99026942
Fax 03 99026578
Email: [email protected]
Abstract
A framework incorporating multiple modalities is proposed for cognitive research into online learning interaction. Its fundamental interaction dimensions are taken to be various combinations of text, graphics, video, sound and immersive virtual reality. These dimensions are then considered at two further levels of analysis, distinguishing between 1-way and 2-way and synchronous or asynchronous interactions.
Useful ways for siting current cognitive learning research and new research directions among these framework components are proposed. Methods for improving learning of cognitively demanding content by reducing the learning cognitive load in these online contexts are derived from recent cognition research. These are goal-free problem solving, worked and completion examples practice, reducing split-attention and redundancy effects, relating the heavy use of worked and completion examples to learners' expertise, the imagination effect, using multiple modalities and identifying and employing optimal conditions for discovery learning.
The implications and scope of effective new learning approaches for online learning and opportunities for research are discussed. The current research is argued to have immediate implications for improving learning in non-online contexts, and be highly suggestive of fruitful areas of research and development in online learning environments, especially if multimedia or virtual reality are employed as interaction modalities. It is also argued that the existing cognition research provides powerful methods of planning, prioritising, developing, implementing and evaluating online learning using judicious combinations of multiple interaction modes and instructional methods.
Introduction TO INTERACTION MODALITIES FRAMEWORK
Some years ago I suggested a very basic classification of online education and training which I will use as a starting point for this presentation in a modified form.
The simplest level of online education and instruction is via text only, such as in archived gopher materials, email, etc. At the next level we may have text with graphics, such as in a typical Web page. Alternately we may have sound-only messages, such as in radio broadcasts (one-way), or telephone (two-way). We could combine text, graphics and sound, as in audiographics or Web pages, and include video, which may be synchronous or asynchronous, and allow for one-way or two-way interaction. However, if we really want to immerse a student in a learning environment, we would use virtual reality enhanced with full surround sound.
We may represent these categories as a diagram, which captures these distinctions in nine cells.
Figure 1 Online eduction and training media
However, all the above online interactions may occur in either synchronous or asynchronous manner, and may include one-way or two-way interactions. We may extend the above classification to incroporate these further dimensions as shown in Figure 2.
Figure 2 Online instructional interactions
Thus the possible categories of educational online interaction range from pure symbolic text or sound presentations in 1-way, asynchronous mode to the totally immersive Virtual Reality with sound 2-way, synchronous interactions. In this paper an attempt is made to specify where in this categorisation obvious applications may be found for the recent cognition research findings. I will also point out the research and development opportunities I see within the online educational technology and cognitive research.
Introduction to research
My educational cognition research over the last few years has been conducted in the Information Processing framework. It has drawn heavily on both Schema Theory and Expert-Novice studies. An important aspect of the Information Processing framework is to see the mind as a computer. The three basic computer concepts, input, processing and storage, in particular, have been useful in developing theories with significant learning and teaching implications.
The basic principles of this line of thinking are based on the three-store model of the human information processing system .
Figure 3 Three component model of mental processing
That is, the human mind receives information from the outside world through the senses (Input stage) which are decoded in the Sensory Memory. The information from the Sensory Memory is then processed in the Working Memory (Processing stage) and stored in the Long-term Memory (Storage). Of course the previous information stored in the long-term memory can also be drawn on, or activated by the working memory to help deal with the
processing in working memory . The processing in the working memory is what we commonly call conscious thought .
One of the most interesting and significant aspects of the human mind is the very small capacity of the working memory. In 1956 G. A. Miller coined the famous term, "the magical number seven plus or minus two" to describe the number of distinct items he thought humans could hold in the working memory at any time . Since then the exact number of items has been shown to depend on a number of factors, ranging from age, health, level of fatigue, the type of item, familiarity with the content, training, etc. . However, without doubt the capacity of the working memory to deal with distinct items is quite limited, whereas the capacity of the long-term memory is very large, in fact no clear boundary has been established for it .
There are two main mechanisms that improve the working memory operation. They are Schema Formation and Automation . If the items of information are grouped together in a meaningful way, e.g. 29011981 vs 29.01.1981 (my daughter's birthday) they become easier to remember and use as one item, called a chunk . Instead being eight separate items, they can be treated as a single entity by working memory. The structure we have used to link these numbers together, the date notation, is an example of a schema, which helps to organise and simplify mental processing.
The second mechanism is automation. By this I mean processing that is so familiar that one does not have to think about the components of the processing consciously. This is the type of processing fluent readers use when reading text, where they do not try to make out individual letters, but process larger groups, words or groups of words, without attending to individual letters or even words separately.
As one develops better schemas and automation one gains expertise in a given field. Then one is able to select and use more elaborate schemas and automated processes to avoid the bottleneck of processing in the working memory, with too many individual separate items of information leading to confusion, and poor processing, due to a mental overload.
Cognitive Load Theory
Based on this view of the human information processing a number of significant advances in improving learning have been developed under the banner of the Cognitive Load Theory . Principally the Cognitive Load Theory work has been aimed at reducing the cognitive load on the working memory during learning.
A. Reduction in extraneous cognitive load
Goal-Free Problem Solving
The first method developed from the Cognitive Load Theory is called the Goal Free Problem solving . For example in the kinematics topic in physics a small number of equations of motion under constant acceleration are taught in senior high school. After being the taught the equations of motion, the students are commonly given problems of the type:
If a body starts from rest under the action of a constant acceleration of 3 m/sec2, find the distance moved over 4 seconds.
This requires the student to manipulate the equations of motion, combining them and substituting known values for the variables until they find the value of the distance variable.
The following alternative way of presenting the same problem was tried:
If a body starts from rest under the action of a constant acceleration of 3 m/sec2, find all the things you can know about the body after it has been in motion for 4 seconds (using the equations of motion).
It was found that the second type of problem led to far more effective learning. This was called the goal-free problem solving. The reason for the improvement in learning was due to reduction in the use of less effective means-ends analysis approach involved in common problem solving. In means-ends analysis the students are comparing the current state of their problem with the final end to be reached, and trying to move closer to the final end, stage by stage. Thus they not only have to attend to the details of the problem, but also to start, end and mid stages. In the goal-free problem solving they can concentrate only on the problem itself, and the next stage only. Thus they have more working memory capacity available to develop better schemas for the material to be learned. In a sense this is a form of Discovery Learning - in fact, one of the few effective forms of discovery learning, according to my research!
However, there are limitations to the effectiveness of goal-free problem solving, thus the mechanism involved and the context where it is to be used need to be thoroughly understood before assuming it will improve learning in a particular context.
This method has been tested and found beneficial in text, and text with graphics one-way, asynchronous presentations in conventional classroom contexts, as shown in Figure 4.
Figure 4 Goal-free problem solving contexts
This instructional approach has not been used in online contexts yet to my knowledge, and so there are research opportunities to test the effectiveness of this method in all the online technology contexts noted above in Figures 1 and 2.
Worked Examples
The heavy use of worked examples, instead of asking students to simply work through problems after a lesson, was found to improve learning , particularly for complex materials. This means that students were asked to study how given problems were solved, by being provided with the problems and fully worked solutions, before they were asked to solve similar problems. Interestingly both the ability solve the same types of problems and other related problems was always better than, or at least as good as for conventional lesson-problems solving approach.
This was another example of minimising the working memory load in learning by reducing means-ends analysis. The search for solutions in conventional problem solving took up so much of the students' working memory space, there were not enough cognitive resources available to learn the schema or structure of the content material. This technique has been found to be effective in a wide range of situations.
The worked examples practice has also been conducted in the text only, or text with static graphics presented in non-online, 1-way, asynchornous contexts, leaving open plentiful opportunities for research in other areas as shown in Figure 4 for goal-free problem solving.
Integrated Instructions
In presenting complex study material in text and graphical formats a learning improvement was found if the materials were integrated, instead of appearing in separate locations which needed to be kept in the limited working memory . This integration was found to be particularly effective for material that required both the text and the graphics to be understood together. The integration also helps in a situation where different parts of a text presentation refer to the same issue, such as a conventional psychology report, where the method, experiment results and discussion refer to the same concepts. If these were integrated, students studying the reports learned the material more effectively.
A particularly interesting application of this principle was found if students were studying to understand a computer program from a textbook, such as a software manual. If students use a textbook at the same time as they work on a computer, e.g. entering exercises from the book on the computer, a detrimental split-attention condition is set up. Instead the researchers working on the basis of the cognitive load theory separated the manual study and computer operation phases and found a significant improvement in learning .
The manual needed to show the steps in a typical activity to be accomplished, e.g. control of a CAD/CAM system using a computer, the screen responses and the operator actions. If the book described these items clearly enough, it was found to be more effective for the students to study the book only, before working on the computer, rather than try to enter information on the computer as they read the book for the first time. The moral of the story for learning complex material is: turn the computer off, read the book (so long as it describes the process thoroughly), and then work through similar problems on the computer.
The integrated instructions have also only used text only, or text with graphics presentations employing 1-way asynchronous, non-online modes as shown in Figure 4 for goal-free problem solving, leaving opportunities for significant research and development in online contexts.
Expertise
It was also found that the optimum format of information presentation changes with the development of expertise . When a student's competence in a particular field improved, e.g. in understanding electrical circuit diagrams, the need for integrated information presentation and structured worked examples reduced. So at the beginning of a topic the students needed to be furnished with many worked examples, with integrated graphics and text. However, towards the end of a substantial study period, perhaps lasting many weeks, they no longer needed as many text explanations, the graphics were adequate to convey the meaning on their own.
Redundancy
As the interest in integrating text and graphics increased a reverse effect was found - adding text to some graphics actually reduced learning efficiency! . On further investigation it became apparent that the graphical presentation (such as the human circulatory system diagram with arrows indicating blood flow) was sufficient in its own right, and adding any more text actually increased the working memory load rather than reducing it, by introducing an extraneous cognitive load, i.e. redundant material.
Thus the suitability of graphical and integrated text-graphics formats need to be carefully considered before they are used in study presentations. For example in experiments investigating the learning of new software, spreadsheets, via computer presented tutorials, if the material presented on the computer screen was sufficiently informative in its own right, expecting students to use a manual or a textbook at the same time made learning more difficult, due to redundant information overloading the students' working memories . The moral of this story is: if the tutorial on the computer is sufficient, do not expect students to work from the tutorial and a manual or workbook at the same time, even if they teach exactly the same content. Work through the screen-based tutorial by itself first, and only then refer to other written support materials. Since this approach has only been used with text only, or text and graphics information presentation formats, employing non-online, asynchronous modes as shown in Figure 4 above, it needs to be tested in a multitude of online learning contexts.
B. Increase in working memory capacity
Instead of focusing on the reduction of the extraneous cognitive load of a study task, a useful alternative approach to improve learning was found to be the increased effective use of the full working memory capacity. Baddeley found the working memory consists of three main components: the phonological loop (space for processing sound/verbal information), the visuo-spatial loop (space for processing graphical/text information) and the Central Executive (the monitoring/planning space). Mousavi, Low and Sweller and Tindall-Ford, Chandler and Sweller compared the use of two modalities, i.e. sound and visual information presentation together, with single modality presentations.
Generally they found that using graphics together with verbal explanations was more effective than graphics and text together. They explained this effect in terms of the working memory being able to process more material if it was presented in two different modalities due to working memory having separate parallel processing spaces for the two differently conveyed information types. These processing spaces could cope with more information together than if either one was engaged on its own. This increase in the amount of material learned in two different modes was called the Modality Effect.
This is a particularly significant finding for multimedia education. Presentations consisting of graphics or text supported by auditory explanations are more effective for complex content learning than either visual or auditory lessons alone. However, if either the visual or the auditory material is comfortably understandable in its own right, then the other material is redundant and interferes with learning. The value of visually presented computer-based instruction being supported by auditory material would be particularly important in situations where the visual material cannot convey all the content on its own, such as in learning a foreign language. The visual display may convey the text of the new words and the auditory channel can present the appropriate pronunciation. This is also important many areas, such
as in learning new computer terminology, where new technical terms are introduced, whose pronunciation is not obvious from the written text.
Recently Mayer and Moreno examined the integration of sound, text and animation in computer presented learning and found that animation with sound was more effective than animation with text, showing that this principle applies particularly effectively to multimedia eduction. None of this work has been carried out in online contexts yet.
Figure 5 Modality effect
C. Reducing intrinsic cognitive load
Instead of trimming the extraneous cognitive load, the non-essential parts of the study materials, Pollock, Chandler and Sweller found that they could reduce the cognitive load due to the essential parts of the learning tasks. They did this by firstly having the students (electrical apprentices) learn the essential basic component tasks in a rote learning mode (measurements of resistance in the earthing of electrical appliances). They then combined the rote learned components into a more sophisticated schema by a learning with understanding approach (how to test the appliances for earthing safety). They found that this two stage process was more effective than a learning via two understanding stages.
Figure 6 Rote vs Learning with Understanding
This is very recent research and the only experiments that have been carried out to examine this issue have involved text and graphics presentation. Clearly there is huge scope for further work, especially in all the online contexts, as shown in Figure 7.
Figure 7 Rote followed by Learning with Understanding contexts
D. Development of automation
The Imagination Effect
Cooper and Tindall-Ford compared the learning of spreadsheet operations between students using the most effective approach they knew, worked examples study, with the use of imagination. They found that if the students were presented with examples of study materials about spreadsheet use in computer aided instruction mode, and then asked to either study the materials for a short time, or imagine themselves working through the steps, the imagination approach produced a clear improvement in learning time and test scores in three out of four cases.
In the one experiment where there was no improvement, the students were found to not have understood the basic steps involved in the processes, thus the request to imagine the process was not effective, because they did not have the basic building blocks for the processing. The imagination effect was a way of developing a more effectively automated use of the spreadsheet operational components, rather than contributing to schema development.
Thus the very recently discovered imagination effect has been shown to be beneficial in the text with graphics context as for the Rote followed by Learning With Understanding approach shown in Figure 7. There is again plenty of room for further research with regard to this effect in all online presentation modes.
E. Discovery Learning and Cognitive Load Theory
My own experiments shed some light on the relationship between Discovery Learning and Cognitive Load Theory . Generally it appears that Discovery Learning has been oversold, like a used car by over-enthusiastic salesmen. The father of Discovery Learning, Bruner , called Discovery Learning 'the most inefficient technique possible for regaining (i.e. learning) what has been gathered over a long period of time..'. Wittrock questioned the confusion between means and ends in Discovery Learning investigations, and the lack of decent
research in the area. Cronbach further questioned the specificity and benefits of Discovery Learning research conducted so far, suggesting better alternatives for experimental work. Thus even though Discovery Learning has had a significant following since the late 1950's, its value is in serious question.
One of the ways to reduce extraneous cognitive load, goal-free problem solving, could be considered a form of Discovery Learning. In this instance one is trying to explore alimited search space given the starting point.
In my experiments I compared exploration practice with worked examples practice for students learning to develop calculation formulas for a database program, FileMaker Pro. In each case the students were given the same lesson by a computer-aided instruction, a HyperCard stack, and the same time to practice the development of calculation formulas on computer. The creation of calculation formulas is a complex task. An interesting outcome was obtained. If the students had even minimal experience with other database programs previously, both the exploration (group test mean = 35.9) and worked examples practice (group test mean = 30.9) were effective for them. But for the students without previous database experience the exploration (discovery learning) practice was quite ineffective (test mean result for group = 15.1), whereas the students reading the worked examples and then solving problems learned as well as the students with previous database exposure (group test mean result = 29.6).
Figure 8 Previous database exposure and practice interaction
This complements nicely the goal-free problem solving results. In the goal-free problem solving the limited search space allowed the exploration ('discovery') to produce useful schema development. In this case the previous work with databases provided the raw materials for the practice, the relevant schema and some automated database procedures, which allowed the students to gain effective benefit from the exploration practice, as well as the worked examples practice.
Thus when does the discovery learning approach produce effective learning? On the basis of the cognitive load theory, and the experiments reported here, we can say it is useful if the search space for the practice task is limited, or when the students already possess appropriate schema or automated procedures that can be applied to a new context. Apart
from that it is mostly an inefficient, overly time-consuming learning approach, possibly leading to confusion and even killing motivation of students involved in using it.
The discovery learning experiments have so far involved text and graphics presentation, as shown for other effects in Figure 7. So clearly work needs to be carried out to test these results in all the other online technology contexts.
Conclusion
Most of the work reported here has used either text only and text with graphics one-way, asynchronous presentation of information. Thus for those aspects of online instruction where these conditions apply, we can expect to gain useful benefits from applying the above lessons directly. However, what might be benefits of these approaches in the different online contexts, where sound is used in conjunction with text, etc.? What would be the outcomes of these methods if Virtual Reality was used online? What about employing two-way communications in these modes, either asynchronously or synchronously? By no means has the current research yet exhausted all the areas. In fact if we were to look at our technology interaction contexts for one final time, we note that most of the areas are still open for discovery and exploration, as shown in Figure 9, especially in the online education.
Figure 9 Interaction contexts investigated so far (shaded)
Another way the cognition research discussed in this paper could be used to improve online learning is to use its concepts and techniques to identify and prioritise educational improvement efforts. For example, if we wished to improve online learning outcomes we can use element interactivity and learning efficiency analysis methods, developed in the cognitive load theory context, to identify topics or course components where the cognitive load is the greatest. Having identified the course components in need of improvement, we may use cognitive load measures to rank the components into a priority order, and employ cognitive load measurement techniques to evaluate online educational programs.
Thus cognition research provides ways of identifying and prioritising online learning needs, as well as suggesting robust techniques for improving and evaluating online education. Overall, the online dimensions of synchronous vs. asynchronous, and one-way vs. 2-way interaction dimensions, coupled with the nine identified forms of educational mediation, open up a multiplicity of contexts for improving online learning, where existing cognition research may be applied or extended.
References
