How to design educational multimedia: a "loaded" question.

This article reviews a wide range of research literature and makes the argument that the existence of "cognitive load" is not necessarily a bad thing in the design of educational multimedia-as opposed to the design of software applications! tools. It also presents a model of educational multimedia learning that delineates different types of cognitive load induced by multimedia and evaluates their differential contributions-or, alternatively, hindrances-to successful learning. Suggested design solutions and techniques are offered based on empirical data, as well as logical outcomes of the proposed model.

**********

The model presented in this article integrates a myriad of research regarding--or indirectly related to-designing educational multimedia for optimal user comprehension and learning. This model is squarely based on theories/ideas/models running through the research literature in cognitive psychology, but does not hesitate to borrow from other lines of research--for example, in educational psychology, cognitive science, human-computer interaction, human factors, and so forth-when and where considered productive.

The author believes one of the reasons the development of such a model is so complicated has to do with the existence of parallel-though not necessarily independent-design considerations. Thus, it appears it is necessary to construct multimedia to fall within different constraints along different dimensions that are acting both individually and in concert with one another.

Simplifying the model to contain only a few critical dimensions requires a great deal of integration and condensation of existing studies and their findings, but can yield some rather complex-though reasonably functional-design dimensions.

Three broad factors or dimensions to consider when designing educational multimedia have been distinguished:

* regulating total cognitive load;

* regulating relative dominance of load types; and

* maximizing concurrent activation (i.e., simultaneous conscious attention) of relevant features of incoming information and (at the same time) relevant prior knowledge.

REGULATING TOTAL COGNITIVE LOAD

A commonly accepted principle in cognitive psychology and other social science disciplines is that working memory capacity-and attention-is fairly limited in humans, and when this limited capacity is overwhelmed, decreased levels of cognitive performance result (Sweller, 1988). The time-tested finding that working memory can contain or maintain around seven "chunks" of information, plus or minus two (Miller, 1956) is accepted, where these chunks might be thought of as schemata and where a more greatly articulated or developed schema can contain a greater amount of information than a lesser articulated or developed schema (Chi, 1988; Simon, 1974).

A common theme in software and multimedia design is minimizing cognitive load for the purposes of usability and learn-ability. The thinking is that keeping cognitive load low allows the user to interact with the software or multimedia without becoming overwhelmed or confused by the options or information being presented. When a software application imposes a large cognitive load on the user, it becomes difficult for the user to learn and/or use the application.

When a piece of software is an application or tool, this thinking makes sense because the application or tool is only a means for interacting with and/or creating content that can stand on its own independently of the application or tool. A user learns the application or tool to allow them interact with it and produce the actual content they are ultimately interested in. Since the capacity of working memory is limited (Miller, 1956), any load imposed by the tool reduces the amount of working memory capacity available for content-related thinking and processing.

In contrast to software applications and tools, educational multimedia actually contains content the user is interested in learning. In the current state of multimedia systems however, content cannot exist independently of a software method for navigating and accessing the content; in other words, a software interface is a necessary liaison between the user and the content he/she seeks to access and become acquainted with.

It stands to reason that the same goal of minimizing cognitive load for software applications and tools should apply to this software interface that must necessarily (at least currently) accompany the targeted content offered or dispensed by the educational multimedia software. As with a software application, the interface is only a means for getting at the content the user is ultimately interested in.

What about the presentation of the educational content itself? Should it also be dispensed in such a way as to impose as little cognitive load on the user-learner as possible--or spoon-feeding the information to users-students?

In fact, studies have shown that presenting content-to-be-learned in a manner that does not challenge students in some way can lead to shallower processing of the instructional materials yielding reduced comprehension of the targeted subject matter (Kozma, 1991; McNamara, Kintsch, Songer, & Kintsch, 1996; Shapiro, 1998). Of course, there is one complication mentioned in these studies: this effect (i.e., of decreased learning from less challenging instructional materials) is most evident in students with prior relevant knowledge regarding the targeted subject matter; students with no prior knowledge of the targeted subject matter tend to benefit from a simpler, low-load imposing presentation of the content; they do worse with the more challenging version of the presentation.

This is both good and bad news for educational multimedia designers. The bad news, of course, is that different learners require different instructional materials for optimal learning, making it effectively impossible to design one optimal static-sequential version for all learners. The good news is the discovery of a consistent principle: learning is enhanced by designing education multimedia that challenges learners without overwhelming them; this principle is true regardless of learners' level of prior content-related knowledge.

Other than complex, adaptive intelligent tutoring systems, is there a way to ameliorate these mismatches between learners and the presentation of the educational content? One possibility is the creation of multi-layered or leveled multimedia that allows the learner to seek out different paths that best correspond to their levels of previous knowledge, as well as their dynamically changing levels of knowledge and understanding gained as they proceed through the multimedia product. An analogy might be drawn with the common provision of color-coded difficulty levels for downhill skiers trying to find ski runs that match their abilities; these color-coded markers (which often include geometrical symbols--like diamonds, squares and circles as well) help skiers determine whether they've acquired the skill level-through enough previous skiing experience--to handle a particular ski run being offered. It's important to point out that these matches may change throughout the day of skiing. (1)

Another means for handling mismatches where a user feels overwhelmed is by providing external memory aids (e.g., an onscreen scratchpad) that allow users to divest themselves of portions of the load (Scaife & Rogers, 1996). The external aid can be used for holding notes (e.g., partial results, intermediate conclusions), drawings, and so forth, so space in working memory can be freed up for other processing and storage. The user can reactivate the information contained on the external aid in working memory simply by glancing back at it.

There is evidence that educational multimedia can stimulate increased levels of cognitive engagement between student and instructional materials (Stoney & Oliver, 1999). Multimedia has the ability to supply a lot of information all at one time (i.e., in parallel versus serially as when reading text); and accompanying this intrinsic ability to readily load sensory registers and working memory or attention, is the concomitant ability to engage and draw a person into the presentation. The more engaged the user is, the more likely they are to absorb some of the underlying content-related information even if the multimedia features attracting them are not directly relevant to the content (a phenomenon sometimes referred to as incidental learning).

The main point to take away from this section is that simply trying to keep cognitive load as low as possible may be inappropriate in many cases for designing effective educational multimedia. There appears to be an optimal amount of cognitive load--somewhere between minimal load and the load at which a user's cognitive processing catastrophically fails--that emerges from, and is mediated by, a particular individual's ongoing or momentary relationship to the instructional material (e.g., how much prior knowledge he or she has of the subject matter).

REGULATING RELATIVE DOMINANCE OF LOAD TYPES

Reflexive Load versus Intentional Load

A "reflexive load" refers to the automatic activation of representations in working memory in direct response to external stimuli. The stimuli that invoke a reflexive load are typically physically salient or dynamic and usually invoke a nearly automatic attention orientation response from the perceiver (Cowan, 1995); for example, many people find it difficult to ignore the moving images on a television in the room--even if they're trying to concentrate on something else. A reflexive load is the result of a passive, bottom-up (i.e., initiated by the stimulus) engagement. The salient and sometimes dynamic features of multimedia stimuli tend to impose such a reflexive load--especially compared to something like plain text. (2)

An intentional load refers to the activation of representations in working memory proactively initiated by the leamer-perceiver--that is, a load that is self-initiated. An intentional load is the result of an active, top-down (i.e., initiated by the perceiver) engagement. An example of this: a student is reading a difficult textbook and finds it necessary to re-read certain sections multiple times before they make sense; to understand these passages, the student must engage in active problem-solving type behavior. In this scenario, the student is proactively taking on processing load in order to gain greater comprehension of the text.

These two types of loads borrow from the same limited attentional and short-term processing resources--that is, there is no room for both a high reflexive and high intentional load; they must appropriate storage and processing from the same resource pool. The upshot of this is that a multimedia presentation that provokes a high reflexive load response is likely to "crowd out" the ability of the user to initiate a more intentional, proactive form of processing.

By its very nature, interactivity promotes intentional load by requiring user input. There is evidence that interactive multimedia presentations engender more learning than multimedia presentations without such interactivity (Kozma, 1991).

For entertainment purposes, a high reflexive load with little or no concomitant intentional load is acceptable and possibly preferred. For educational purposes, however, a lack of an accompanying intentional load usually translates into shallower processing of the underlying content and a reduced rate of learning. Thus, it is often advisable to "pull back" from the full range of high-powered features and possibilities available through multimedia when presenting content for educational purposes (Wickens & Baker, 1995). This is not to say that advanced features of multimedia never have a place in educational content; they can be very useful for presenting certain types of information (see next section on knowledge activation) or as a periodic means for keeping the learner engaged--especially when the content is particularly difficult or "dry."

Examples of media at two ends of the Reflexive-Intentional scale (Figure 1):

Figure 1

Media examples

Difficult Expository Text        Music Video

low reflexive, high intentional  high reflexive, low intentional

This idea of the importance of a self-initiated load for learning is central to theories of Constructivist Learning (Cunningham, Duffy, & Knuth, 1993).

Interface-Related Load versus Content-Related Load

Cooper (1990) stated that "Cognitive load theory suggests that effective instructional material promotes learning by directing cognitive resources towards activities that are relevant to learning rather than to processes that are an adjunct to learning." Thus (as alluded to in a previous section). it should be more productive to have a greater portion of the total cognitive load associated with the targeted educational content rather than the interface. For entertainment content, it's possible for the interface or navigation to actually be a facet of the entertainment.

It may also be possible for the navigation for educational multimedia to contribute something toward comprehension of the targeted subject matter, or for the navigation (or lack thereof) to be a means for engaging the learner so that they'll be drawn into the content (Shapiro, 1998). But relegating the interface to a very small portion of the total load--by relying on, for example, very standard icons, GUIs, and layouts--is most consistent with existing empirical data on the subject.

MAXIMIZING CONCURRENT RELEVANT KNOWLEDGE ACTIVATION

This principle may seem to run counter to the first principle of "Regulating Total Cognitive Load." But though knowledge activation and cognitive load are generally positively correlated, they are not strictly correlated; a greater amount of knowledge activation can be achieved without a concomitant increase in cognitive load. A particular level of cognitive load that engenders a greater amount of knowledge activation may be considered a more efficient load.

A greater amount of concurrent knowledge activation is preferable (when possible without overloading working memory) because it maximizes the learner's opportunity for knowledge integration among simultaneously activated knowledge in and through working memory. It is assumed here that working memory can contain both representations of knowledge, and pointers to knowledge representations contained in long-term memory (Ericsson & Kintsch, 1995). The more effective a pointer is, the greater the amount of relevant long-term knowledge that is activated (or "pointed to") by it. Also, the greater the potentially relevant knowledge activated at any one time, the greater the opportunity for the learner to notice patterns and connections across or among these knowledge representations that can lead to better integrated mental models or deeper, more comprehensive understandings of the subject matter (Mayer, 1997) (Figure 2).

Integration can occur between incoming information and prior knowledge (i.e., linking the new with the old; a sort of vertical integration), or it can occur across the incoming information (e.g., among its different modal representations; a sort of lateral integration [Paivio, 1986]). A greater amount of knowledge integration translates into a greater amount of comprehension of the subject matter to which the knowledge relates.

What kinds of external representations enable the greatest amount of manageable internal knowledge activation? Answer: Those that enable the most efficient and relevant activation of knowledge in or through working memory (again, activating as much knowledge in short and long-term memory as possible without overwhelming working memory).

Multi-modal representations (existing externally or internally) have the potential for being the most efficient means for storing and activating knowledge. A number of studies have suggested that human beings have independent parallel systems for processing different aspects of a stimulus; there is evidence for dissociations among visual, spatial, auditory, and verbal information as it is processed by the human brain (Baddeley & Logie, 1999; Shah & Miyake, 1996; Logie & Marchetti, 1991). Since a human can process these different types of information in parallel, a stimulus can combine these different types of information without necessarily increasing the corresponding cognitive load that is induced. In fact, Mousave, Low, and Sweller, (1995) described this effect as "increasing the effective size of working memory" by using multiple rather than a single modality.

A key factor that is just beginning to become evident is that targeting multiple modalities is most effective when the multi-modal information is viewed as a single, integrated representation by the perceiver--for example, an animation with integrated voice-over narration leads to better comprehension and recall than when narration is presented before or after the animation (Moreno & Mayer, 1999). Even when all of the information is in one modality, comprehension, retention and learning are enhanced when the different pieces of information are perceived as a unified whole (e.g., an experiment and literature review by Moreno & Mayer, 1999] concluded that "Students learn better when onscreen text and visual materials are physically integrated rather than separated"). In contrast, information that is presented in such a way so as not to appear integrated creates a greater cognitive load accompanied by relatively less overall knowledge activation.

There is also some evidence that certain types of representations are better suited for particular types of information. For example, Larkin and Simon, (1987) showed that diagrams can be much more efficient for representing geometry problems than can sentential, verbal descriptions. Hegarty and Just, (1993) discovered that subjects required a diagram of a machine configuration (in addition to a text description) to construct a functional mental model of the machine. There's also some indication that dynamic information is best communicated by animations or video (Hansen, Narayanan, & Schrimpsher, 2000).

Other benefits of multi-modal representations include information features attuned to the different cognitive styles of different learners (Liu & Reed, 1994), and multiple types of retrieval cues for accessing a greater range of potentially relevant prior knowledge--since retrieval cues for accessing knowledge can be of very specific types or content depending on how they were originally encoded (i.e., "encoding specificity" [Tulving & Thomson, 1973]).

Control of Attention

Attentional and working memory constraints demand the strategic allocation of their processing and storage resources. Naive students who don't know yet "what is important" need attentional control assistance (e.g., pointers, arrows, schematic diagrams versus full-detail images, etc.); they don't have the needed experience or background knowledge to effectively parse complex displays or multimedia information into effectively meaningful chunks of information. Part of knowing what is and what isn't important independent of knowing something about the subject matter, is being multimedia literate; this goes for both the designers and the users.

By concentrating activation within relevant knowledge versus irrelevant knowledge, more meaningful or critical patterns and connections are likely to be detected by the learner. There is also a minimization of unnecessary or unproductive cognitive load.

In the words of Cognitive Load Theory (Sweller, 1988) effective instructional material facilitates learning by directing cognitive resources toward activities that are relevant to schema acquisition. Instructional material that requires learners to direct cognitive resources to activities other than schema acquisition may thus be viewed as inferior.

Directing cognitive resources to the most central or important aspects of the instructional materials can occur from within the learner or from without. It's primarily a matter of attentional control. The importance of attentional control is the need to suppress or deactivate irrelevant knowledge that might be reflexively activated by the many salient and dynamic features multimedia has to offer.

Factors that determine the learner's internal attentional control are previous experience with both (a) the targeted content (i.e., the subject matter being taught) and (b) the medium through which it is being disseminated. Either type of preexisting experience assists the learner in knowing which aspects or features of a presentation to focus on to best gain new knowledge or information from it. As we alluded to before, being familiar with how to use and interpret multimedia is a facet of what is sometimes referred to as media literacy. The result is that users with previous experience with the topic or the medium require less guidance about where to focus for optimal learning or information uptake.

A number of studies have shown that naive learners-ones that have little prior knowledge of the topic and/or the medium-require external assistance (i.e., via special design of the instructional materials) to be able to focus on the features or parts of the presentation that engender a productive learning experience; this appears to be especially true in the case of presentations that maximize usage of the wide range of capabilities multimedia has to offer.

At least a couple of studies have shown how less (content) can yield more (skill and knowledge). Lazonder and Van der Meij, (1993) found that "Minimal Manuals" (i.e., manuals that provide concise step-by-step procedural instructions for using software without including accompanying explanations and background information) yielded better procedural skill faster, with better error-recovery than manuals that did provide such explanations and background information. Mayer, Bove, Bryman, Mars, and Topangco, (1996) discovered that "multimedia summaries" (i.e., text + drawing summaries of scientific textbook passages) led to significant enhancements of learning by focusing the learner's attention on the most crucial points in the text passages.

A study done by Jeung, Chandler, and Sweller, (1997) is particularly relevant to controlling attention when presenting multimedia that requires a high degree of visual search. They found that presenting information in text + diagram form was inferior (in terms of the subject learning the underlying topic) to presenting the information by way of text or diagram alone when inferring connections between the text and diagram required a lot of visual search. Only after they added visual indicators (e.g., arrows) to show the portion of the diagram that corresponded to the portion of text the subject was reading, did the subject learn the topic better compared to viewing either the text or diagram alone.

A central lesson of this section is that maximizing relevant knowledge activation often means minimizing irrelevant knowledge activation.

CONCLUSION

Designing instructional materials with multimedia is a bit like walking a tight rope: the designer tries to provide a rich, informational learning experience for the learner while at the same time not exceeding his or her ability to process and assimilate the information. Add to this the complication that some amount and types of working memory load are actually good for learning, while others are not. In other words, not all cognitive load is bad; and efforts at reducing cognitive load to minimal possible levels may be misguided in regard to the design of educational multimedia. (This is in contrast, for example, to the design of usable/learnable software applications and tools).

In this article, the authors attempt to reduce the stigma attached to the existence of cognitive load in multimedia designs--at least when the multimedia is being designed to serve educational purposes. By examining the nature and even the potential advantages of cognitive load in more detail, the hope is to stimulate a greater sensitivity and awareness of the complex and subtle considerations involved in the design of effective educational multimedia.

[FIGURE 2 OMITTED]

Notes

(1.) One complication here is the dubious ability people may have to accurately gauge their own skill or knowledge level regarding a particular subject or task (Sternberg & Kolligean, 1990).

(2.) It is sometimes possible for humans to become accustomed to certain reflexive loads and to learn to suppress and/or screen them out.

References

Baddeley, A., & Logie, R. (1999). Working memory: The multiple component model. In A. Miyake & P. Shah (Eds.), Models of working memory: Mechanisms of active maintenance and executive control. New York: Cambridge University Press.

Chi, M. (1988). Overview. In M. Chi, R. Glaser, & M. Farr (Eds.), The Nature of Expertise. Hillsdale, NJ: Erlbaum.

Cooper, G. (1990). Cognitive load theory as an aid for instructional design. Australian Journal of Educational Technology, 6(2), 108-113. Available: http://celo.murdoch.edu.au/gen/aset/ajet/ajet6/su90p108.html

Cowan, N. (1995). Attention and memory: An integrated framework. New York: Oxford University Press.

Cunningham, D., Duffy, T. M., & Knuth, R. (1993). Textbook of the future. In C. McKnight (Ed.), Hypertext: A psychological perspective. London, Ellis Horwood.

Ericsson, K., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102(2), 211-245.

Hansen, S., Narayanan, N., & Schrimpsher, D. (2000). Helping learners visualize and comprehend algorithms. Interactive Multimedia Electronic Journal of Computer-Enhanced Learning, 1(2) [Online]. Available: http://imej.wfu.edu/

Hegarty, M., & Just, M. (1993). Constructing mental models of machines from text and diagrams. Journal of Memory & Language, 32(6), 717-742.

Jeung, H., Chandler, P., & Sweller, J. (1997). The role of visual indicators in dual sensory mode instruction. Educational Psychology, 17(3), 329-343.

Kozma, R. (1991). Learning with media. Review of Educational Research, 61(2), 179-211.

Larkin, J., & Simon, H. (1987). Why a diagram is (sometimes) worth ten thousand words. Cognitive Science, 11, 65-100.

Lazonder, A., & Van der Meij, J. (1993). The minimal manual: Is less really more? International Journal of Man Machine Studies, 39, 729-752.

Liu, M., & Reed, W. (1994). The relationship between learning strategies and learning styles in a hypermedia environment. Computers in Human Behavior, 10(4), 419-434.

Logie, R.H., & Marchetti, C. (1991). Visuo-spatial working memory: visual, spatial or central executive? In R.H. Logie & M. Denis (Eds.), Mental images in human cognition (pp. 105-115). Amsterdam: Elsevier.

Mayer, R. (1997). Multimedia learning: Are we asking the right questions? Educational Psychologist, 32(1), 1-19.

Mayer, R., Bove, W., Bryman, A., Mars, R., & Topangco, L. (1996). When less is more: Meaningful learning from visual and verbal summaries of science textbook lessons. Journal of Educational Psychology, 88, 64-73.

McNamara, D., Kintsch, E., Songer, N., & Kintsch, W. (1996). Are good texts always better? Interactions of text coherence, background knowledge, and levels of understanding in learning from text. Cognition & Instruction, 14(1), 1-43.

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.

Moreno, R., & Mayer, R. E. (1999). Deriving instructional design principles from multimedia presentations

with animations. Proceedings of the IEEE International Conference on Multimedia Computing and Systems, (pp. 720-725), Florence, Italy. Piscataway, NJ: Institute of Electrical and Electronics Engineers.

Mousave, S., Low, R., & Sweller, J. (1995). Reducing cognitive load by mixing auditory and visual representation modes. Journal of Educational Psychology, 87, 319-334.

Paivio, A. (1986). Mental representations: A dual coding approach. New York: Oxford University Press.

Scaife, M., & Rogers, Y. (1996). External cognition: How do graphical representations work? International Journal of Human-Computer Studies, 45, 185-213.

Shah, P., & Miyake, A. (1996). The separability of working memory resources for spatial thinking and language processing: An individual differences approach. Journal of Experimental Psychology: General, 125, 4-27.

Shapiro, A. (1998). Promoting active learning: The role of system structure in learning from hypertext. Human-Computer Interaction, 13, 1-35.

Simon, H. (1974). How big is a chunk? Science, 183, 482-488.

Sternberg, R., & Kolligean, J. (Eds.) (1990). Competence Considered. New Haven, CT: Yale University Press.

Stoney, S., & Oliver, R. (1999). Can higher order thinking and cognitive engagement be enhanced with multimedia? Interactive Multimedia Electronic Journal of Computer-Enhanced Learning, 1(2) [Online] Available: http://imej.wfu.edu/

Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12. 257-285.

Tulving, E., & Thomson, D. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80, 352-373.

Wickens, C., & Baker, P. (1995). Cognitive issues in virtual reality. In W. Barfield & T. Furness (Eds.), Virtual Environments and Advanced Interface Design (pp. 516-541). New York, NY: Oxford University Press.