TERRY MAYES

Institute for Computer-Based Learning

Heriot-Watt University, Edinburgh EH14 4AS

This paper discusses the idea of mindtools for learning. These are essentially comprehension tasks which require a learner to analyse material at a deeper conceptual level than would normally follow from a simple instruction "to learn". Deep learning results as a kind of by-product of the mindtool task, as indeed it does from any such analytical search for meaning. An empirical and theoretical underpinning for this approach is provided in the cognitive psychology literature by the levels of processing approach, and by the enactment effect. An attempt is also made to position mindtools in the context of instructional theory, and an example is given of a mindtools approach to learning from hypermedia.

"Subjects remember not what was 'out there', but what they did during encoding"

Craik and Tulving, 1975

The concept of a mindtool, as used in the workshop that has produced the current volume, is easy to describe. It is simply a device, or technique, for focusing the learner's analytical processes. A mindtool can be regarded as an instructional technique in so far as it involves a task, the explicit purpose of which is to lead to active and durable learning of the information manipulated or organised in some way by the task. The primary task is not learning per se. To instruct someone 'to learn' is in effect to say, 'perform some activity which results in understanding of, and durable memory for, this material'. Our idea of a mindtool is of one that gives the learner just such a ready-made task.

The mindtool concept also carries the implication, as with any tool, that the user will become more skilful with practice, and the tool will therefore be more effective in the hands of an experienced user. This definition is broad enough to encompass a wide range of activities as mindtools, ranging from verbal debate to the playing of computer games. The idea is predicated on the assumption that learning is not a particular, discrete activity that can be turned on and off. Rather, learning is an inescapable by-product of comprehension. Nevertheless, we have to acquire techniques for comprehending. Many aspects of human cognitive development can be regarded as the acquisition of such techniques.

It is helpful to consider this at two levels. First, there is the development of skills for making sense of the world. These are metalearning processes. Then there is the business of actually employing these in the acquisition of knowledge. The idea underlying the present workshop was that computers can offer interactive tasks which are effective at both these levels. The result is assumed to be a deeper level of comprehension of the currently analysed material, and as a consequence of this, deeper learning and thus more durable memory for that material. There is also the benefit of an improved capacity for such thinking in the future.

The cognitive psychology of mindtools

"...it is legitimate to say that all the cognitive processes that have been considered, from perceiving to thinking, are ways in which some fundamental 'effort after meaning' seeks expression."

F C Bartlett, 1932

Anderson (1990) describes a personal experience which illustrates the essential nature of learning very well. In a verbal learning experiment in which his sophomore class was required to learn paired-associates such as DAX-GIB, Anderson, determined to outperform the rest of his class, tried to burn the information into his memory by an intense process of repeating the paired associates over and over to himself as quickly and loudly as possible. By this method he achieved the worst performance in the group. His method neglected to make the pairs meaningful in any way. In fact, countless experiments have now demonstrated that meaningful information is better remembered than meaningless information. Only a very small amount of information can be retained by rote, and only for as long as it can be maintained in working memory by conscious attention. As soon as attention is diverted it is lost. It is almost impossible to recall anything if it has no meaningful structure to guide retrieval. By the same token, it is hardly possible not to learn something which has provided meaning. In fact, it does not seem to matter much whether there is an actual intention to learn or not. There is a long history of experimental findings on learning that demonstrate this.

It is evident that we learn all the time without making any special effort. Ask football devotees to tell the results of last Saturday's matches and not only will they do so, usually in great detail, but they would be astonished to be regarded as having had to make any kind of effort to learn that information. Once a sufficiently rich framework of understanding is in place, the pickup of information is indeed effortless. In fact, the whole concept of "processing" information has rather too deliberate and intentional a connotation. As we build a framework, or schema, for comprehension, we build a mechanism for automatic learning. New information is simply an elaboration, or a filling in of the slots, of what is already understood. No 'effort' is involved, beyond attending to the information in question. It is not so surprising, then, to find that learning done with the intention to remember is hardly any more effective than that done without such intention. Research on this question compares intentional learning with incidental learning.

A well-known experiment that illustrates the role of intention was performed by Hyde and Jenkins (1969). In this, the subjects were read a 24-word list and were subsequently asked to free recall as many of the words as they could. There were seven different groups of subjects, each receiving different instructions before hearing the list. One group was given intentional learning instructions. They knew that they were to be asked to recall the words subsequently. Three of the groups were not told that they would be asked to recall the words. They were given 'orienting tasks' which ensured that they would pay attention to the words as they were presented. One group rated each word, as it was presented, on a 'pleasantness-unpleasantness' dimension. The second group had simply to decide whether each word contained a letter "E". The third group estimated the number of letters in each word. Three remaining groups were given mixed intentional and incidental instructions. They were required to perform one of the three orienting tasks, but in addition were told that they would have to recall the words. From the results it was clear that learning was as effective when words were rated for pleasantness as when subjects were told to learn the words. The other two orienting tasks - detecting E's and estimating the number of letters - produced poor learning. Also, when combined with the intentional learning instructions, the inefficient orienting tasks interfered with learning. The intention to learn, per se, seemed not important. Much more important was attending to the material in a particular way. A pleasantness-unpleasantness judgment requires thinking about the meaning of each word. The other tasks merely require thinking about surface or structural features. These apparently interfere with the processing of meaning that is necessary for learning.

Other studies of incidental learning have shown that people often remember rather little about familiar objects. For instance, the study by Nickerson and Adams (1979) demonstrated the remarkably poor visual memory most subjects have for the detailed features on the face of a familiar coin. Of course, such detailed features are quite incidental to the functionality of a coin, receive only shallow processing and are therefore poorly remembered. This argument suggests that although it may contradict our model of our own minds that we seem to notice so little about the environment we "know", it is functional not to process the meaning of everything that impinges on us. An HCI example of this was provided by Mayes et al (1988b) who tested what users remember of the detailed content of the MacWrite interface. They found that even experienced users can recall little of the menu contents, even though during use those menus are the instruments of their successful performance. It seems that the necessary information is picked up, used, and discarded; it is not learned in the sense that commands are learned. More exactly, users retain only enough information for recognition, not the much greater amount required for recall. This study differs from previous studies because the learning which the subjects seemingly failed to do was not incidental but apparently central to their purposeful and skilful behaviour. Thus, it seems that users do not learn even things which are vital to their performance if they reliably find them in the environment when needed. Much of the "knowledge" that underwrites their performance seems to be left in the world, which is therefore used as a kind of extended memory. There is no point in 'comprehending' the visual detail on the face of a coin in order to use the coin. Similarly there is no need to comprehend the detailed features of a computer application, unless those features are crucial at the functional level of the user's task.

We learn as a by-product of understanding. Yet we can normally get by with 'understanding' less than we may like to think. Much successful performance can be based on an interaction with information in the environment. Only by requiring a learner to perform explicit comprehension tasks, where deep processing is necessary to complete the task, can we be sure that the learner is not constrained by the context of the particular learning experience.

The basic point here, as with much else in psychology, was succinctly made by William James (1890) who wrote: "...the one who thinks over his experiences most, and weaves them into systematic relations with each other will be the one with the best memory..". In its modern form in memory theory, the notion is one of levels of processing.

THE LEVELS OF PROCESSING APPROACH

Craik and Lockhart (1972) argued for the understanding of human memory as a by-product of perceptual analysis and that the durability of memory would be a positive function of the depth to which the stimulus has been analysed. Thus, "..deeper levels of analysis are associated with more elaborate, longer lasting and stronger (memory) traces...". Normally only the results of the deeper analyses can be regarded as learning, the by-products of preliminary or 'surface' analysis are discarded. What is needed later is meaning, and the extraction of meaning involves the deeper levels of processing. Craik and Lockhart viewed processing levels as a continuum of analysis. At one extreme, sensory analysis in the visual or auditory analysis systems will give rise to memory traces that are transient and easily disrupted. At the other end of the continuum, the process of semantic analysis will lead to more or less permanent memory.

The levels approach assumes that information that seems immediately meaningful, perhaps because it is highly familiar, is easily remembered because it is compatible with previously existing cognitive structures. Such material will be easier to process to deep levels, and faster, although speed of analysis is not itself a very good predictor of subsequent retention. Depth of processing will be affected by several things: the amount of attention given, the relation to existing cognitive structures, and the amount of time available for perceptual analysis and processing.

Craik and Tulving (1975) undertook a long series of experiments in order to gain empirical evidence and to refine the levels of processing approach into a full-scale theory. The usual procedure followed in their experiments was to present words to subjects and to ask a variety of questions designed to influence depth of processing. Shallow levels of processing were achieved by asking about the nature of the typescript, (is the word in capital letters?); intermediate levels by asking for a judgment about phonemic similarity (does the word rhyme with...?); deep encodings were encouraged by asking whether the word would fit into a certain sentence frame or semantic category (is the word a member of the following set...?). For both recall and recognition tests, the deepest level of encoding took the longest time and produced the highest subsequent retention. Actually, time to encode was shown not to be the critical factor. A complex but shallow task will take longer to perform but will still yield lower memory scores than an easy but deeper processing task.

Craik and Tulving carried out one experiment where their subjects were asked to judge the appropriateness of the target word to fit sentences of varying complexity. For example, does the word "pen" fit the sentence " she dropped the......", or can the word "watch" fit the sentence "the old man hobbled across the room and picked up the valuable...from the mahogony table"? There was a strong tendency for the more elaborated sentences subsequent to be better recognised. This, and other related findings, led Craik and Tulving to suggest that what is critical is not simply the presence or absence of semantic coding, but the richness or elaboration with which the information is encoded.

In the literature on the cognitive psychology of memory, the levels of processing approach led to a rather arcane debate about the mechanisms involved in producing the striking effects observed in the experiments. We need not concern ourselves with this level of theory[1]. The relevant point for the present discussion is that both empirical and theoretical underpinning for the idea of mindtools is evident in this work.

One further finding from this literature is worth highlighting. That is the enactment effect. Cohen (1981) showed subjects a series of objects. Under one condition they were asked to perform an action on each object. For example, they might be shown a match and then asked to "break the match". Subsequent recall was significantly higher if the instruction to break the match was actually carried out rather than simply being read. Other studies have confirmed that the enactment effect is is large and robust (Nilsson & Cohen, 1988). As Craik and Tulving put it: people remember what they did.

Mindtools and instructional theory

Glaser and Bassok (1989) have discussed the various instructional approaches that have emerged from the main thrust of work in cognitive science over the last few years: the concentrated work on competence, on the study of the growth of expertise in complex domains such as medical diagnosis, geometric proofs and computer programming. How do mindtools fit in?

THE DEVELOPMENT OF COMPETENCE

One line of instructional development is based on the study