Last changed 7 Sept 1998 ............... Length about 3,000 words (20,000 bytes).
This is a WWW document by Steve Draper, installed at http://www.psy.gla.ac.uk/~steve/interactivity.html. How to refer to it.

Content and Interactivity

Stephen W. Draper
Department of Psychology
University of Glasgow
Glasgow G12 8QQ
U.K.
email: steve@psy.gla.ac.uk

This was written as a response to discussion on the ITFORUM email list around 29 July 1996. (Link to list of ITFORUM participants.)

The original paper by Rod Sims, and discussion of it was some months earlier. A later message by Barney Dalgarno triggered off a discussion that led to my comments. Another web version of this note with better cross links is on the ITFORUM site.

Introduction

For me at least the recent exchange triggered by Barney Dalgarno on interactivity is very stimulating, as the conflicting views show me that I believe a number of things that seem contradictory. This long piece is my personal synthesis of the issues (in other words, stand ready to use the delete function as writing this was good for my learning, but reading it may not be useful to you). The way I read it, one group (Ian Hart) see it as just another instantiation of a contrast between medium and content, which they see as already resolved by Clark (1983) in favour of content determining learning and medium being irrelevant. A lone, but needed, voice said what about situation or context. Yet Barney, like Rod Sims, obviously thinks that a classification by types of interaction would have predictive power about learning outcomes. Interaction types are not quite the same as the medium vs. content issue. Learner actions are not just "medium", and learning depends not just on content but on the actions taken in relation to the content. What we have here, I think, is a long spectrum of hypotheses from the crudest versions of media determinism to classifications of learner actions. The crudest renderings of the issue are about sensory modality: is it better to receive material through the eyeball or the eardrum; then through the "medium" e.g. books vs. TV vs. computers; then the genre e.g. pictures vs. words. Then, where this discussion started, through different interactive modes that the computer can offer a learner. Behind this spectrum is another one from regarding instructional material as something that is done to the student to regarding the issue as about what actions the learner takes. Interactive modes are not quite the same as "media" because they more clearly concern learner actions which, unlike physical definitions of media, really do affect learning outcomes. Clark argued against the relevance of media as an independent causal factor in learning. Kozma's (1991) rebuttal of this as a general proposition amounts to the following: what matters to learning is learner actions, learner actions depend on particular instructional situations, and these cannot be separated from the properties of media that define the range of possibilities. So on Kozma's view, we can't dismiss the media factor just because some of its forms are unproductive; and the issue of interactive modes is at least more promising than the worst of these because it relates to learner activity. My view, then, is that the factors that really do determine learning outcomes are learner actions. The questions are: what kind of learner actions are the important types i.e. the main determinants of learning, and are computer-learner interaction modes of this type? I shall begin by arguing that classifications by interaction type usually are irrelevant to learning outcomes, but nevertheless this shouldn't stop us seeking a useful classification of learner (inter)action types.

CATEGORIES OF INTERACTIVITY

OVERT BEHAVIOURAL TRANSACTIONS

I have seen quite a few categorisations of learner-computer interaction by type, but I believe they are all of little use in education. The crudest are overtly machine-centered, that is they categorise interaction with humans in terms of a machine's technical characteristics. In the end, I think this applies even to Barney's and to Rod Sim's categories in his IT forum paper. The basic reason they are attractive both to computer scientists and to psychologists who like to measure overt behaviour is just that: that overt physical actions by humans or machines are unproblematic to observe and record. The trouble is that learning does not depend upon these things: you get learning with and without any of the interactive types mentioned. One example that shows this concerns TV. TV is completely non-interactive from a physical viewpoint: the viewer has no control over anything except perhaps viewing distance and sound volume. Yet occasionally it can support an intense learning episode. Some years ago there was a series on UK TV by Bryan Magee on "The great philosophers". Each week there would just be two men talking. By all the normal standards of TV production this was the least engaging TV ever, with the pattern on the wallpaper behind them the only thing of visual interest (actually I still remember that); but it had me on the edge of my seat straining in concentration, as it was pitched just at the edge of my understanding (I knew very little philosophy). This was one of the most intensely engaging TV experiences I have had, and ranks with my best learning experiences too, but would score zero on all these scales of interactivity and should probably not be used in any general way to advocate two talking heads as a general approach to good teaching and learning. It was interactive in the way that mattered (between my prior understanding and the ideas being discussed) but that interaction was completely invisible on physical and behavioural criteria. Although Barney's and Rod's schemes are much more sophisticated than the crudest in this area, I don't think they can escape this fundamental problem: observable interaction is just not the interaction that matters. A second case to ponder here is how the very same material (and hence its interaction modes) can have different effects on the same person on different occasions. A really great novel or film, but still more a good paper or textbook, can have a markedly different effect on you the second time you read it than the first. This is a fairly common experience, and again shows that observable interaction does not determine learning. A third case to consider is that posed by the literature on deep and shallow learning (Marton et al.; 1984). Here the same material has different effects on different learners because they interact with it differently, but again the interaction is not directly observable. It also holds a disturbing lesson for those who worry about motivation as a factor. On my reading, at least, it seems that students who want to pass tests say they goal is to learn, and these are the ones who do well on many tests but have short, shallow retention. Those who end up with long retention are those who were not trying to learn, but just to understand. In any case, the different learning outcomes depend upon different learner cognitive actions, but not on easily observable ones, and not on the interactive modes overtly afforded by the materials. Before leaving this topic, it should be said that such technical analyses of available interaction modes are not without their own interest if done properly. Consider an ordinary textbook. These normally offer at least five different kinds of interaction: they may be read sequentially (beginning to end), they have a hierarchical structure (e.g. part, chapter, section), they have an index for accessing by content, and they have page numbers which are routinely used by instructors assigning reading. In addition learners can and do make marks on them: highlighting, marginal notes etc. Very, very little software offers all 5 of these modes of interaction: computers have not in the main yet caught up with print as an interactive medium. And it is not so much the separate properties of each of these modes that matters, but that one product offers all these alternatives, so that one learner has multiple ways of interacting with the same material. Again, classifications of interaction types usually direct us away from this plurality and flexibility and in so doing are probably directing us away from a property crucial to supporting learning. As a transitional example, consider someone studying a map, diagram, or graph that is for them a rich source. Simple classifications say that such static printed representations are non-interactive, but that is wholly at odds with the subjective experience. Eye movements might seem to give the dedicated observer a handle, but this is misleading: firstly because what matters is attention, and this can move over an after-image independent of eye movements, and secondly because what really matters is the structures implicit in the printed representation (e.g. the relationship between two peaks on a graph, the slope of ground implied by countour lines). The interactions that matter are cognitive, not observable.

DEPTH OF PROCESSING

As we all know really (don't we?) when not distracted by technophoria, what matters for learning is depth of processing (not the number of mouse clicks, or the number of hyperlinks followed, or whether it is a simulation or a computer tutorial). So if we don't mind abandoning checklists of technical features and directly observable behaviour, then we could have a classification on internal interaction: by what levels of interaction are engaged. I think this has a real chance of predicting learning outcomes. It would begin with Craik & Lockhart (1972) levels e.g. syntactic vs. semantic processing, and continue with the shallow vs. deep learning distinctions (Marton et al.; 1984).

TYPES OF PROCESSING

Actually there is an interesting alternative theory, and hence a rival possible scheme for classifying interaction types. Perhaps what matters for promoting learning is the number of varieties of processing a learner does, rather than just "depth". What would be significant, then, about interaction and learner actions in general would be their requirement to use knowledge in a different way. In the simplest example, understanding what a teacher says requires one kind of processing (following links from words to meanings), re-expressing what they said requires another (creating links from meanings to words). This theory has at least two justifications. Firstly, many models of memory would predict that multiple different links to an item will lead to longer retention. Secondly, most definitions of "understanding" boil down to the idea that the more different ways you can access and use an item, the greater (or "deeper") your understanding (or "transfer").

EXTERNAL ACTIVITIES LIKELY TO PROMOTE ...

Whether you go for depth of processing or variety of types of processing, both the above are kinds of hidden mental interaction between a learner and knowledge. There is reason to believe that they predict learning directly. They are the classifications of cognitive interaction that we want, but do not correspond to what is easily observable (the sensori-motor interactions). However there may be an intermediate, bridging class of interaction: types of interaction that demonstrably promote learning, at least statistically speaking. For instance activities like reading a text, writing an essay, doing an exercise might be instances in such a classification. Eric Smith's and Andrew Fluck's very different lists might also both be seen as attempts at such intermediate classifications. For me they are variations on the notion of mathemagenic activity.

MATHEMAGENIC ACTIVITIES

"Mathemagenic activity" is a term coined by Rothkopf (1970) and means an activity that gives birth to learning. In his words "You can lead a horse to water but only the water that gets into his stomach is what he drinks". In fact a teacher is in an even worse position: not only can a teacher not cause learning directly, they cannot even perceive it happening directly unlike horse minders who can at least see and hear whether a horse drinks and how much. This learner autonomy and the indirectness of teacher power is of course an aspect of constructivism. It is linked here with the idea that learners' actions have a big effect on learning. The impulse behind categorisations of types of CAL, types of learner-computer interaction, or indeed types of educational intervention is that it would be a big help to teachers if types of learner activity could be associated with types of learning. The notion of mathemagenic activity and attempts to classify their types is the essence of this approach to understanding teaching and learning. Laurillard (1993 p.103) takes up this notion and proposes a model in which there are 12 mathemagenic activities. These apply to all subject areas. The basic approach is that no software to date covers all 12 activities, so any complete approach will combine software (multimedia, etc.) with other activities. This model is not without its problems: to explain how learning occurs without overt support for and observation of some activities, I have to hypothesise that these may be internalised. This leads to further predictions that have not yet been properly tested e.g. that learners will report these internalised activities when interviewed, and that study skill training would equip learners to perform such activities internally without further support. Nevertheless, I find this model to have real predictive power. That is, rather than studying the effects of media or of interactive modes on learning, I vote for studying these mathemagenic activities as primary factors determining learning.

AN EXAMPLE: SIMULATION

As an illustration, consider whether simulations are good for learning. Inspired by how revealing a simulation is to them, some teachers implement one and hand it to students by itself. No learning occurs for almost all students (the blank screen phenomenon). So then the teacher provides a worksheet, as for many labs. Students go through the worksheet, but learn rather little, again as in labs. However the situation is actually worse, because in labs at least students will be learning what materials look like and how to handle apparatus, which is one set of objectives (corresponding to activities 6-9 in the model), whereas in the simulation they will be learning how to operate the software not the real thing. So the third step is to go for the best (but still not very common) practice for labs, and put on "pre-labs" that get students to activate the theoretical concepts relevant to the simulation or lab in advance, so as to maximise the chance of their making connections between concepts and practice (activities 10, 11). It would not be technically difficult to deliver worksheets and pre-labs in software as well, but the point here is that this model suggests that learning outcomes depend on all these aspects being delivered, whatever the delivery medium of each. It also, unlike Clark's delivery truck metaphor, says what needs to be delivered and allows us to analyse a situation to identify the missing elements; and suggests that these elements are not "content" nor "media" nor "interactive modes" but activities.

NICHE BASED SUCCESS

I am also finding that this model illuminates why some bits of courseware are merely OK while others are big successes. For instance CAL projects that are basically conceived of as whole courses, when tested, generally generate results that Clark would expect: about as good as non-CAL delivery, plus sometimes a bit of positive halo effect. In contrast the big successes seem to be when the designer identified a defect in the previous non-CAL delivery corresponding to one of the Laurillard activities, and then thinks of a way of exploiting technology to fill the gap. One example (McAteer et al., 1996) here at Glasgow concerned teaching Portuguese. Everyone nowadays thinks that the way to learn a second language is through conversation practice, but it is too expensive to provide each student with many hours of a Portuguese speaker. So a piece of courseware was built to provide a conversational context (a photo of a South American street market, then some canned utterances) and then invite the student to utter a reply (activity 7) which is recorded, and so on. This has produced a dramatic improvement in student learning. In a way it seems obvious: but most courseware actually does not have a clear pedagogic analysis of a need that it is designed to fill, and does not produce big improvements in learning. Success in CAL seems to me to depend upon the specific niche it addresses: clever fits score big, the rest produce the same patchy performance that the non-technological solution did. The Laurillard model helps to analyse what the needs are. Note too that the use of technology here seems compatible with both Clark and Kozma's views. It doesn't do anything that couldn't be done better by humans, but in practice in this particular situation, the learners get many hours practice per week, instead of just 2 in a conversation class. You can see how one set of people might try to analyse this as due to the multimedia properties of the technology, while another might analyse it as due to a motivational effect (after all, why didn't these students just practice on each other or with simple audio tape materials?), and a third as a social effect due to the successful scaffolding by the courseware not achievable by other students (insufficient language proficiency) or simple tape materials (not a convincing conversation). What I like about the Laurillard model is that it lets me spot the missing or weakly supported activities in any particular situation while remaining neutral about whether to redress them by technology. Success then follows when a designer sees how to use technology to cover a gap that human delivery is failing in practice to cover. But I don't see how classifications of interaction types have helped design successful CAL so far.

REFERENCES

Clark, R.E. (1983). "Reconsidering research on learning from media" Review of Educational Research, vol.53 no.4 pp.445-459.

Craik, F.I.M. & Lockhart, R.S. (1972) "Levels of processing: a framework for memory research" Journal of Verbal learning and verbal behavior vol.11 pp.671-684

Kozma, R.B. (1991). "Learning with media" Review of Educational Research, vol.61, no.2, pp.179-211.

Laurillard, D. (1993) Rethinking university teaching: A framework for the effective use of educational technology (Routledge: London).

McAteer,E., harland,M. & Sclater,N. (1996) "De Tudo Um Pouco Ñ a little bit of everything" Journal of Active Learning vol.3

Marton,F., D.Hounsell & N.Entwistle (1984) (eds.) The experience of learning (Edinburgh: Scottish academic press)

Rothkopf,E.Z. (1970) "The concept of mathemagenic activities" Review of educ. research vol.40 pp.325-336