Computers serve as a kind of mirror. They can replicate human abilities, but only if those abilities are capable of being comprehensively described. When we try to do something on a computer which we have previously only done in some other way, we sometimes learn a great deal about ourselves. Using a computer to teach has forced us to think rather carefully about what teaching consists of.
Early attempts to use computers in education concentrated on the teacher's role as the controller of an interaction, making statements, verifying through question and answer whether those statements had been understood, then extending the learners' command of the principles by drills and exercises requiring inference and analogy. This is what Kemmis described as the Instructional Paradigm (Kemmis et al, 1977), and he said of it that it "involves the belief that the knowledge students need to acquire can be specified in language and learned by the transmission and reception of verbal messages". The Instructional Paradigm, as a general procedure in computer-assisted learning, can be said to have failed. (For evidence of this failure, see Higgins 1983.) The failure forces us to ask what went wrong? Did it fail because machines do this job less well than humans, in other words because we have failed to capture part of the process in our computer simulation of teaching? Or did it fail because this kind of teaching is itself ineffective?
I think I favour the first explanation. A good human teacher has two attributes which are extremely difficult to reproduce on a machine. One is enthusiasm. Good teachers love the subject; they want their students to share that love and 'join the club' of people who command those skills. A computer is a machine, completely neutral as to the program it is carrying. However many 'user-friendly' messages are incorporated into the program, no sensible user will believe that the machine is capable of genuine warmth or enthusiasm. The second attribute is sensitivity. Good teachers listen to their students, and notice what they do. They see them, too, and read their feelings about the lesson from their posture or their faces. When a mistake is made, they know if it is due to carelessness, to genuine misunderstanding of a principle, or to a desire to experiment. They will know what to do in each case. A computer has far fewer channels of communication and, today at least, a far less adequate mental model of the nature of the learner. In standard quiz or drill-and-practice programs, the computer is given no model of the student to work with at all. It simply matches responses to a bank of acceptable responses and reports the result mechanically, "Well done, Jimmy!" or "Wrong, try again."
Does this mean that computers are useless in teaching, in particular language teaching? The kind of teaching role I have described—presenting, questioning, drilling, evaluating—is one I have called magister. In this role the machine is manifestly inferior to the human. There is, however, another teaching role in which it may be superior. This second role I call pedagogue. Look the word up in a dictionary and you will find that its original meaning is "the slave who escorts the children to school." In classical times he would walk five paces behind the young master, waiting to be summoned to answer his questions, play word games with him, or even, if that was what the young master wanted, give him a test. The moment the young master had had enough, the pedagogue went back to his place. In more recent times a pedagogue role has been played by, for instance, ballet and fencing masters at aristocratic courts, nannies and private tutors in rich families, and au pair girls and 'assistantes'. The pedagogue has knowledge but no authority, must always be available but can be ignored or sacked if he fails to please. He leads a wretched life, and modern mankind is rightly guilty about the form of exploitation he represents. In any case, he can only serve one learner or a small group at any one time, so it has become too expensive to use human pedagogues on a large scale.
Computers make bad magisters but good pedagogues. We need have no bad conscience about exploiting them. But, in order to use them, we, both as teachers and learners, need to re-learn a skill which we have largely forgotten, the skill of exploiting a slave. The slave, after all, is passive. He takes no initiatives, asks no questions until commanded to, but obeys mindlessly whatever instructions are given to him. In order to use computers as pedagogues learners will have to learn to take initiatives, to dig out knowledge which will not be offered freely. Teachers, perhaps, will have to learn to stand back and not interfere with the exploration process.
What types of activity fit the pedagogue role? The slave can be a copyist, copying out the young master's work whenever he changes his mind about what he has written. This is what a word-processor does. The pedagogue can be an encyclopedia, answering questions but not telling him what questions to ask. This is what happens when the young master consults a database. Both are natural and common applications of computers. Or the pedagogue can play games, and this is where the computer can be of greatest relevance to language learning.
I must first dispel some of the associations of the word 'game'. Games are more respectable than they used to be in language learning, but we still tend to think of them as Friday afternoon activities, recreation rather than serious learning. The games I am going to describe are not of that kind. They are challenges, problems which can engross learners and give them very intensive forms of practice while they remain largely unaware of what language skills they are practising.
The first group of activities include simulations, logic problems and adventure games. The computer presents a problem. Learners, often working in groups, take a decision. The computer now displays the outcome of the decision. Real-time simulations are familiar in the form of arcade games and flight simulators. They are of little relevance to language learning. However, there is another category of simulation, move-based simulations, which are highly relevant. In such activities one has, for instance, to run a company, deciding year by year how many employees to engage and what price to charge. The real language practice may not be what one reads from the screen and types at the keyboard but rather the talk among learners as they discuss the next move. Logic problems are similar; MURDER, for instance, presents a simplified police investigation. The task is straightforward, but the talk it engenders can be very rich. There is a special type of maze exploration game generally known as Adventure. These have been well exploited in language classes, notably by Armando Baltra (1984). In Adventures one has to communicate one's decision to the computer through brief verbal instructions (often two-word sentences) and a large part of the activity consists of trying out different forms of words until one discovers a pattern which the computer will accept and act on. The relevance of this to language learning is obvious.
The second group of activities can be labelled Generative. (The term is due to Tim Johns; see Higgins and Johns, 1983). In these the computer is given some form of grammar and vocabulary and will assemble pieces of language. The learners' task is to study what it produces, respond to it in various ways, decide what is meaningful, and perhaps modify the rules. This may amount to 'teaching the computer'—notice how the conventional relation of learner to magister is reversed here. In Johns's A/AN, for instance, the machine will add 'a' or 'an' appropriately before any noun or noun phrase the learner types in. The challenge is to beat the machine, to find some exception to its rules and force it to make a mistake. (This will be difficult: the program is sensitive to the difference between A UNIFORMED MAN and AN UNINFORMED MAN, for instance.) Advanced learners will become fascinated by the rules themselves, and will want to see 'how the trick is done'. Another classic program in this area is ANIMAL, which I have developed and extended under the name JACKASS. In this the computer challenges the learner to a guessing game, trying to guess what is in the learner's mind. It is almost bound to fail, of course, but after each failure it asks the learner to tell it what he or she was thinking of; it adds the new information to its memory and gradually grows 'cleverer' under the learner's eyes. At any stage the learner can interrupt the process and ask the machine either to write an 'essay' incorporating what it has learned or to reverse the roles and set the learner a guessing task, based on the information the learner has given it.
The third category of activity is labelled 'analytic'. These are activities in which the computer takes pieces of text, analyses them in some fashion, and then sets a task based on its analysis. The analysis can be either intelligent, ie taking account of the meaning of the text by using some kind of semantic parsing, or unintelligent, i.e. simply using space and punctuation to identify word, phrase and sentence boundaries. The latter kind is far easier to implement on a small computer, and may be just as rewarding. For example, a program called CLOSE-UP (now renamed PINPOINT) takes a short piece of text and masks it except for one randomly chosen word. The screen also displays eight potential titles. The learner may be able to guess the title of the text at once, but more probably he or she will need to see more of it. Pressing a key will reveal the words before and after the starting word, and this process can be repeated, opening the window on the text wider until the learner has enough information to guess intelligently. The process is designed to encourage guessing, but the scoring system will penalise wild guessing. Learners gradually evolve a winning strategy as they discover how much or how little information they need in order to identify the topic of the passage.
In its pedagogue role, the machine cannot plan a lesson. It does not know what the young master will want to do or how long he will want to continue an activity. All the machine can do is offer a range of activities. Nor does it have any record‑keeping or evaluating role beyond that demanded by game-scoring. Effusive congratulation or coy criticism are equally inappropriate. It is a magister's function to say "Well done" or "You must try harder". The pedagogue's equivalent is "You win" or "You've lost", and even these are not always needed. Sometimes the only message needed is "You've finished", and that, for a learner who has been pursuing a challenging task, may be congratulation enough.
The benefit to learners from the availability of computers in language learning is, I believe, not just that they can enjoy a wider range of extra-curricular activities, practice tasks which can be fun. It is also that they will be forced to re-assume initiatives which they cannot take in the conventional classroom. A class of thirty does not allow each student to use the human teacher as a pedagogue, to ask questions as they occur, to demand a particular kind of work at a moment's notice. Therefore learners whose only exposure to a language is in the classroom may acquire a habit of passivity, of answering questions but never asking them. Such passivity will be of no service in front of a computer (other than one which is carrying a drill-and-practice exercise or a piece of programmed learning). But the computer is endlessly patient and capable of unlimited repetition. With a problem to solve, the learner can try out a range of inputs, and become an experimenter. The computer can be a laboratory in a sense in which the Language Laboratory never was, namely a place where one can try out a range of different procedures and observe the results. It can put the trial back into trial-and-error.