David N. Dobrin
Over the last year, I have gained a certain notoriety by proclaiming my disapproval of text analysis programs like AT&T's WRITER'S WORKBENCH (WW). My disapproval is based on an in-principle claim about the capabilities of such programs. They cannot accurately simulate the kinds of analysis we do (the claim), and therefore they are pretty much useless for the purposes we have in mind for them (the disapproving consequence of the claim). Rather to my surprise, most of the many responses to my work have admitted the claim, but denied the consequences. In this short article, I would like to describe the general form of these responses and suggest why they may be mistaken.
Let me begin by reviewing the relevant part of my argument. Whenever an analysis program is known to be inaccurate, its output is untrustworthy. Before a user can act, therefore, on any explicit or implicit recommendation made
by the program, the user must evaluate the output. If, for instance, a spelling checker says that the word "metamorphoses" is misspelled, the user must decide whether it is in fact misspelled, or merely unrecognized, before he or she changes it.
So much may be obvious, but let me remind you that "must" is a big word. Neither the WW or spelling checkers come provided with a whip. If the user is feeling lazy that day, he or she can take the computer's word for it, and no one will be the wiser. Nor is there any guarantee that the evaluation will be correct. The words the fact that in the sentence "Sherlock Holmes discovered the fact that had eluded everyone else" are used correctly, but the WRITER'S WORKBENCH will identify the usage as a potential mistake. If the not-so-expert author of that sentence happens to believe the computer, no one will ever know. Correct evaluation is not all that easy; usually it requires some sophistication. Most of us, for instance, need a peek at the dictionary to tell about "metamorphoses." Where there is less agreement about what an error is and no final authority like the dictionary, evaluation is even more delicate. We may have to know a lot about grammar and English and even about the habits of computer programmers who try to be English teachers before we can tell why a recommendation was made and whether it is correct. Indeed, so much sophistication is required that usually people who can evaluate the output correctly are the people who don't need the programs in the first place.
Admittedly, there may be people who don't know the rules perfectly, people whose memories
might be jogged by the computer program but who will not be misled by it. In a recent article (l), I argued that these people were few. Whether they would be helped, moreover, depended crucially on the kind of analysis being done. Where analysis was 99% accurate and the rules tested for were strict, as in spelling checkers, there might be many people for whom an analysis does more harm than good. Where accuracy drops and controversy rises, there are far fewer. Nevertheless, even with spelling checkers, some harm is done. Most users of spelling checkers tend to excise rather than correct words they can't spell, and all users of spelling checkers proofread less carefully because they depend on the spelling checker.
In this world, moreover, people are going to use such programs even though they don't have the sophistication to use the packages properly. This not only has the consequences for text or style as mentioned above but it also has consequences for individuals' views of language. At best, these programs encourage various false, silly, or limited ideas about language and give people new opportunities for petty tyranny as they try to impose those ideas on others. At worst, the programs give users exactly the wrong idea about language, that writing well is just satisfying some syntactic rules. (How fortunate it is that style analyzers can tell us when those rules are broken.)
Text-analysis programs, I conclude, might be helpful to a few people in a few situations, but the dangers they pose to the uninformed, combined with the questionable benefits they provide for those who can use them, make them a doubtful sort of tool, rather like those exotic
gadgets one gets at Brookstone. This is not a terribly strong claim; the only people who ought to be made unhappy by it are people who have wild ideas about computers revolutionizing writing. Frankly, I was originally a little embarrassed about saying something so obvious.
The trouble is that people really do have those wild ideas. We simply have no consensus about what the machines can do and how they should be used. Until I became notorious, I had no idea how disparate perfectly serious opinions held by apparently rational people are. Before I go on, let me list a few, so you can see why I am still hammering on this point. In the past six months, I have been told the following:
I must confess that neither the chemistry professor nor the philosophy professor saw the light, but in a way this confirmed my argument. Evidently they have some mistaken ideas about language, and the programs encourage them in these misconceptions. I do not know about their propensity to petty tyranny, but I have my suspicions.
Of the people who take my argument thus far seriously--many, thank God--most have argued along the following lines. Admittedly, they say, the programs do some harm. But, in the right hands--presumably those of knowledgeable teachers--these programs can be more of a help than a hindrance. True, they often give wrong answers, but they also often give right answers. Where a teacher is around to show the wary user the pitfalls, he or she can emerge from the bush, correct answers in hand, almost unscathed.
The thing I like the least about this reply is the phrase, "In the right hands." There is something unaccountably naive about anyone in the 20th Century using an argument that begins that way. And even if the general form were not naive, it takes an extraordinary will to ignorance to believe that the right hands are ever going to come very near these programs. For the very large companies that sell them, the academic market is a mere backwater. AT&T currently makes the WW a cornerstone of the marketing structure for the 3B2 (an inferior minicomputer). IBM has spent years developing EPISTLE (a better and much more expensive WW); they could never recoup their research money by selling it only to a few English teachers.
Nevertheless, the "in the right hands"
argument has a certain plausibility, and I'd like to go into it a little more. The general idea is that we use the programs as one more tool in the classroom. We teach people about a program, describe what it does, describe its limitations, and then have the students use the programs for the rest of the semester. The expectation is that students' papers will improve in the areas that the programs cover. This might well happen. Using the computer is certainly one way of teaching things that very well might need to be taught, that sentences should vary in length, that "shouldn't ought to have" is occasionally an execrable usage.
Most teachers who have actually tried it claim another benefit. Using the programs gives you something to talk about. Because the program does make mistakes, you can spend time analyzing the computer's analysis, pointing out errors it makes, arguing the justice of its "correct" criticisms, and so on. Some teachers like to take a piece of Hemingway's prose or the Gettysburg Address and run it through the WW. The results, I am sure, are amusing, as well as instructive. Notice that in the last activity, the teacher and student are in league against the authority of the program. No longer is the teacher the enforcer of (apparently arbitrary) rules, the computer is. The teacher is off the hook. And, many teachers report, not only are the rules no longer "their" rules, but the students are more likely to believe in the rules, however badly applied. The rules gain authority when they are embedded in a computer program rather than in the teacher's pronouncements or the Harbrace College Handbook.
Finally, the teachers claim, the students are more likely to go back and look at their text when they have to have the computer analyze it. The program encourages the students to take a critical attitude toward their own work. Because the students have text editors as well as the WW, they can easily change things, and, miracle of miracles, they often do.
So much for the benefits. One way of evaluating such advantages is to ask whether similar or greater benefits can be gained at lower cost. It does seem probable that they can because none of these claimed benefits depend at all on the specific characteristics of the program. Imagine that you had a program that flagged sentences randomly, rather than whenever it found, e.g., a diction error. The program would still encourage the student to look at a sentence again, to discover what was wrong with it (if anything), to discuss it with the teacher (on a equal level) and to change it, if necessary.(2) If you were to use this new program, let's call it RANDOM, you might have to lie a little. You could say, for instance, that this new program did a complete evaluation of the paper, but was inaccurate 40% of the time. This figure is probably roughly right (there's surely something wrong with 60% of my sentences anyway), but in any case, it's just what is claimed for EPISTLE, which is much more expensive.
As long as we're on the subject of getting a similar effect at less cost, why don't we ask students to read other students' papers and flag apparent errors? This again is much cheaper than the WW. It would also produce far more comprehensive error-checking and far more accurate. Remember, the WW only checks a very few
things, and its error rate is at best 80%. Why don't we? First of all because the WW can be counted on to produce some output, whereas students can never be, and second, because the WW has far more authority than the poor students do. If students (or even the teacher) do the error-checking, frequently they will not be believed, whereas for some odd reason, in 1986, Lorinda Cherry, a programmer and documentation specialist at AT&T (who developed the WW), will be.
For me, anyway, there would be an important third reason for avoiding such a program. There is something distasteful about having students waste their time doing that kind of error-checking. Having a computer analyze a paper for errors and then analyzing that analysis is a form of game you can play with the computer: see, it got that one; blew it there. It's not appropriate to play the same game with people, as well as much less fun.
If you did have students read each other's papers, they ought to be spending their time on more important things, like deciding whether anything sensible is being said and whether it can be said better. But of course the same is true for teachers and for the original authors. My real objection, you see, is not the fact that the WW does nothing for you that can't be done better for less. Rather, it is that the WW is a distraction. In a writing class the focus ought to be on writing, not correctness in the few areas that happen to be subject to relatively accurate (80%) grammatical analysis. True, after hours spent rather enjoyably learning the program and learning its limitations, students under the guidance of the right hands will
probably not use "The fact that" when they shouldn't ought to have. But every thinking teacher should ask himself or herself: is it worth the time? Does learning that really tell a budding writer whether the thoughts make sense or are worth saying, and whether they are expressed well? And that's what writers need to learn in writing class (and everywhere else).
Teaching these concepts to writers takes far more than the three hours a week we have for them and far more generosity than any but the best of us have, for, alas, adolescents rarely have much to say and rarely say it well. It is painful to have to respond like a human being each week to 30, 60, or 150 three-page recapitulations of the obvious, laced with drivel and ignorance in equal proportions. Yet if, as they often do, these pages are gradually going to become informed rather than ignorant, it will be because someone like you insisted. And you can't be spending your time insisting if instead you are discovering that the WRITER'S WORKBENCH thinks that the first sentence of the Gettysburg Address could be shortened.
Now obviously many things need to be taught at once, and at least some of those are grammatical techniques. But grammatical technique must be taught, like all technique, as being in the service of something. Errors are errors, after all, because they impede communication. If a computer could teach (or even find) such errors, then I can see that it might aid the overall cause, as long as it wasn't terribly expensive and didn't take much time to use. But even then, as long as the range of potential errors found is limited, the program could end up pushing your attention away from the important
problems and toward the problems that the computer happens to be able to identify.
I have sat on several panels where people have proclaimed the success of the WRITER'S WORKBENCH. In my original article, I ran parts of the WW on the article itself, so I could show the reader how much garbage the program generates. When panelists show samples of the program's successes, however, they usually remove the garbage. What is remarkable about these "good" examples is that very often even they are a distraction. The teachers show a paragraph with errors in three of the five sentences, which the WRITER'S WORKBENCH has correctly found. They don't point out that the paragraph itself is devoid of common sense, repetitive, or merely infelicitous, and that the best way of correcting it is to combine sentence one and three. They seem actually to have missed that simple fact. The WW has distracted them.
To be a help in the classroom, I want to argue, the WW must not only find errors (or not find them), but it must find the important errors, the ones that prevent people from saying what they want to say or what they might have said. If we want to teach people to be skilled evaluators of their own and others' work, the presumed goal of the people who make the "in the right hands" argument, then we have to teach them the most important skill, namely deciding where the problem lies in a paragraph. If the problem happens to be one that the WW identifies, well and good. If, however, the fault the WW identifies does not happen to be the problem, then the WW is a distraction. I
leave it to you to figure out which is more likely to occur.
I know that some teachers find it pleasant to have the WW and programs like them. But I hope that in the course of time, they will discover that like Valium, the WW is both expensive and rather bad for you.
1. College English (forthcoming, 1986).
2. This idea came to me when I discovered, after using the WW for a while, that I was frequently correcting the sentences that the WW flagged but rarely making the corrections it suggested.