Computers and Society
David D. Thornburg, Contributing Editor
Computers, Writing, And AI
Of all the myriad applications for personal computers, I would guess that, by far, the most popular use is word processing. In the early 1980s, the availability of good word processing programs on micros virtually drove the dedicated word processors from the marketplace. Even though there has been a resurgence of dedicated word processors, there is little question that the personal computer will continue to be king in this application area.
As a writer who feels naked without his computer, I occasionally startle people when I say that I don't use a spelling checker. I thrive on the computer's ability to let me move text around to my heart's content; I enjoy the ability to format my output in any fashion I want, but I don't use the computer to check my spelling or usage.
Casting A Spell
My reluctance to use spelling checkers has nothing to do with my spelling skills; I make plenty of mistakes. My problem is that spelling checkers don't catch the kinds of errors I commonly make. Many (but not all) of my spelling errors result in other words. For example, stop becomes spot. A word-based spelling checker is useless in such cases. I would love to have a spelling checker that catches errors like these, but such programs have to be context sensitive.
Another problem with most spelling checkers is that they can lead you to wrong choices. For example, with the word alot, most spelling checkers would offer the choice allot instead of the correct split into two words, a lot. The word payed would be even harder to handle since spelling checkers that scan letter by letter would end up with payer, and so on, and not offer the correct past-tense form, paid.
The complexities of English are such that a good proofreading tool needs to be based on something far more sophisticated than a word list. It needs to have access to rules of grammar—an ability to deal with phonology, orthography, morphology, syntax, and semantics.
This is a tremendously complex task.
Because of the rule-based nature of much of English grammar, it is appropriate to seek some solutions for this problem in the domain of expert systems—sophisticated programs that are an outgrowth of years of research in the field of artificial intelligence.
A Model Emerges
A few months ago, I was privileged to visit with Dr. Robert Arn, the creator of just such a program for the ICON computer.
The ICON is manufactured by Unisys to meet the needs of educational computing in Canada. Dr. Arn designed the ICON, and he holds a doctorate in the field of linguistics. His interests in technology and linguistics came together in a project that has kept him and his colleagues busy for the past three years. During this time, he designed Englishl—a rule-based program that sits in the background of a word processor, ready to offer advice when asked.
To use Englishl, the user simply selects the text to be analyzed and requests that it be examined. Words that seem incorrect are highlighted. The user can then select any highlighted word and ask for an explanation of what may be wrong. The computer looks for spelling and usage errors and provides as much coaching and advice as it can. The reason for providing extensive coaching (as opposed to simply correcting the error) is that this program is designed to serve as a writing tutor. The goal is to help the writer avoid errors in the future, not just to correct them.
The program version I used worked with a base of rules for all but the semantic level of language. Dr. Arn is working in that area now. Using our previous example, it correctly identified that alot should be a lot and that payed should be paid. This latter error was detected by first catching the misspelling, noticing that the -ed ending may imply a past-tense form, and then exploring rules for forming past-tense forms.
In addition to providing an intelligent spelling checker, Englishl also examines the text for words that present usage problems for some writers, for example, fewer/less, seasonal/seasonable, and so on. Consequently, this program can serve as a writing coach to a student who is working on an essay. English1 does not make any changes in the author's text; it makes only observations and suggestions.
English 1 Is RAM Tough
In the form I saw, English1 takes five megabytes of disk space. The new version that includes the ability to deal with semantic issues will be distributed on CD ROM since it is going to use about 500 megabytes. These sizes indicate just how complex natural languages are.
Even though Unisys is starting to sell the ICON in the U.S., it is targeted to classrooms, not businesses. Consequently, it is unlikely that we will have access to tools as sophisticated as English1 for some time. On the other hand, the fact that English1 exists at all proves that good proofreading tools can be created to run on microcomputers. Continued work in this area will result in programs that run on the machines we have on our desks.
Until then, I'll have to be content to do my grammatical analysis with the one computer that outstrips them all—the human brain.