[conclusion of Wresch article at top of this page]

Many of the qualitative issues facing computer-assisted instruction designers in English composition may be resolved through the ongoing research in the field of artificial intelligence. Like rhetorical invention to an extent, artificial intelligence is the study of intelligent, human strategies for solving problems in order to design and develop computer systems that simulate intelligent behavior. As I was programming the three open-ended invention programs named TOPOI , BURKE, and TAGI, I attempted to match expected keyword strings from a writer's natural language in order to keep the writer in a simulated dialogue. These software routines were my introduction to this field of artificial intelligence. Now at the Air Force Human Resources Laboratory at Lowry Air Force Base in Denver, Colorado 80230, I am continuing this investigation by directing programs in research and development in applied artificial intelligence. Two research areas in particular, it seems to me, will be crucial to future research in instructional computing in English composition; they are natural language processing and intelligent computer-assisted instruction.

Natural language procesing refers to computer routines that attempt to "understand" ordinary English (or French or German) sentences and to "produce" English sentences. While many syntactic routines do an adequate job of parsing, the semantic routines in such AI languages as LISP need more development. Moreover, pragmatic routines--software that accounts for the context of the natural language are most problematic. The application of natural language processing in composition software could occur in word processing software and in CAI that demands "user-friendliness." The state of the art of computational linguistics is relatively young; nevertheless, computer-assisted instruction and recursive word processing software ought to be developed in an articulate, simulated natural tongue.

Designers of intelligent CAI attempt to imitate intelligent teachers. Consequently, intelligent CAI in composition would combine knowledge representations of the subject, heuristics for manipulating or arranging the knowledge effectively, and individualized motivational instructional strategies for writers. The pseudo-Socratic dialogues of my invention programs had no sense of a writer's subject matter, a fair representation of three well-known rhetorical heuristics, and a friendly--though often random--set of motivational responses. Making those programs more intelligent could mean improving any one of these three areas, but each improvement has both good and bad consequences the humanistic composition teacher should consider. For example, integrating specific knowledge helps a writer learn or recover the content issues, but knowledge occupies more of the computer's memory and may make the program operate slower. Having the program judge the order of the heuristic questioning based on the previous reply, the topic, or the probability of a successful answer from the student rather than a random order or a predetermined sequence certainly would improve an invention program, but, again, such intelligent improvements require memory and time to execute. Finally, one major goal of intelligent CAI is that such a system would "learn" or at least acquire the experience of its own runs. In other words, the software would be constructed to augment the knowledge base, examine the order of the heuristic presentation and modify the order if necessary, and build diagnostic routines or assimilate self-motivational cues from the writers themselves. Obviously, a good deal of research is needed in ICAI, but I feel

composition CAI offers an ideal research arena for such self-improving software because writing must mix knowledge products with linguistic processes.

I'll close this note with an observation made by Douglas Hofstadter in Godel, Escher, Bach: An Eternal Golden Braid (New York Basic Books, 1979):

The idea that, if AI is to be achieved, the actual hardware of the brain might one day have to be simulated or duplicated is, for the present at least, quite an abhorrent thought to many AI workers. Still one wonders, "How finely will we need to copy the brain to achieve AI?" The real answer is probably that it all depends on how many of the features of human consciousness you want to simulate. (572)

I for one, believe composition teachers can use the emerging research in artificial intelligence to define the best features of a writer's consciousness and to design quality computer-assisted instruction--and other writing instruction--accordingly.