On some supposed contributions of artificial intelligence to the scientific study of language

Current work in Artificial Intelligence (AI) does not deal with the essential questions that should be attacked by an inquiry into the workings of lang. The principles on which such work is currently based provide no basis for psychological reality claims. Current work does not provide valuable contributions because of its incorrect theoretical foundations. The basic goals of the scientific study of language are outlined in the first section. A valid theory must be observationally, explanatorily, & descriptively adequate. It must explain the principles of human language organization & account for the facts of language acquisition. The stated goals of a number of prominent researchers in AI are outlined in a second section. One major goal is the "practical desire to have a usable language understanding system." Universal criteria need not enter into such a pragmatic schema at all. Present language understanding systems are mislabelled in that they actually only 'understand' language in some very limited context. Other stated goals of "gaining a better understanding of what language is" & developing new theories must necessarily fail to offer valid psychologically-oriented theories, because they stem from the first criterion. The recent work of several prominent researchers is critiqued in some detail in the third section. T. Winograd has proposed a model based on Systemic grammar which has a level of syntactic structure & a level of surface structure containing feature & function markers, connected to the syntactic level by realization rules. In both detail & overall organization, this proposal avoids or arbitrarily decides questions essential to a true understanding of lang. However the very features causing this theory to be inappropriate for linguistic purposes render it very useful for carrying out limited language tasks. M. Minsky's 'frame' theory is built on frames consisting of a top, unchanging portion, & a lower portion which may be altered. Possible nodes, possible frame configurations, & what items may be changed within a frame are never defined: "There is no theory of anything at all, and especially not a theory of human thinking." R. C. Schank has claimed to transcend the computer program format, with a set of conceptual dependency networks he calls C-diagrams. It is shown that any given diagram is only one of an infinite number of possibilities & the model thus explains nothing. The work of E. Wanner et al, on Augmented Transition Networks (AT