The `understanding´ of natural language in CAI and analogous mental processes

The use of open-ended natural language interaction is essential in computer-assisted instruction (CAI) aimed at teaching clinical decision-making and in interactive tests that emulate oral examinations. Lessons learned in developing CASIP (computer-assisted socratic instructional program) a CAI authoring tool, about the scope and limitations of computerized natural language interaction are examined. Different approaches were tested, and the most effective of these was found to involve a conditionally branched search of key words from among clusters of contextually synonymous words. This mode of man-machine interaction offers a multistage model for certain aspects of human verbal interaction. Possible ways of testing this model are suggested as well as uses for diagnosing certain cognitive dysfunctions, especially in children