Argues that the best strategy for building integrated, personalized information systems consists of linguistically oriented knowledge and inference, rather than artificial intelligence techniques. The discussion addresses difficulties in the evaluation of such systems and strategies for devising performance studies. (five references) (CLB)