ABSTRACT
Identification and representation of knowledge is a major problem in the experimental and theoretical studies associated with artificial intelligence research. The spectrum of tasks studied in artificial intelligence range from puzzle solving at one end to speech and visual perception tasks at the other. The sum total of accumulated knowledge used in puzzle solving is usually miniscule, while each of us spends a significant part of our life acquiring knowledge necessary to perceive and understand speech and visual stimuli. Unlike problem solving, the perceptual tasks are characterized by high data rates, large amounts of data, (possibly) errorful input, and the need for real-time response. These characteristics make it imperative that we use every available source of knowledge in the design of perception systems. In this paper we report on a system that attempts to understand speech, as an illustration of the role of knowledge in an intelligent system.
