HUMAN-MACHINE COMMUNICATION BY VOICE

One way to gain an understanding of natural spoken language is to derive a constructive theory of it by building a language engine. The more nearly this artificial language engine simulates human lin-guistic behavior, the more of language it may be said to explain. A constructive theory of language must have two important features. It must capture known linguistic structure and it must express this struc-ture in an elegant and computationally tractable mathematical frame-work. Complete constructive theories of language do not yet exist. However, interesting ones do exist and there is reason to hope that they will improve. This paper describes the state-of-the-art in auto-matic speech synthesis and speech recognition and explains some of the mathematical models on which their underlying theories rest. 1