In an attempt to overcome the lack of natural means of communication between student and computer, this thesis addresses the problem of developing a system which can understand natural language within an educational problem-solving environment. The nature of the environment imposes efficiency, habitability, self-teachability, and awareness of ambiguity upon such a system. The major leverage points that allow these requirements to be met are limited domain, limited activities within that domain, and known conceptualization of the domain. Semantic grammar is introduced as a paradigm for organizing the knowledge required to understand and permit efficient phrasing. The need for succinct formalism for expressing semantic grammars led to the use of the Augmented Transition Networks (ATN). This led to the design and implementation of a general ATN compiling system which in turn translates an ATN into a program in a runnable computer language (LISP). The ATN compiler is also capable of producing programs which have been optimized to the features used by a particular ATN. The ability of ATN-based semantic grammars to perform satisfactorily in an educational environment is demonstrated in the natural language front-end for the SOPHIE system. (Author/WBC)