Automated derivation of program control structure from natural language program descriptions

Abstract

This paper describes a system which organizes a natural language description of a program into a conventional program control structure, as a part of a larger system for converting informal natural language program specifications into running programs. Analysis of the input program fragments using a model of a human “reader” of specifications has been found to be a very successful adjunct to conventional “planning” methodologies. Natural language descriptions of programs can frequently be characterized as “rubble”—a very loosely organized set of almost independent description fragments [Schwartz]. Such specifications are often quite robust, due to a large degree of redundancy; they are also frequently quite concise, due to reliance on the readers' innate knowledge and their knowledge of the application domain. This paper discusses a paradigm for structuring the portion of “rubble” program descriptions which maps into conventional programming language control constructs and definition facilities.