Automatic Problem Description from Model Program for Knowledge-based Programming Tutor

Abstract

Knowledge-based programming tutors are supposed to analyze the students programs using knowledge of the concepts of programming language, skills to build programs, misconceptions of novice programmers, and information about the programs to be analyzed. The last one provides the programming tutor with the intentions of programmers, and this allows the tutor to do an intention-based diagnosis (Johnson, 1996). This is given to the system in the form of a problem description by human instructors. However, it is very hard for instructors to write a problem description. For instructors, the simplest way to describe a problem may be to write a model program of that problem. This paper describes the system named GOES, a GOal Extraction System, which extracts the purposes (Goals) of a model program automatically. GOES is a susbsystem of C-Tutor, a knowledge-based C Programing Tutor(Hahn et al., 1996). GOES extracts implemented plans from a model program, and then extract goals from the plans according to goal/plan hierarchies. The efficiency of GOES depends on the process of plan extraction. In GOES, only pairs of labels and variables of each statement are used to extract candidate plans. Exact matching is followed only for theses candidate plans. In addition, to extract plans more efficiently, we introduce the concept of necessary and sufficient conditions of a plan to the model program. Using this method, plan candidates can be extracted efficiently and successfully.