Extraction of problem description from sample program for knowledge-based programming tutoring

Abstract

In building program tutoring system, problem description about program task is inevitable to analyze students`` buggy programs. Problem descripiton is provided an instructor to the system and should include enough information to deal with various students`` programs. However, it is very hard for an instructor to write a problem description in system understandable form. The simplest way for an instructor to describe a problem may be to write a sample program of given problem, which cannot provide enough information to analyze various programs. In this dissertation, implemented is GOES(Goal Extraction System) which analyzes sample programs and generates problem descriptions. When analyzing sample programs, GOES proposes "Label-based Plan Recognition" to recognize programs more efficiently and control the search space which grows explosively as the size of the program and the knowledge base become larger. GOES is a bottom-up program understanding system which analyzes sample programs with no advanced information about the programs, and generates high-level descriptions. There are two issues in bottom-up program understanding. First issue is how to represent programming plans which are the basic semantic blocks in programming understanding. In GOES, plans are represented as code-base form which provides the understandability and extensibility of knowledge base. The drawback of code-based representation is that it has difficulty in coping with syntactic variability. GEOS solves this problem by removing syntactic variations during parsing and preprocessing. Second issue is how to control the explosively growing search space efficiently. GOES uses the "Label" which denotes the functionality of each program statement as a beacon of plan extraction. The label is used to check the necessary condition of a plan to a program line, and the sufficient condition of a plan to the program. Variables of each program statement is used to group the noncontiguous plan fragments....