The focus of the investigation described in this paper is the development of a concise, yet rich knowledge representation paradigm that could be effectively and efficiently used to model the intelligent behavior of simulated agents in a simulator-based tactical trainer. The behavior of these agents would be similar to that of an adversary who would react to a student's action in a manner representative of enemy tactics. The availability of this feature would be of significant utility to the training process for two reasons: 1) the student would face a realistic enemy who is knowledgeable about tactics in the domain of interest and, 2) the instructor would not have to be burdened with playing the part of the enemy in those training systems where this is commonly done.

The hypothesis presented is that whereas tactical knowledge is highly dependent upon the context (i.e., the situation being faced), a combination of script-like structures and pattern-matching rules in an object-oriented environment could serve as a concise means of representing the knowledge involved, as well as an efficient means of reasoning with that knowledge. This hypothesis was tested through the development of a prototype system that implemented the knowledge of a submarine tactical officer on a patrol mission. The prototype was implemented in CLIPS 5.1, a rule and object-based expert system shell developed by NASA. The results of the prototype show that the combination of scripts and rules in an object-oriented environment promises to meet the requirements described above.