1986 年 1 巻 2 号 p. 228-237
This paper describes mechanizing abduction which means hypothesis generati,on and selection for given observations. We consider hypothesis generation and selection problem in a similar framework to a theory formation system Theorist. The experimental system is implemented in DEC-10 Prolog. It generates tentative hypotheses from a set of hypothetical formulas and selects one to form consistent explanations of given observations. The differences between our system and Theorist are follows : (1) In our system,hypotheses and given knowledge are represented in an is_a hierarchical structure and appropriate hypotheses for given observations are derived based on the structure. On the other hand,in Theorist,hypotheses are not supposed to have such a structure. (2) If multiple possible hypotheses are to be given by the system,our system asks new observations to the user in order to select one of such possible hypotheses. The important thing is that meta-programming methodology enables the user easily to define the criteria for hypothesis generation and selection mechanism. (3) Our system can make use of negative knowledge (constraints) in the same way of Theorist-S. Our system has a clearer semantics and an easier implementation way than Theorist-S.
