ABSTRACT
Two Kinds of Inferences
Inferences are typically generated by applying some inference schemata to a body of knowledge. In deductive reasoning, a schema like the modus ponens (P, P->Q, | Q) may be applied to a rule set and thereby produce some inference (e.g. Q). Inductive inferences are generated by an inference schema which replaces specific by more general terms. The inference of "fruits are eatable" may thus be induced from the assertion "apples are eatable". In a way, such inferences are already implicitly contained in the knowledge base and the inference schemata. In other words, inferencing is often the explication of implicit information. Creative inferences, on the other hand, are constructed as novel knowledge units which are not even implicitly contained in the terminology of the knowledge base. Such creative inferences may be produced by relating separate terminologies to one another. For example, Boden (1991) has described Kekulé's discovery of the benzene ring as such a creative insight (or inference) where the knowledge about strings of carbon atoms became associated with the knowledge about snakes. A snake which bit its own tail could thus generate the idea of a ring (rather than a string) of carbon atoms. This would be impossible to deduce with the original terminology about carbon atoms. In this paper, we describe a computational model which produces creative inferences.
