Formalizing nonmonotonic reasoning systems

摘要

In recent years, there has been considerable interest in nonmonotonic reasoning systems. Unfortunately, formal rigor has not always kept pace with the enthusiastic propagation of new systems. The argument has long been made that, because of the general intractability of formal systems, it is unreasonable to consider them for practical applications. This is taken as support for the use of systems such as semantic networks which, although not completely understood, can compute quickly. We suggest that this argument is not entirely convincing, and that formalizing such systems may yield dividends in terms of both clarity and correctness. We argue that formal systems, such as Reiter's default logic, provide useful tools for the specification and description of nonmonotonic systems. We present new results which enhance this usefulness. To illustrate the benefits of this approach, a theory of inheritance networks is developed. This yields a notion of correct inference, and sufficient conditions for the coherence of network inference representations.