This paper proposes a system that attempts to
combine inductive learning from examples with prior
knowledge in the form of precepts, and default
reasoning. In particular, the system deals with linear
and tree-structured inheritance, including exceptions
to exceptions. It handles conflicting defaults (such as
the famous Nixon Diamond [17]) through either static
priorities given a priori by an external source, or
dynamic priorities obtained by counting matching
examples in a real-world training set. The system's
execution combines rule-based reasoning and
similarity-based reasoning, as in [20]. If a new
situation can be handled by an existing rule (given a
priori or learned), it follows the rule, and if there is no
matching rule, the new situation follows the most
similar (for some similarity measure discussed later)
existing situation. Learning is effected incrementally
and inductively by retaining those situations that are
not accounted for by the current knowledge base. Of
interest is the fact that new knowledge may cause
generalization of the existing knowledge base.