In description logics, default knowledge is exclusively treated as incidental rules. However, as few concepts are definable using only strict knowledge, imposing strict definitions leads to terminological knowledge bases that mostly contain partially defined concepts. This is a real problem because such concepts can only be inserted as leaves of the terminology. Moreover, instance recognition is biased as these concepts must be explicitly mentioned as properties of these instances. It follows that partially defined concepts are described with necessary but not sufficient conditions. As a solution to these problems, we propose to integrate defaults in concept definitions and we argue that this is essential for our diagnosis application. We introduce a description language ALεδ with default(δ) and exception(ε) connectives. The cornerstone of our approach is the introduction of a definitional point of view where a default can be part of a concept definition, whereas in the classical inheritance one it is only viewed as a weak implication. We go on to describe a map between the definition of a concept and its inherited properties, and we show that the combination of these definitional and inheritance levels considerably improves the capabilities of classification processes. In particular this allows us to distinguish sure from probable instances and typical from exceptional instances. Finally we provide a specific operation, object refinement, which consists in enlarging object descriptions with exceptions in order to find additional concepts the object is an instance of. This operation is useful for our diagnosis application.

We propose a treatment of coordination based on the concepts of functor, argument and subcategorization. Its formalization comprises two parts which are conceptually independent. On one hand, we have extended the feature structure unification to disjunctive and set values in order to check the compatibility and the satisfiability of subcategorization requirements by structured complements. On the other hand, we have considered the conjunction et (and) as the head of the coordinate structure, so that coordinate structures stem simply from the subcategorization specifications of et and the general schemata of a head saturation. Both parts have been encoded within HPSG using the same resource that is the subcategorization and its principle which we have just extended.

A semantic network is a structure for representing knowledge as a pattern of interconnected nodes and edges. This paper focuses on the means linear logic offers to represent these networks. In order to compare our inferences, we have chosen one non-monotonic logic: default logic serves as a reference. The main result proves the equivalence between linear logic and default logic in taxonomic default theories. We hope this will help better to understand the relations between non-monotonicity and defeasible knowledge representation.