A knowledge representation language for large knowledge bases and “intelligent” information retrieval systems

摘要

This paper describes the “conceptual” Knowledge Representation Language (KRL) proper to an environment for the construction and use of Large Knowledge Bases (LKBs) and/or Intelligent Information Retrieval Systems (IIRSs). The coding of the elementary events occurring in the real world (“Snoopy is Charlie Brown's beagle”) is organized around a “semantic predicate” (“own,” “move,” etc.) identifying the “basic” type of elementary situation to be described. In the literature, this type of information is labelled by using several names: episodic memory, assertional data, etc. In this paper, we shall speak of “descriptive data”: they are the domain of the “descriptive component” of the KRL. The “entities” (Snoopy, Charlie Brown, beagle…) which are mentioned in the elementary event and which are, at least partly, proper to a particular application domain fill the “roles” (slots, facets, cases) associated with the semantic predicate; the entities are, therefore, the “arguments” of the semantic predicate. Roles include “subject,” “direct object,” “source,” “destination,” “modality,” etc. The conceptual predicative units constructed in this way take the name of “predicative occurrences”; the semantic predicate, the arguments and the conceptual unit as a whole can be characterized by “determiners,” for instance “temporal determiners” which quantify time duration or frequency of the elementary events. Moreover, the predicative occurrences translating the elementary events can be associated under the form of “binding conceptual units” using logical, causal, etc. relationships, giving rise to complex conceptual constructions (“binding occurrences”). The entities of the application domain which are used in the predicative occurrences are defined in terms of a specialization hierarchy (“lexicon” = “A beagle is a sort of hound/a hound is a dog …”). Thus, every entity is represented as the specialization of one or more parent-entities from which it inherits properties (attributes) and behaviors. Analogously, occurrences are defined according to their own specialized hierarchy (“grammar”) by referring them to “abstract conceptual units” (“templates”) which describe their general characteristics. The information inserted into these two hierarchies is normally identified by speaking of semantic memory, terminological data, etc. We will make use here of the term “definitional data”; these data are the domain of the “definitional component” of the KRL. A compromise between an “object-oriented approach” and a “logic-oriented approach” is proposed for implementation purposes.