Aspekte der rechnergestützten Generierung von Inferenzregeln durch Regelschemata

Abstract

One of the main problems in designing Natural-Language-Systems depends on the question “Where do all the inference rules come from?”. In this paper we describe a subsystem of the NL-QAS BACON supporting the task of writing inference rules. The domain inference rules are seen as transformations from logical expressions to logical expressions. The inferential process is controlled by the operators in the domain of the expression to be transformed. For any operator exists a set of relevant rules. A relevant class of inferential rules depends on “semantic relations between words”. A set of basic concepts of such relations is introduced. Each of these concepts is paird to a rule-schema by which it is possible to generate special inference rules for the operators of the semantic representation language. The inferential component and the rule-generating process (checking some types of inconsistencies) are described. The problem is discussed from different viewpoints: the system designer’s developing an AI-system and the cognitive scientist’s interested in the basic cognitive concepts of humans.