Contextual relevance in analogical reasoning: A model of legal argument.

摘要

Analogical reasoning is used in rational argument and problem-solving. It is used by knowledge workers in organizations for many kinds of problem-solving tasks. Providing information systems to support the information requirements of knowledge workers requires an understanding of action and representation knowledge required in such tasks. Analogical reasoning is a knowledge-based, problem-solving process in which intelligent agents apply knowledge from precedents to problems. The justification for use of a precedent is relevant similarity between a problem and a precedent.; The contextual relevance theory of analogical reasoning explains the process by which relevance judgment varies with a change in problem solving context. Relevance judgment is the selection of a precedent as relevant to a problem case. The theory explains a method for measuring similarity that is computationally practicable in information processors with bounded resources.; The contextual relevance theory is a knowledge level theory of analogical reasoning. The unit of analysis is knowledge possessed by an agent as opposed to a particular symbolic representation of the knowledge. In the contextual relevance theory conceptual knowledge is the knowledge that permits effective similarity judgment and that affords shifts in relevance judgment with changes in context. Conceptual knowledge includes heuristic schemata for recognizing possible instances of a concept and includes constraints on instances of a concept. The similarity of a problem to a precedent is measured with respect to a concept. Situations that contain objects that are instances of the same concept are similar with respect to that concept. A precedent is useful when it provides knowledge that helps to meet a goal in a problem. Relevance is based on conceptual similarity and usefulness.; BRAMBLE is a model of the contextual relevance theory of analogical reasoning. It is also a design specification for a computer program implementing the theory. BRAMBLE is realized as a logic program written in MRS. The knowledge-base in BRAMBLE is derived from data collected in a study of legal reasoning about corporate acquisitions. The program demonstrates and tests the contextual relevance theory, showing the power of conceptual similarity to respond to changes in context and to generate contextually plausible analogical arguments.