(14) A Study on Legal Reasoning based on Goal-Dependent Abstraction

This study presents a new algorithm to find an appropriate similarity under which we apply legal rules analogically. Since there may exist a lot of similarities between the premises of rule and a case in inquiry, we have to select an appropriate similarity that is relevant to both the rule and a goal of our reasoning. For this purpose, a new criterion to distinguish the appropriate similarities from another useless ones is proposed and tested. The criterion is based on GDA, a goal-dependent abstraction, to select a similarity such that an abstraction based on the similarity never loses the necessary information to explain the goal. In order to cope with our huge space of similarities, our GDA algorithm uses two criterions that originate from the terminological structure of our legal knowledge. The first one is called a similarity inheritance condition by which similarities between general concepts are calculated from the similarities between the most concrete concepts. The second one is concerned with role-filler relationships between concepts, and is called a value restriction preservingness. By this condition, our similarities are required to preserve the relationships. Some experimental results show that our GDA also the two criterions help us reduce the huge space of possible similarities.