Susumu Kunifuji

The Japan Advanced Institute of Science and Technology, Hokuriku (JAIST) was founded in 1990 as a national graduate institution for the advancement of the frontiers of science and technology. JAIST offers three graduate programs of information science (from 1992), materials science(from 1993), and knowledge science (from 1998), which lead to master's and doctoral degrees. The schools of Information Science and Materials Science have 17 chairs, and the school of Knowledge Science has 12 chairs. The author is a professor of the Schools of Information Science and Knowledge Science, and a director of the Center for Information Science.

Research currently focuses on knowledge acquisition and learning, groupware, and creative thinking support systems, e.g., abductive and inductive logic programming, legal expert system, group decision support system, brainstorming support system, awareness support system, and convergent thinking support systems. This article describes his researches on knowledge acquisition and learning.

To conquer a knowledge acquisition bottleneck problem in the knowledge-based system, we are interested in developing a knowledge acquisition support system with learning capabilities. We designed and implemented a knowledge base management system KAISER and a hypothetical reasoning system HRS. Based on the experiences of these two systems, we proposed new logic programming languages with extended inference functions, abduction and induction, in 1995-97 and tried to implement a legal expert system using abductive logic programming in 1995-96.

Abductive Logic Programming is an extension of Logic Programming to perform abductive reasoning. The main objective of ALP is to find a plausible explanation which is a set of ground atoms that can explain given observations. An original abductive proof procedure for ALP was proposed by Esghi and Kowalski. We adopted Kakas and Mancarella's procedure that allows not only negative literal but also arbitrary literals as abducibles. Using this, we developed a new legal reasoning system in ALP. The system can deal with ambiguities of the described facts and exceptions which is not described in relevant articles. In addition, the goal queried to our legal reasoning system is different in compliance with the user who is a plaintiff or a defendant. In order to overcome these difficulties, ALP is used in our system which can deal with the implicit exceptions and generate the presumptions according to the user's demand.

Secondly, even if ALP is used as a legal reasoning mechanism that makes up for a lack of legal knowledge, it generates multiple hypotheses set. Then, our system has to select one suitable hypotheses from this generated hypotheses set, with a hypotheses selection module executing fitness calculation from the similar precedents. We implemented an experimental system for the contract law such as CISG (united nations Convention on Contracts for International Sale of Goods), and demonstrated our system at IJCAI97 at Nagoya.

Recently, we are investigating an integration of abduction and induction. One idea is to use an extended inductive generalization with abduction. Our approach can deal with the situations when background knowledge is insufficient to induce definitions of given examples. Most existing Inductive Logic Programming systems implicitly assume that the given background knowledge is enough to induce definitions of given examples. But, generally speaking, background knowledge is not usually supplied sufficiently in real-world situations. In order to solve this difficulty, we propose an extended framework of Inductive Logic Programming which uses the methods in Abductive Logic Programming. For more information, please contact us by e-mail at kuni@jaist.ac.jp or kanai@jaist.ac.jp.

School of Knowledge Science, JAIST

Tatsunokuchi, Ishikawa 923-12, Japan

Email: kuni@jaist.ac.jp