Weakening conflicting information for iterated revision and knowledge integration

摘要

The ability to handle exceptions, to perform iterated belief revision and to integrate information from multiple sources is essential for a commonsense reasoning agent. These important skills are related in the sense that they all rely on resolving inconsistent information. In this paper we develop a novel and useful strategy for conflict resolution, and compare and contrast it with existing strategies. Ideally the process of conflict resolution should conform with the principle of Minimal Change and should result in the minimal loss of information. Our approach to minimizing the loss of information is to weaken information involved in conflicts rather than completely discarding it. We implemented and tested the relative performance of our new strategy in three different ways. Surprisingly, we are able to demonstrate that it provides a computationally effective compilation of the lexicographical strategy; a strategy which is known to have desirable theoretical properties.