Processing Inconsistency of Knowledge on Semantic Level Ngoc Thanh Nguyen (Institute of Control and Systems Engineering, Wroclaw University of Technology, Poland thanh@pwr.wroc.pl) Abstract: Inconsistency of knowledge may appear in many situations, especially in distributed environments in which autonomous programs operate. Inconsistency may lead to conflicts, for which the resolution is necessary for correct functioning of an intelligent system. Inconsistency of knowledge in general means a situation in which some autonomous programs (like agents) generate different versions (or states) of knowledge on the same subject referring to a real world. In this paper we propose two logical structures for representing inconsistent knowledge: conjunction and disjunction. For each of them we define the semantics and formulate the consensus problem, the solution of which would resolve the inconsistency. Next, we work out algorithms for consensus determination. Consensus methodology has been proved to be useful in solving conflicts and should be also effective for knowledge inconsistency resolution. Keywords: Inconsistent knowledge, conflicts, consensus methods Categories: E.1, H.2.1, I.2.4, I.2.11 1 Introduction In many practical situations in order to solve a problem one often has to gather knowledge from different resources for realizing the task. Nowadays owing to modern computer technologies gathering knowledge is not a hard task at all, but there may be two features of this knowledge, which often cause the decision making process difficult. The first feature is related to the big amount of knowledge, which on one hand contains many useful elements, but on the other hand it often contains also a lot of useless elements. For this problem many methods for information filtering and ordering have been proposed. The second refers to the consistency of the gathered knowledge. Some elements of this knowledge may refer to the same subject, but they are not coherent. Hunter [Hunter, 98] describes the resources of inconsistent knowledge as situations in which one obtains “too much” information. Generally, a problem of knowledge inconsistency may be formulated as follows: For a given set of logic formulas, one should find out if these formulas are inconsistent or not. As stated in [Nguyen, 04a] inconsistent knowledge can be considered on two levels: syntactic level and semantic level. On the syntactic level inconsistent knowledge as a set of logic formulae is assumed to have no model [Kifer, 92], that is on their basis one can conclude false. For example, for the set of formulae {¬α∨β, α, ¬β, α∨¬β} with the inference engine of standard logic one can easily notice that no Journal of Universal Computer Science, vol. 11, no. 2 (2005), 285-302 submitted: 31/7/04, accepted: 15/12/04, appeared: 28/2/05 © J.UCS