On the merging of Dung's argumentation systems

In this paper, the problem of deriving sensible information from a collection of argumentation systems coming from different agents is addressed. The underlying argumentation theory is Dung's one: each argumentation system gives both a set of arguments and the way they interact (i.e., attack or non-attack) according to the corresponding agent. The inadequacy of the simple, yet appealing, method which consists in voting on the agents' selected extensions calls for a new approach. To this purpose, a general framework for merging argumentation systems from Dung's theory of argumentation is presented. The objective is achieved through a three-step process: first, each argumentation system is expanded into a partial system over the set of all arguments considered by the group of agents (reflecting that some agents may easily ignore arguments pointed out by other agents, as well as how such arguments interact with her own ones); then, merging is used on the expanded systems as a way to solve the possible conflicts between them, and a set of argumentation systems which are as close as possible to the whole profile is generated; finally, voting is used on the selected extensions of the resulting systems so as to characterize the acceptable arguments at the group level.