Dung’s abstract argumentation model consists of a set of arguments and a binary relation encoding attacks among arguments. Different acceptability semantics have been defined for evaluating the arguments. What is worth noticing is that the model completely abstracts from the applications to which it can be applied. Thus, it is not clear what are the results that can be returned in a given application by each semantics. This paper answers this question. For that purpose, we start by plunging the model in a real application. That is, we assume that we have an inconsistent knowledge base (KB) containing formulas of an
abstract monotonic logic
. From this base, we show how to define arguments. Then, we characterize the different semantics in terms of the subsets of the KB that are returned by each extension. We show a full correspondence between maximal consistent subbases of a KB and maximal conflict-free sets of arguments. We show also that stable and preferred extensions choose
some consistent subbases of a base. Finally, we investigate the results of three argumentation systems that use well-known attack relations.