Probabilistic argumentation frameworks: Basic properties and computation

Abstract

In this paper we analyze probabilistic argumentation frameworks PAF, defined as an extension of Dung abstract framework in which each argument n is asserted with a probability pn and where an argumentation semantics is used to compute arguments' status. We start by extending recent definitions of PAF removing the hypothesis of arguments independence, extending the computation to preferred semantics and defining the distribution of various probabilities induced over arguments acceptability status. We then prove some basic properties linking grounded and preferred PAF and we describe the first algorithm to compute the probability of acceptance of each argument. We end our work with an application of PAF to legal reasoning. © Springer-Verlag Berlin Heidelberg 2013.