Probabilistic Argumentation: An Approach Based on Conditional Probability –A Preliminary Report–

A basic form of an instantiated argument is as a pair (support, conclusion) standing for a conditional relation ‘if support then conclusion’. When this relation is not fully conclusive, a natural choice is to model the argument strength with the conditional probability of the conclusion given the support. In this paper, using a very simple language with conditionals, we explore a framework for probabilistic logic-based argumentation based on an extensive use of conditional probability, where uncertain and possibly inconsistent domain knowledge about a given scenario is represented as a set of defeasible rules quantified with conditional probabilities. We then discuss corresponding notions of attack and defeat relations between arguments, providing a basis for appropriate acceptability semantics, e.g. based on extensions or on DeLP-style dialogical trees.