In this paper, starting from a central intuition of the French linguists Anscombre and Ducrot (1983), I address the question of how to represent the types of dependence which connect propositions in linguistic argumentative orientation. First, I present the notion of probabilistic confirmation (2.1), before applying it to scalar predicates. Next, I present some potentially problematic observations (3.1) and show how the confirmation-based approach can be extended to cover such cases (3.2).