Argumentation frameworks have aroused intense interest from the AI community over the past years. Dynamic aspects of argumentation frameworks have received some interest from the community, but none of these works tries to address the recovering problem, that is, what shall we do when the new addition leads to the loss of all extensions. Such problem is typically a belief revision problem. In this paper, we propose a revision operator to revise an argumentation framework by another one, with the guarantee that the result of the operation will be an argumentation framework which has at least one stable extension. We also propose an algorithm to compute the revision operation outcome.

We generalize in this paper the Dung's abstract argumentation theory in order to represent, in addition to the attack relation, a particular kind of support relation which captures knowledge of the form : “argument a is necessary to obtain argument b”. Unlike a general unspecified support, the necessity relation has the advantage to ensure that its interaction with direct attacks generates new (indirect) attacks having exactly the same nature of the direct ones. We discuss the advantageous consequences of this specialization of the support relation on the acceptability semantics of the underlying bipolar framework. Then, we show a suitable translation of any such bipolar framework into a logic program without passing by a preliminary translation into the classical framework of Dung.

This paper shows that we can take advantage of information about the probabilities of the occurrences of events, when this information is available, to refine the classical results of diagnosability: instead of giving a binary answer, the approach we propose allows one to quantify, in particular, the degree of non-diagnosability in case of negative answer. The dynamics of the system is modelled by a reducible Markov chain. A state of this chain contains information about whether it is faulty (resp. ambiguous) or not. The useful refinements of the decision about diagnosability are then obtained from the asymptotic analysis of this Markov chain. This analysis may be very useful in practice since it may lead to take the decision of tolerating some non-diagnosable systems, if their non-diagnosability is not critical, and thus allows one saving the cost of additional sensors necessary to make these systems diagnosableThis work is part of DIAFORE project supported by ANR under grant ANR-05-PDIT-016-05.

We describe a heuristic method to order the events evoked in texts describing road accidents. The method consists first in generating, by non monotonic inference rules, a set of precedence and simultaneity constraints between temporal references linked to particular words of the text. Then, the proposed algorithm updates the narration order of the events to satisfy these constraints and give the real order of the events described in this text. The temporal information is represented according to a linear temporal model which is enhanced in a first order reified language.