Modal provability foundations for argumentation networks (Q2269511)

An \textit{argumentation network} is a graph where an arrow from a point \(A\) to a point \(B\) is interpreted as: argument \(A\) attacks argument \(B\). The author reminds the reader of the notion of a \textit{Caminada labelling} that assigns to each argument either 1 (argument accepted), 0 (argument rejected), or ? (argument unjudged), with constraints such as: if an argument is accepted then any argument it attacks is rejected; if all arguments that attack an argument are rejected, then the latter is accepted, etc. Extra conditions can be imposed on a labelling to yield desirable properties. The paper presents a framework to capture the Caminada labellings in a modal-theoretic setting as follows. Atoms represent arguments (points of the network). Intended interpretations are restricted to Kripke frames consisting 3 possible worlds ordered by the accessibility relation as a chain \((M_1,M_2,M_3)\). Some axioms guarantee that any atom \(q\) can be interpreted in one of 3 possible ways: true in all possible worlds, false in all possible worlds, or true in \(M_1\) and \(M_3\) but false in \(M_2\), making \(q\) \textit{of type} \(1, 0\) or ?, respectively. A modal formula \(\mathbf{m(P)}\), function of an argumentation network \(\mathbf P\), is then defined and proved to have as models of the kind described precisely the Kripke frames that correspond to a Caminada labelling of \(\mathbf P\), in the sense that an argument is of type \(1, 0\) or ? in a model of \(\mathbf{m(P)}\) iff it is accepted, rejected or unjudged in the associated Caminada labelling. The author explains how the correspondence stems from the work he did on an interpretation of logic programs with negation in a more general modal-theoretic setting, where the accessibility relation of the intended Kripke frames has the structure of a tree rather than the structure of a chain. Then the author discusses the benefits of his approach. By removing one of the axioms, he makes it possible to have a fourth type of assignment, where an atom \(q\) can also be made true in \(M_1\) and false in \(M_2\) and \(M_3\). He provides an example where two argumentation networks can be distinguished in terms of their sets of models in the relaxed setting, whereas they have the same Caminada labellings. He also explains how natural logical concepts, such as logical consequence and substitution, lead to interesting results or considerations, that call for a corresponding interpretation in terms of Caminada labellings. The meaning to give to substitution, where an argument network ``takes the place'' of an argument in another argumentation network, is particularly delicate. The author provides the fundamental intuitions on substitution and announces results to be given in another paper.