A logical account of formal argumentation (Q2269515)

The paper starts with a reminder on notions relative to an \textit{argumentation framework}, that is, a binary relation \({att}\) on a set \({Ar}\); intuitively, \({att}(A,B)\) expresses that argument \(A\) attacks argument \(B\). A set \(\mathcal{A}\mathit{rgs}\) of arguments is conflict-free if it contains no pair of arguments one of which attacks the other, and it is said to \textit{defend} an argument \(A\) if for all arguments \(B\) that attack \(A\), some member of \(\mathcal{A}{rgs}\) attacks \(B\). A conflict-free set \(\mathcal{A}{rgs}\) of arguments is (i) \textit{admissible} if it defends all its members; (ii) \textit{complete} if it defends all its members and its members only; (iii) \textit{grounded} if it is complete and \(\subseteq\)-minimal; (iv) \textit{preferred} if it is complete and \(\subseteq\)-maximal; (v) \textit{stable} if it is complete and attacks all arguments it does not contain; (vi) \textit{semi-stable} if it is complete and its union with the set of arguments it attacks is \(\subseteq\)-maximal. Then the authors introduce the notion of an \textit{argument labelling}, that is, a function that maps every element of \({Ar}\) to either \({\mathbf{in}}\) (argument accepted), \({\mathbf{out}}\) (argument rejected), or \({\mathbf{undef}}\) (argument unjudged). They characterize all previous notions in terms of argument labellings with specific properties: (i) \(\mathcal{A}{rgs}\) is admissible iff it has a labelling such that if an argument is labelled \({\mathbf{in}}\) then all its attackers are labelled \({\mathbf{out}}\), and if an argument is labelled \({\mathbf{out}}\) then some of its attackers is labelled \({\mathbf{in}}\); (ii) \(\mathcal{A}{rgs}\) is complete iff it has a labelling such that an argument is labelled \({\mathbf{in}}\) iff all its attackers are labelled \({\mathbf{out}}\), and an argument is labelled \({\mathbf{out}}\) iff some of its attackers is labelled \({\mathbf{in}}\); (iii) \(\mathcal{A}{rgs}\) is grounded iff it has a complete labelling whose set of arguments labeled \({\mathbf{in}}\), or equivalently \({\mathbf{out}}\), or equivalently \({\mathbf{undef}}\), is \(\subseteq\)-minimal; (iv) \(\mathcal{A}{rgs}\) is preferred iff it has a complete labelling whose set of arguments labeled \({\mathbf{in}}\), or equivalently \({\mathbf{out}}\), is \(\subseteq\)-maximal; (v) \(\mathcal{A}{rgs}\) is stable iff it has a complete labelling with no argument being labeled \({\mathbf{undef}}\); (vi) \(\mathcal{A}{rgs}\) is semi-stable iff it has a complete labelling whose set of arguments labeled \({\mathbf{undef}}\), is \(\subseteq\)-minimal. In the second part of the paper, the authors embed the argumentation framework into various logical settings. Kripke models allow one to let possible worlds represent arguments, the accessibility relation, and the attack relation. The frames of the modal logic are fixed to chains of length 3; complete labelling can then be axiomatized. Another modal approach investigated in the paper borrows from Löb's modal provability logic, in which \(\boxminus\) is the modal operator of provability, to define a formula that logically implies \(\boxminus a\) if \(a\) is labelled \({\mathbf{in}}\), \(\boxminus\neg a\) if \(a\) is labelled \({\mathbf{out}}\), and neither if \(a\) is labelled \({\mathbf{undef}}\); that formula is obtained as a fixed-point solution in a suitable provability logic. Alternative representations of argumentation frameworks are also discussed: one representation is in a first-order setting with the class of intended interpretations reduced by circumscription, another representation is as logic programs.