University of Liverpool
4pm Tuesday, 2nd of June, 2009
Room 4.31/4.33, Informatics Forum
Computational argumentation considers models by which formal notions of "maximal collection of acceptable arguments" can be defined. In a now widely adopted approach -- that of Dung (1995) -- such concepts are set within the context of a directed graph (X,A) whereby X defines a (finite) set of "arguments" (treated as atomic entities) and A describes a binary "attack" relation, so that (p,q) in A (read as the argument p attacks the argument q) captures a sense that that p and q are incompatible. In this setting appropriate collections of acceptable arguments are given by subsets, S of X, that satisfy particular criteria, E(S), w.r.t. (X,A).
Recently a number of proposals have been advanced defining alternative criteria to those studied in Dung's original work, e.g. the semi-stable semantics of Caminada (2006), the Ideal semantics of Dung, Mancarella, and Toni (2007), the resolution-based semantics of Baroni and Giacomin (2008). In addition models by which the basic argument/attack structure is extended have been put forward, notably value-based frameworks (Bench-Capon, 2003) and weighted systems (Dunne et al., 2009).
In this talk, after reviewing the principal decision questions arising in such extension based approaches and complexity properties of the traditional semantics the characteristics of these novel proposals from a complexity perspective will be considered.