Closed tfgordon closed 12 years ago
the standard fixpoint approach may be helpful here, where evaluation is not recursive but goes only two steps: statements acceptability (in/out) -> arguments applicability -> statements acceptability (in/out)
Argument construction and argument evaluation are now completely separate. A protocol for argument evaluators has been defined and an the Carneades evaluator has been reconstructed as a structure implementing this protocol. This new version of the Carneades argument evaluation structure now handles cyclic argument graphs. Graphs without cycles should have the same semantics as before. In graphs with cycles, the labeling of the statement and argument nodes can be partial. The scintilla of evidence proof standard has been removed, to enable all statements to be :in (P acceptable), :out (not P acceptable) or :undecided (neither P nor not P acceptable). Similarly, argument nodes now also have three labels: :in (applicable) :out (not applicable) or :undecided.
An argument evaluator implementing Henry Prakken's ASPIC+ model is planned for the near future, to be implemented using the Dung-O-Matic implementation of Dung Argumentation Frameworks from Chris Reed's group at the University of Dundee. This evaluator will also be able to handle cycles.
Provide support for evaluating arguments in graphs with cycles. See the extensions to Carneades proposed by Gerd Brewka (Abstract Dialectical Frameworks) and Henry Prakken (based on ASPIC+).