Eecient Defeasible Reasoning Systems 7. Current and Future Work Eecient Defeasible Reasoning Systems Levels(n)

On the theoretical side, our eeorts concentrate on the establishment of connections to argumentation systems 22;41 , and the development of semantics for defea-sible logic 21;42. Finally we are investigating the applicability of defeasible logic (and its variants) as a modelling languages in the domains of regulations, business rules, electronic commerce, and the legal domain. We believe that eecient connict resolution rule-based systems with priorities are a useful tool that can represent in a natural way many solution procedures that are currently applied manually. 8. Conclusion We have presented two new implementations of defeasible logic, based on substantially diierent techniques. Our experiments on query-answering implementations have demonstrated that both Deimos and the existing d-Prolog system can handle very large rule sets, although d-Prolog is eeective on only a narrow range of rule sets. Deimos is clearly superior in the more realistic situations when some rules connict. We have seen that the complexity of computing consequences in defeasible logic is linear in the size of the input theory. Our experiments with the partial implementation of Delores have connrmed this claim. Indeed the partial implementation of Delores was clearly the faster system in almost all experiments on which it could be run. However, the transformation implemented in full Delores did not behave linearly. Since theoretically it is of linear complexity, there is clearly an engineering issue to be addressed here. In summary, both Deimos and Delores show promise as high-speed implementations of defeasible logic, and Deimos has already partly fulllled its promise. Consequently it appears that defeasible logic provides rule prioritization and defeasible reasoning in an eeciently implementable way. Work is continuing on both systems. For Deimos, we are implementing memo-ization using mutable arrays, instead of a balanced tree, in order to eliminate the O(logN) factor. For Delores, we are addressing the problems of initialization and the pre-processing transformation that were exposed by our experimental evaluation. Acknowledgements We thank Scott Brady and Chris Herring for their work on a preliminary all-conclusions system, and Guido Governatori for his work on the transformations used in Delores and for discussions on defeasible logic. EEcient Defeasible Reasoning Systems tree(10,3)). The timings of such problems were the only ones to vary signiicantly when experiments were repeated. It turns out that the initialization of S consumes the bulk of this time. Furthermore, it is on those problems that contain many diier-ent atoms that Delores performs worst. This is …