|
Abstract : |
Most rule-based systems are imperative, or, if they are mainly declarative, also provide some imperative features. It is argued that a strictly declarative formalism with a sound logical basis is preferable. This paper addresses the task of implementing forward chaining efficiently for declarative rulebases. Deductions that can be invalidated by new information are particularly difficult to deal with consistently and efficiently. The paper shows that efficient forward chaining is possible, by utilising the implicit structure of the rulebases to optimise the most common circumstances. A language, called SDML, which utilises such declarative rulebases to simulate multi-agent models, is described. 1, |
