In this paper an inference calculus containing a Specialisation Inference Rule in the paradigm of multiple-valued logics is presented. The calculus is implemented using techniques of partial deduction, and is shown to be sound and complete for atom deduction.
The communication between autonomous agents based on this calculus is much more cooperative than the classical one: The answer to a query is now a set of specialised rules and propositions. Our system is thought for the cooperation among agents via the communication of knowledge, not just data, in a similar way to other systems [2], where the communication is about lamdba-formulas; or the communication of inductive inferences as in [3], a work on multi-agent learning systems.
The specialisation calculus is also related to other work on conditioned answers [4, 16, 19] and on the treatment of unknown information [21]. It allows us to obtain conditioned answers after the specialisation of a rule base with the known information. Our system is able to give back useful answers even in the case of partially known information.
The main difference of specialisation calculus with respect to other uses of partial deduction, is that it is based on a multi-valued propositional language and it is oriented to the improvement of the communication among agents, not just efficiency.
This specialisation calculus can also be used to make validation of rule bases. Consider that a physician has a general rule base for pneumonia treatment, and that he wants to check it in a restricted context such as: `women with gramnegative rods'. The specialisation mechanism allows him to obtain a new rule base specialised for pneumonia treatment in the particular case of women with gramnegative rods. The expert should agree with the behaviour of the new rule base so obtained, in that restricted context, because it is a specialisation of its original one, otherwise he must revise it. To check the behaviour of this reduced rule base he can apply any classical method (v.g. by case analysis), but to a much more reduced one, and this is the advantage of the use of the specialisation calculus. This specialisation mechanism can also be understood as a way of modularisation, by contexts, of flat and non-structured rule bases. This methodology gives then a more comprehensive and systematic way of validating rule bases than the standard methods.