摘要：This paper defines in a fonnal and computational way the notion of 'consilience', a tenn introduced by Whewell in 1847 for the evaluation of scientific theories. Infonnally, as has been used to date, a model or theory is 'con"silient' if it is predictive, explanatory and unifies the evidence. Centred in a constructive framework, where new tenns can be introduced", we essay a fonnalisation of the idea of unification based on the avoidance of 'separation'. However, it is soon manifest that this classical approach is vulnerable to the introduction of fantastic concepts to unify disparate sub-theories. Our second approach is constructed by using a detailed evaluation of the relationship between the theory and the evidence by means of reinforcement propagation. With the use of reinforcement, fantastic concepts can be better detected and the role of consilience for theory construction and revision can be specialised for different inference mechanisms like explanatory induction, abduction, deduction and analogy
