C. MacNish, Antony Galton, and John Gooday
In Erik Sandewall and Karl-Gustav Jansson (eds.) Scandinavian Conference on Artificial Intelligence, IOS Press, Amsterdam, 1993, pages 30-37.
It is widely recognised that qualitative knowledge alone is insufficient in reasoners which are to design, understand and predict the behaviour of physical systems. There is little consensus, however, on how quantitative information should be integrated with qualitative reasoning.
The philosophy of the qualitative reasoning community has traditionally been to admit limited quantitative "landmarks" or to delay consideration of quantitative information until a complete qualitative model has been built, thereby allowing the development of qualitative models which have no physical realisation. The constraint logic programming (CLP) approach, on the other hand, embeds quantitative constraints directly within the clauses of a Prolog-like deductive system. This has the effect of disguising the qualitative model and brings with it the difficulty of incorporating rich computational domains within a traditional deductive system.
This paper proposes an alternative approach, called quantitatively scoped qualitative reasoning (QSQR), in which quantitative information is used to guide qualitative deductions. This operational view of quantitative constraints has the advantage that quantitative information can be used to prune the space of qualitative models, while at the same time remaining outside of the qualitative description.}