Toward Higher-Order Qualitative Representations

Qualitative reasoning typically proceeds by instantiating type-level constraints to produce a complete propositional model of a scenario, and then reasoning over this propositional representation. As we pursue the problem of learning qualitative models and applying them to openended construction domains, the limitations of this approach have become increasingly acute. Consequently, we have been developing representations and techniques for higherorder qualitative reasoning, which supports qualitative reasoning without needing to generate a full propositional scenario model. We present our type-level vocabulary and describe our experience with it in learning qualitative influences in a strategy game domain.