In this paper, we propose a functional approach to overcome the lack of expressiveness of predicative approaches for extensional reference resolution. This later approach indeed forces designers of dialogue systems to use specific heuristics when solving some reference cases, which they can't integrate in a unified representation framework. As we focus on extensionalreference resolution in a generic framework for human computer dialogue system, we point the fact that the process of resolving referential expressions needs to take context into account even for expressions involving intrinsic characteristics of elements. Thus we propose arepresentation model for the semantics of intrinsic referential extractors. This model is built on three functions, the first one serves to calculate the similarity ratio between two elements, the two others serve to partition a reference domain which is previously sorted by the first function.