How Hybrid Should a Hybrid Model Be?

We present a perspective on the nature of hybridness withinhybrid symbolic/connectionistmodels of cognition. We discuss the qualities that differentiate apparently hybrid models from those which truly accrue properties of the complementary types of technology, and argue that the latter type of model is constituted by a stringent set of mappings between the component systems, mappings which preserve the semantic content of representations and algorithmic relations over them, as well as critical functional properties of the underlying technology. These mappings are exemplified in the techniques of skeletonisation and implementation, but are absent from cluster-analytic techniques and from hybrid models arising from the juxtaposition of separate connectionist and symbolic physical systems. The development of additional constraints for defining appropriate mappings between hybrid models and the resultant types of hybrid models remain areas for future research.