G. Halford, J. Wiles, M. S. Humphrcy, W. H. Wilson
This paper presents a computational model, using a Parallel Distributed Processing architecture, of the role of memory retrieval and analogical reasoning in creativity. The memory model stores information as the tensor product of up to three vectors representing, for example, context, cue, and target. The model can retrieve information in the form in which it was stored, or it can use two cues to generate an item that has been separately stored with each of them, but has never been associated with both cues jointly. This means that the intersection of two sets can be computed, without having been stored, thereby providing for the generation of novelty. For analogical reasoning, predicate-argument-bindings are represented as the tensor product of vectors representing relations and their arguments. The model can simulate the transfer of relations from one domain to another, as occurs in the creative use of analogy.