“She offered no argument”: Constrained probabilistic modeling for mnemonic device generation

A common aspect to creativity as described by creative theorists is the juxtaposition and balance of two opposing qualities, namely novelty and typicality. Practical models of computational creativity are needed that effectively leverage the contributions of each of these qualities in a synchronous manner. We discuss the effectiveness of constrained probabilistic models in representing this duality in generative models of creativity. We illustrate constrained Markov models as an example of a constrained probabilistic model and demonstrate its application to computational creativity in the elaboration of a system called NhMMonic for generating mnemonic devices. We demonstrate the effectiveness of the system¹ using a qualitative survey. Our findings suggest that the constrained Markov model is particularly effective at generating mnemonics that exhibit novelty and typicality in grammatical and semantic flow with the overall result of more effective mnemonics for the purpose of memorization. Source code as well as our mnemonic device generator are both freely accessible online.