In this dissertation I present research on the creative aspect of language use. I focus on Blind Variation and Selective Retention, a process whereby a creative system produces an outcome by first generating with no imposed constraints all variant outcomes and then retains from these variegated candidates the optimal outcome by gradually introducing selective constraints. I carry out a dynamical systems analysis of a model incorporating a maximum entropy construction based on pairwise correlations among interacting elements of the system and a Metropolis walk in energy space. I conduct computational and behavioral experiments to test the validity of the outcomes emerging from the model. The scientific motivation is to understand the processes underlying creative cognition, including how previously impossible outcomes can be discovered and produced by a creative system, and what factors contribute to the viability of creative outcomes.

First I study and analyze a Potts Hamiltonian model of Blind Variation and Selective Retention. When systems operate near a critical point, I show that the energy landscape described by the model can provide reasonable estimates of empirically observed data. I also show that the energy space decomposes into several clusters that promote discovery of viable unobserved states. I compare my results with other computational models and findings from human subject experiments, and show that my model consistently predicts outcomes that are novel, surprising, valuable, and intelligible.

The model I develop demonstrates the idea that creative processes should be viewed as emergent, collective phenomena. This idea represents the crux of a long-standing debate in creative cognition. The challenge has been to move beyond concepts by developing robust mathematical models that can be probed and analyzed. The approach I propose provides a framework for meeting this challenge.