People of all ages and demographics exhibit impressive creative abilities when playing. Pretend play exemplifies this creativity: people blend imaginary worlds with local objects to act our various scripts and roles. While many have researched aspects of social interactions and cognition in pretend play, there do not yet exist formal models of the process of engaging in pretend play. Computational play is our effort to develop computational models of pretend play implemented in agents that can co-create imaginary play worlds with humans. These models open the possibility for more playful interaction with the increasingly prevalent artificial intelligences in our world (e.g., Apple's Siri).
As a first step toward full computational play agents we developed a model of how objects are used in pretend play. Props such as cardboard tubes are often employed as stand-ins for imagined objects like lightsabers. Object blending is a model of the cognitive processes involved in mapping pretend object properties onto real-world objects. This model combines insights from Fauconnier and Turner's Conceptual Blending Theory (2003) with mental models of categories based on Rosch's prototype theory (1999). Object blending provides a computational grounding for play agents to use a set of real-world props in imaginary scripts—e.g., using a cardboard tube as a lightsaber.
Our work on computational play models led to further insights on the nature of Conceptual Blending Theory. In particular, developing these models highlighted a gap in existing models around goal-driven blending. Together with Boyang "Albert" Li's work on methods to generate pretend gadgets we developed a model of goal-driven conceptual blending that extends Brandt and Brandt's (2005) approach to Conceptual Blending Theory.