The ability to conceive objects that do not exist is a hallmark of human creativity. Many science fictions, fantasies, and fairy tales contain imaginary objects such as lightsabers in Star Wars and the magic mirror in Snow White. These objects contribute significantly to story developments and the fun of reading. Indeed, the "gadget story" is considered as a major subgenre of science fiction. The cognitive construction of such objects, which we call gadgets, is a creative act that we attempt to imitate with an Artificial Intelligence.
As an effort to reverse engineer gadgets in real stories, we study Doraemon, a hugely successful Japanese manga. The story theme is that Doraemon, a robot from the future, helps primary school students to cope with their daily problems such as exams and bullies by using futuristic gadgets indistinguishable from magic. It is estimated that the entire Doraemon series contains over a thousand gadgets. The abundance of gadgets makes Doraemon a unique opportunity for studying and imitating cognitive construction of imaginary objects of human.
We surveyed five volumes of the manga and proposed 9 techniques that the author may have used to create gadgets. These techniques are unified into a 5-step process. We designed an algorithm for generating gadgets based on our taxonomy, utilizing a combination of planning, analogical reasoning, and knowledge of everyday objects and tools to create gadgets serving narrative purposes. The algorithm can reproduce gadgets in Doraemon. To our best knowledge, this is the first attempt at the generation of novel gadgets as part of AI storytelling.
We also found that the theory of conceptual blending can explain many gadgets in the manga. Our algorithm closely resembles the goal-driven conceptual blending process proposed by Brandt and Brandt (2005).
The ACM C&C paper contains an overview and an analysis of Doraemon. The INT4 paper explains the algorithm in details. The ICCC paper contains my thoughts on the algorithm's link with conceptual blending theory.