On Thursday, distinguished cognitive scientist Mark Turner visited campus and gave a great lecture on conceptual blending. I was a little familiar with this from Fox Harell’s work, but Turner’s lecture was very revealing on the cognitive roots of conceptual blending.

The gist of it works like this: Human cultural development only really began recently in our evolutionary development. For 800,000 years on earth, biological humans used the same stone tools in the same way without variation. It is only extremely recently, in the past 50,000 years, that our range of potential behaviors began to expand. But: it began to expand dramatically. Turner’s point of interest is that humans began to develop culture and language, but it is bewildering to understand how and why they exist.

So, the real question is how we form new concepts, and create new behaviors. Turner’s solution to this is conceptual blending, specifically double-scope blending, which can combine two conceptual domains (which are in conflict), and produce a new and unique conceptual domain, where new meanings can be made. This idea is great, but it is necessary to pull back to a couple of interesting ideas that are touched on.

One is that a conceptual domain, or a frame, can be much more broad and general. Turner gave examples of memories, structured expressions in language, and also physical engagement. These have the properties of conceptual models. The other thing about models is that the types of models represented here are not abstract and propositional, but they are embodied (generally) and procedural. Thought involves running a model, or simulating it. Mammals have the capacity to simulate models: think of playing fetch with a dog. A dog can catch all manner of objects flying through the air. Some sort of mental calculation is taking place, and it is easily argued that this is an execution of an embodied model. so this modeling is a very basic and intrinsic ability.

A conceptual blend occurs when there are two conflicting conceptual frames or models at work in a situation. Turner noted that there is a capacity for humans to hold two different frames of thought in mind simultaneously. When he did this, I immediately thought back to AI and cognitive architectures focused around planning. Generally, these only define one sort of cognitive frame, and have difficulty when modeling two thoughts at once. Examples of multiple thoughts are thinking of memories and going about everyday tasks. Some work has been done regarding this recently, but I’ll get into that later. The point is that it is a complete departure from the models of commonly used AI.

What is interesting about conflicts in models is that they are not mentally discouraged, they instead trigger thought. This is especially the case in children, who learn concepts and combine them very rapidly during development. In a double-scope blend, the two domains must be in conflict. For instance, a good example that Turner mentioned is Harold and the Purple Crayon. The story combines two domains: drawing with a crayon, and the physical world. The trick is that anything Harold draws becomes real. So, these domains are immediately in conflict, because, we know (and kids know too) that things that are drawn do not become real. That is the blend that occurs in this domain, though. Elements from the domain of drawing, and from the domain the physical world are selectively combined. New meanings and properties emerge that are totally new, for example: Harold wants to get home, and sees the moon in the sky, and remembers that he can see the moon from his window. When he draws a window around the moon, suddenly he is home. This logic is magical, but it is absolutely consistent with the model formed by the blend.

The topic of conceptual blending is of limited use in the simulation work that I am trying to do, but it is very useful from the perspective of understand how real people might make sense of models represented within a simulation game, and apply those to the external world. It also does something to explain the value of adaptations in general. You can think of a fictional artifact as defined by a model, which is a blend of two things: the model of the medium, and the underlying model that defines the work. An adaptation should take that underlying model, and combine it with a new model that is the new medium. An individual’s interpretation of a work is going to form a new blend, though, which will be between the individual’s experience, and the perceived work. When we account for the idea of individual and cultural interpretations, we can have a new model, which is a blend of the interpretations of a community. This idea is running away with the idea of conceptual frames that Turner originally defined, which are all internal, much smaller and more precise, but it is a reasonable direction for thought.

It would be good to think more about formal and computational models for conceptual blending. I kept wanting to ask Mark Turner about computational models when he was taking questions, and then realized that is exactly what Fox Harrell‘s dissertation is all about. That would be good reading material. Relating blending to AI, is a major topic in Jichen Zhu‘s dissertation as well.