This is a comprehensive review of human factors affecting interaction with nearly any "formal" system where "formal" means anything from a color wheel to a prolog database.
I have repeatedly encountered these issues while working on a formal language for specifying scientific protocols. I knew I had my work cut out in terms of finding a way to create an interface that non-expert users could work with, but this makes it clear just how critical it is for the tools not to distract the user from their expert thinking.
This poses and incredible design challenge, and that is without any discussion of variability between users!
What predicts the aptness or ineptness of a given formalism? It has to begin with an idea of what the user already knows, and with the fitness of the formalism to the user's concept of the job to be done. Perhaps for the majority of the HN readership: "Informality Considered Harmful".
In your example clustering is probably an abstraction but the rules for choosing names might be a formalism. Unsupervised vs supervised learning is another comparison that this connotes..