Names come from people, so the probability being distributed is the probability that you'll get a given name from a given person for a given thing. Take your sp

Names come from people, so the probability being distributed is the probability that you'll get a given name from a given person for a given thing. Take your sphere-to-disk example. The bigger the hole gets, the less likely it is that a given person will describe it as a hole. But at various points you'll get spikes where most people say "hole," "bowl," "disk," etc. Personally I think it's silly to take this as proof that the definition of hole "isn't working". Probability distributions are good enough for quantum mechanics and fluid dynamics, but not language? Also look around: people communicate about holes productively (i.e. to accomplish work) all the time, so obviously the definition is working just fine under many conditions. Language is one of those things that works way better in practice than in theory.