Every apple is a real apple.

Take every apple that ever existed, whether real or imaginary. Combine them all into a single structure - not exactly by averaging, but by overlaying the different pieces of information together to get a cohesive whole. The result would be something that, basically, averages out to an apple in theory but doesn't necessarily resemble a physical apple.

That's okay, though, because we're not after a physical apple. We're after a potential apple - in essence, a kind of mathematical function that defines the limits within which "apple" exists. The limits are fuzzy, just like the edges of our structure: some characteristics don't start and stop so much as fade, and fractal math states that such edges must be infinite in gradiation. In linguistic sense, you're left with a bunch of traits that are "apple", a bunch that are "apple-ish", and some that "tend to be apple" at this end but gradually become less apple-y at the other.

Now, let's use that structure - that equation or description, however you want to think of it - as our definition of "apple". Suddenly, every apple is a real apple, because we're not looking for specifics so much as trends or patterns.

This is hard for a lot of people to grasp: we've traded something hard and explicit for something vague, but in doing so end up with a better definition and, in actual fact, a better understanding of that to which we're referring. Whereas before nothing could be said to be an apple (and thus the definition useless), we can now define with reasonable precision what is or isn't an apple and use that data for further explorations (such as defining what is or isn't an apple pie).