Yes, it is all fuzzy logic from the get-go in the real world: grammatical classes are fuzzy (with very steep membership functions, to be sure), semantic webs are fuzzy, even physical characteristics (and indeed physical bodies) are fuzzy. But, happily, fuzzy things can be dealt with relatively unfuzzily at various practical levels. So, here, we can make specifications for 'olin', say, which work fine and also tell what sort of story is required to dodge the bullet. That doesn't make 'olin' a class, just a separate word, as it already was (tp is blessedly short of synonyms).
This has been a useful discussion, precisely because it brings these questions up and they need to be discussed at some point (maybe later, when there is a bit more culture to mess with). there are more pressing issues at the moment -- of the how to say correctly sort, for example.
By the way, don't confuse the value of the membership function with the the truth value of the claim that something is in the class. That is only one possible tv assignment, and often not the most useful one. For most practical cases, something about an acceptance function also plays a role, where acceptance presumably varies with membership but not necessarily directly. Of course, the strongest acceptance functions are those that give 0 up to a certain membership value and 1 thereafter, i.e., the preciding functions.