How is the reference of a virtue predicate like ‘honest’ fixed? How is the reference of a vice predicate like ‘cowardly’ fixed? Two candidates from philosophy of language suggest themselves: descriptivism and direct reference. Very roughly, on the descriptivist view, the meaning of a trait term is given by a set of associated predicates which, when satisfied, indicate that the term applies. For example, someone is honest if and only if she never lies, never cheats, and never steals: H(s) iff ~[L(s) v C(s) v S(s)]. On the direct reference view, by contrast, the meaning of a trait term is fixed by an inaugural act of directly referring to a property. Honesty is whatever trait she has.
Linda Zagzebski argues somewhere (I forget where) that virtue and vice terms are better understood on the direct reference model than the definite description model. I find that claim preposterous as a universal generalization, but I think it might apply to some trait terms: eponymous ones. Some examples are ‘maverick’, ‘quixotic’, ‘chauvinistic’, ‘sadistic’, ‘draconian’, and ‘quisling’. You might even think that ‘Christian’ fits the mold. One interesting thing about such trait terms is that, if their meanings really are fixed by direct reference, then it should be possible to make new ones. And if it’s possible to create new virtue terms, it might just be possible to create new virtues.