There is a forthcoming volume edited by B. Samuels entitled Beyond markedness in formal phonology which addresses the question. Basically, this is a case where the term is taken to be primary, and the referent is taken to be "open to discovery" – which is a nice way to say that it doesn't have a fixed meaning, and depending on school of thought, it refers to unrelated facts. Trubetzkoy's imported the concept of a "mark" into linguistics, and used it to refer to a distinguishing property that a linguistic unit has. In phonology, it was intimately tied to privative features, where a voiced consonant might "have the mark" and voiceless ones might lack it. Subsequent developments retained the term but radically altered what it was about. This was especially necessary when Trubetzkoy's privative analysis was replaced with Jakobsonian binary features, whereby all oppositions become equipollent. The nature of being "marked" then had to change, and it changed in the direction of being "more basic" versus "less basic", with a presumed acquisitional bias in favor of the "more basic" value.
Greenberg explicitly tied the notion of "marked" with frequency of occurrence, and Chomsky and Halle in SPE followed up on this assumption by creating a formal theory of "markedness" whereby rules were simpler to express if they produced unmarked results – consequently, the thinking was, unmarked outputs will be more frequent. Generally speaking, since that time the term has been taken to mean "happens most often", and the puzzle then is, what is the nature of the fact that causes something to be more frequent (i.e. "unmarked"). SPE held that it was a list of context-sensitive specifications that come "for free". More recently, "markedness" has been taken in OT to mean "a configuration that is to be avoided".
It is some note, IMO, that there is negligible interaction between the concept of markedness in phonology, and its use in semantics.