The first extended transformational-generative grammatical study of any aspect of a language written by anyone other than Noam Chomsky was the study of nominalizations by Robert B. Lees in his MIT dissertation, published as a monograph in 1960. In it, Lees attempted, among other things, to offer a detailed treatment of noun-noun compounds. Other early studies in generative grammar followed. Part of what they were attempting to do was to give a syntax for nominal compounds that would explain what patterns of meaning were available in noun-noun compounds: tree house means "house in a tree" (a location relation), while lion king means "king who is a lion" (a predication relation), and tax collector means "collector of taxes" (a verb-object relation), and so on. I never thought such research was on the right track. It seemed to me that the semantics of such noun-noun combos was so protean that nothing could ever come of it. And I was reminded on this the other day when I saw this headline in a British newspaper:
Detective attacks jailed canoe wife who lied to sons
What, I hope you are asking yourself, is a canoe wife?
It turns out to mean (as those following stories in the popular press within the UK would know) to mean "wife of the man who pretended to have drowned while out on the ocean in a canoe".
The story involved a dishonest couple, John and Anne Darwin, of whom the lead detective in the case disapproved strongly enough to publicly call them "despicable". Facing a certain amount of debt, they faked the husband's death, making it look like a canoe outing off the Durham coast in northern England had ended in a drowning tragedy. Executing the plan involved Anne Darwin telling their two grown sons mournfully that their father had perished, and watching them go through the grief process (one of them took it especially hard). John Darwin hid out in his home town, living secretly with his wife, for four or five years, and then the couple started making plans to move overseas. On a visit to Panama the couple were stupid enough to let a property agent photograph them, and the photo was put on the web. In December 2007 John Darwin conceived of the brilliant idea of simply walking into a police station (while Anne was back in Panama) saying he had amnesia and didn't know who he was or where he had been. But he had overestimated police gullibility. Detectives rapidly suspected that he was a liar. The game was finally up when someone simply typed "John, Anne, Panama" into the Google search box and Google images came up with the Panama photo. John was charged and convicted. Then they looked into the Anne's culpability. In the end, convicted of six counts of fraud and nine of money laundering, she actually got a slightly longer sentence (six and a half years) than he did.
[This account of the Darwins' fraud was revised and corrected with the help of the Wikipedia entry after this was first posted.]
In my view, there is no sense in trying to develop a taxonomy of possible semantic relations that noun-noun compounds can express, given that one of them would apparently have to be a relation that permits N1 N2 to hold of a person x iff N2 is the name of the relation that x bears to some person y such that y was involved in an incident in which an object of the type N1 played a salient role. Define the notion "natural semantic relation" as you will, this surely isn't one.
It looks to me like a noun-noun compound N1 N2 can be formed given just about any relation between N1-type things and N2-type things that turns out to be a relevant one for the description of some situation.
Addendum: By the way, irrelevantly, Howard Jacobson wrote a column in The Independent in which he quoted a headline saying "Despicable canoe couple sent to prison" and added that it "gave one pause". But his reason for puzzling over it was idiotic: he said, "What's a despicable canoe? In fact it was the couple who were "despicable" in the judge's view on account of their deception of their sons." He's just trying (and failing) to make a stupid joke, I think. The point is not to find perversely misguided parses. The interesting thing is the noun-noun compound canoe couple, not the fact that (like any noun) it can be modified by an attributive adjective.