As slippery as an eel
So on to the promised language-oriented thing:
I've been interested in idioms for a long time but like many researchers, have never had a really solid criterion, or even a semi-solid criterion, for drawing a boundary line between an idiom and a metaphor. Some things are clearly idioms ('bell the cat'), some clearly metaphors ('All the world's a stage'), but for many expressions, it's quite unclear whether to locate them in one box or ther other. In the course of a lecture in Stuttgart, a student was asking about an expression in Japanese, 'make X dance', meaning to control someone as a puppet, pull their strings, so to speak (there's another one!). We were trying to decide whether that was a real idiom or not...
...and Michal Starke suggested that one possible way to distinguish might be to ask whether the interpretation is available after literal translation from the language under investigation into some other language. That seemed to draw some clear distinctions—'X kick the bucket' doesn't translate, 'fear gripped X' does translate; we decided the 'make X dance' expression also translates.
This is clearly not going to make things black and white -- metaphors can depend on culture-specific entities (like stages or puppets) for which literal translation equivalents might not be findable in the target language because of cultual differences. Furthermore, closely related cultures/languages, in contact for one reason or another, might borrow idioms--I recently learned that the British English idiom 'pull the chestnuts from the fire', meaning to solve the problems or save the situation, also exists in Italian, which seems like a possible candidate for a borrowed idiom. But nonetheless, it seems like a possible start.
Aside: Did you notice that in that story about the 3.3 million-year-old Australopithecus afarensis infant fossil, they found a hyoid bone, to which tongue and throat muscles attach? The bone differs significantly between apes and humans, and the differences may be related to the language adaptations of the human vocal tract. The little female's hyoid looked more apelike than humanlike, although she walked upright. So did her arms and shoulders.
Update: Ah, of course the Language Loggers noticed. Hey, could one call the Language Loggers 'Language Lumberjacks'? That'd make a fun Ling 201 constitent-structure bonus question.