For example, within my lifetime, fun switched from being just a noun to also being an adjective. That reanalysis happened in people's heads, and you can't detect it in people's heads, because that was fun makes sense whether fun is a noun or an adjective. You only realise, as an outside observer, that fun is now an adjective, because people have started using fun in contexts where a noun makes no sense: that was so fun, that was the funnest thing ever.
If you're trying to date linguistic change, you have a problem. Because the initial reanalysis happens in people's heads, you can't see it in textual evidence. The only thing you can see in textual evidence is how the reanalysis has spread to novel contexts. In fact, the only thing any other speaker of the language will notice is when the first speaker starts using the construction in novel contexts.
So the actual change to language as one person's internalised system happens with reanalysis. But language as spoken by a community only changes when everyone is in on the reanalysis; and for everyone to be in on it, the expression has to be used in unambiguously novel ways.
Reanalysis itself can happen in two ways. One is the way I've just described: the interpretation of the expression happens imperceptibly, because the expression has a structural ambiguity. Anything in English following that is can be either a mass noun or an adjective. The phrase anorange can be chopped up as either a norange or an orange. (And norange was the original form—as in Arabic nāranj, and Greek νεράντζι.) I will originally meant that you want to do something; but (if you have anything to do with it), it also contains the expectation that the something is going to happen in the future.
On the other hand, reanalysis can happen when someone takes an expression that is established in one domain, and starts using it in a similar way, in a completely different domain. For example, a rocket in Italian was originally a spindle, and a well established term in the domain of weaving—until someone got the idea of using rocket to refer to spindle-shaped projectiles, in the unrelated domain of ballistics. A head has a well established meaning in anatomy, which gradually built up connotations as the most important part of the body. (That's why κάρα in Ancient Greek came to be used in phrases like "the divine head of Jocasta is dead.") That notion of importance led "head" to be used in domains unrelated to either anatomy or containing a brain and an oral cavity: head waiter, head of the beach.
The second kind of reanalysis uses a concept in a new domain, with some but not all of the same meaning. That is of course exactly what a metaphor is, and this is metaphoric change. The first kind exploits ambiguity within the same domain, rather than making a conceptual leap. Because it's not metaphor, but an "adjacent" meaning, linguists have come up with the cleverness of calling this metonymic change, extending the original notion of metonymy. (The thinking is outlined in Hopper & Traugott's Grammaticalization textbook.) I can see why they did so: the White House and the US government are both on 1600 Pennsylvania Ave; wanting to do something and having something happen in the future are inherent in the same phrase I will. It's still overclever to call it that, but I don't get to write the textbooks.
Both metaphoric and metonymic change lead to reanalysis; but they work in different ways. A metaphor is an act of conceptual creativity; it only takes one clever speaker to come up with a new way of seeing things, and others will be impressed enough to follow suit. Metonymy, on the other hand, doesn't need cleverness; it can be positively hampered by cleverness. Metonymy relies on a different interpretation of an expression in the same domain; it relies on people misunderstanding what what meant. Metonymy builds on the ambiguities available from the context; the context is what speakers are scrambling for, because they usually didn't understand the original expression. The point of metaphor, on the other hand, is ignore the context: metaphor invents a new context for the expression.
I'm going through all this, because it comes up in my next post, which is reporting on a couple of blog posts and associated discussion, trying to explain how the Modern Greek expression αμέτι μουχαμέτι has developed. The expression, we can be reasonably sure, originates in the Ottoman Turkish ümmet-i Muhammed. The Turkish phrase means "nation of Muhammad", from the Arabic Ummah. The Modern Greek expression is close to "come hell or high water": it points to unreasonable insistence, pigheadedness: το έβαλε αμέτι μουχαμέτι να κάνει τον γιο του δικηγόρο, "he set out, ameti mouhameti, to make his son a lawyer".
If you're trying to work out how "Nation of Muhammad" ended up meaning "Come hell or high water", you can take it as a metaphorical change. My godfather, Tasos Kaplanis, is going to be annoyed with this sequence of posts, because I'm going to take his postings in vain twice. When the phrase first came up on the Magnificent Nikos Sarantakos' Blog, Tasos posted a comment that allows a metaphorical explanation for what happened. (I'm simplifying in saying so, but I'm arguing a point here.)
My hypothesis is that the phrase reflects the power relations in the ottoman domain, when the dominant and privileged ümmet-i Muhammed could impose its will arbitrarily on the subject rayahs [tax-paying class, particularly non-Muslims]. I presume that many decisions were taken only because that's what the ümmet-i Muhammed wanted, and many people were privileged only because they belonged to the ümmet-i Muhammed; that might explain Koutsonikas' [early] rendering, in which (α)μέτι μουχαμέτι meant "through the force of Muhammad". (By the way, that rendering supports my hypothesis on power relations, I believe.)
Tasos' explanation has a Greek-speaker take an expression from one domain and context (Muslim war cries, Ottoman ruling class perspective), and apply it to a different domain, with a different connotation (Imposition of arbitrary rule > arbitrariness, Christian subject class perspective). That's metaphorical extension.
The fact that the phrase originated with the Ottoman ruling class is certainly part of why the phrase has the colouring it has. But it's a fairly abstract application of the phrase, and I don't think it's enough to explain what has happened. It's not that metaphorical change doesn't happen: the best of slang is founded in metaphor, and it's the alteration in words' context that makes slang vivid. But for a bilingual Greek to take a war cry and start using it to mean "because I said so" seems a stretch.
Particularly because there is another instance when a foreign phrase is used to mean "because I said so", and there's nothing metaphorical about it: με το άστα ντούε, "[he did it] with an asta due", which is merely the Arvanitika for "that's how I want it" (ashtu dua). (The expression is more common now in fully Greek guise, με το έτσι θέλω.)
What seems more plausible to me is a metonymic change: instead of one ingenious bilingual speaker, suddenly flipping the meaning of ümmet-i Muhammed out of context to mean "hell or high water", I think it likelier that Greeks heard ümmet-i Muhammed in its original use, in contexts that would allow them to reinterpret it (or misconstrue it) as "hell or high water". And for that to happen, you want not ingenuity, but dullness—and felicitous ambiguities. In fact, it helps metonymic change if the speakers who changed the meaning of the expression did not understand Turkish at all. The fact that ümmet-i was changed to ameti indicates that they didn't; ameti sounds more like Amhed (as Fauriel already misunderstood in his 1824 translation), or αμέτε "go on".
More on αμέτι μουχαμέτι later. αμέτι μουχαμέτι is why I'm going through all this; but I'm worked up about all this, because it was a motivating question behind my thesis. My thesis was on the Modern Greek connective που, whose primary meaning is as a relativiser, but which also has various uses as a connective and a complementiser. In Standard Modern Greek, the various senses of που are hard to gather together—the shades of meaning are particularly subtle for the complementiser; but they can be gathered under a general notion of the clause being taken as a given, or presupposed.
I don't want to get too deeply into this, but: που can mean "when, because, since, given that", but not "if, until"; you can be happy που Χ, but not hope που X; you remember που events, but remember πως facts ("I remember going; I remember that I went").
Iris Papadopoulou wrote her 1994 PhD thesis giving a metaphorical account of how που came to have this range of meanings. που is derived from Ancient Greek ὅπου "where" (which survives as Modern όπου); present-day που introduces clauses that are given. The metaphorical account is that που came to have its present range of meanings, as a metaphor from Located-In-Space to Given-In-Discourse: that "because X" or "when X" or "who did X", which provide background to understanding an event, can be expressed as a literal background for the event ("where X").
It's possible; after all, I've just used "background" in that metaphorical sense. But it struck me, again as a rather abstract metaphor to explain the spread of που. It made more sense to me that the givenness of που came along for the ride through metonymic change: since relative clauses contain given information, any reanalysis of a relative clause as a causal, or a complement, would carry the givenness along with it. The immediate cause of που having the range it does is that it was a relativiser; if the immediate cause can explain the semantic range, the ultimate etymology need not be brought in.
There is some circumstantial evidence in that the Pontic relativiser ντο, which has nothing to do with ὅπου, has a similar range of meanings. For that matter, so does Middle French que.
It turns out that Greek does have a clear metaphorical use of "where", extended from Location-In-Space to Location-In-Discourse; it is used in storytelling, to link together chunks of the story. This use is fairly restricted; the only evidence I have found is from Zante (which is where Tzartzanos' Syntax had found it from.) And that metaphor was a modern metaphor, so it involved the modern word for "where", which still has its initial vowel: όπου, οπού.
- Την αυγή πάει ο ταβερνιάρης και βλέπει την καταστροφή. Οπού αρχίνησε να θυμώνη. "At dawn, the innkeeper went and saw the damage. So he started getting angry."
- Να γένουμε, λέει, αδέρφια· πέντε εμείς και ένας εσύ έξι. Όπου λοιπόν τα συμφωνήσανε. "'Let's become brothers', he said. 'There's five of us and one of you; that makes six.' So they agreed."
In fact English has the same metaphor: whereupon. But that metaphor is rather more restricted in scope than can be claimed for που overall; and (I think) more intuitive than applying it to "I remember when".
The metonymic take on language change relies on going through several instances of the ambiguous phrase through time, and pinpointing when the change in meaning would have happened: it is not immediately obvious, because the reanalysis happens in peoples' heads, and does not alter the context. Metaphor, on the other hand, can be detected in just a single sentence, applying the notion in a novel context for the first time. When I do come to writing up the story of αμέτι μουχαμέτι, it will be from a metonymic point of view.