Grammaticalisation theory posits that there is a regular process in language of content words becoming function words and then bound morphemes.
Opponents of grammaticalisation theory (e.g. Lyle Campbell, Brian Joseph) posit that grammaticalisation theory is not particularly meaningful if there are counterexamples (degrammaticalisation), whereby function words or bound morphemes become content words. Their ultimate point is that grammaticalisation is not a discrete phenomenon, and doesn’t have explanatory power.
Proponents of grammaticalisation theory counter that the preponderance of change is in the direction of grammaticalisation, so it is still a meaningful claim. That’s the point of Kasper Geeroms’ answer. FWIW, I agree that grammaticalisation is a preponderance, although I think “rare” is an overstatement; and it is certainly a distinct process from degrammaticalisation, which needs to be made sense of.
Olli Mann’s comment would be an extreme approach, to deny it is happening. But it does happen; to up the ante and the pros and cons are the more obvious example from English, and Campbell has brought up lots of Finnish examples. My recollection is that grammaticalisation theorists tried to deny degrammaticalisation at the beginning, then just harrumphed it was an exception.
Oli Mann asks:
Do you know any striking examples of degrammaticalization?
Yes. (And my PhD thesis was done in the grammaticalisation framework, so not speaking as a hater.)
I’ll give an example in Greek, since I can speak to it; someone has written up a paper on it, but I don’t remember who.
Homeric Greek: ex and ana : two distinct prepositions/adverbs/verb prefixes (like in German).
Classical Greek: ex-ana– : verb prefix; ex and ana: prepositions. No more treatment of verb prefixes as separable morphemes (adverbs).
Early Modern Greek: xana-: verb prefix. ex and ana do not exist as separate morphemes. (Well, okh is a reflex of ex as a preposition, though it is regionally restricted.)
Modern Greek: xana– : verb prefix. AND adverb.
You could argue that xana– became an adverb by analogy with other productive prefixes, which also remained prepositions (apo, para). But they’re prepositions, not adverbs.
You could argue that this is just reanalysis. And so it is; but it’s reanalysis going the wrong way.