For my job I have to work through an enormous amount of Journals, collecting every single article I can find that may be relevant to the Etymological dictionary of Greek. After collecting them I put them into a database and pass them on to R.S.P. Beekes. Needless to say, I see a lot of articles, not always directly related to Greek. But every now and then there's one that stands out a lot.
A couple of weeks ago I found one of those; It was an article proposing Prenasalised stops within Indo-European or some earlier stage of Indo-European. This obviously drew my attention, because it sounded absolutely ridiculous. Nevertheless, the idea's that were proposed were actually quite convincing. The only reason why I'm hesitant to reconstruct prenasalised stops now, is simply because it would double the consonant inventory, which is rather big already for a natural language; especially one which apparently only has two vowels.
The article I'm talking I found in BSL 86  p361-365 By André Martinet called "Finales nasales mobiles et prénasalisées indo-européennes". As you may deduce from the name of the article, it also discusses a n-mobile. The proof for this I did not find particularly convincing, and I will not discuss it here, if you want to know more, go read the article!
The first pre-nasalised stop Martinet proposes is the *ⁿt which would be the source for the r/n-heteroclitics.
Normally the paradigm would look something like this:
Nom-Acc sg. *uód-r 'water'
Gen sg. *ud-n-ós
This r/n alternation is obviously strange, But if we assume that this changing letter was actually the reflex of *ⁿt in two different positions you'd get a paradigm like
Nom-Acc sg. *uód-ⁿt
Gen sg. *ud-ⁿt-ós
It's more likely for a stop like *ⁿt to be come an /r/ word finally than a nasal. Another interesting thing about this theory is that it would explain away the rather odd paradigm found in Greek of this word:
Nom-Acc sg. ὕδωρ
Gen sg. ὕδατος
The Genitive is generally explained as ground for *udn- with added *t as found in the n-stems (σῶμα, σῶματος 'body' from PGr.*sōmn, *sōmntos); This would be perfectly acceptable if there was any reason for the *t to be there in the n-stems; which it isn't.
But assuming that a *ⁿt was actually retained into Proto-Greek as *nt ὕδατος would be instantly explained as *udntos. It works incredibly well.
Later then the t was actually carried over from the r/n-stems to the n-stems (Not particularly convincing either, but lack of better explanation makes it necessary).
Having just indications for the *ⁿt wouldn't really be enough though. Therefore Martinet also proposes another prenasalised stop namely *ᵐbʰ And this one would then be used to account for the odd *mus versus *bʰus distribution for the dative plural ending. One language would have the *ᵐbʰ with *m and the other with *bʰ. This is also very plausible.
An enormous problem is though; it's impossible to think that these would be the only two prenasalised stops, it would leave a massive asymmetry in the phonemic system. Besides that of course, any proof for prenasalised stops is inexistant except in the r/n-stems and the said dative plural ending. This might be an indication that it was a dinstinction already very early on in Indo-European and that these are archaic remains. It's still difficult to prove.
Nevertheless the possibility that Indo-European may have had prenasalised stops at some point should be taken into account when going about pre-Indo-European buisiness.
In Retrospect, I think this theory is too far out there to be in any way plausible. I wanted to write about this article way sooner than I did. But maybe it's good that I gave it some thought. Although it's a very nice attempt at solving some of the issues in Indo-European; There's simply not enough proof to actually make this believable.
Nevertheless I'll keep it on here, maybe other people will find a way to make the theory more plausible, or even less plausible.