Word final *n in indo-european n-stems
*n stems in PIE traditionally end in *-ōn in the nominative which comes from an earlier *on-s (Known as Szemerenyi's law). While this is a very common reconstruction, much of the reconstruction is dependant on one language, Greek.
Greek n-stems indeed end in -ων, but I have my doubts whether that n isn't just an analogy from the other cases (for example the genitive is -ονος). If we look at Latin, Sanskrit and Gothic you would certainly get this impression.
For example Sanskrit rājan- 'king' loses its word final n in the nominative as rājā राजा. Also the Latin word for man homō, hominis clearly shows an absence of an n.
Lastly there's Gothic who has quite a few n stems. but the most interesting is guma 𐌲𐌿𐌼𐌰, mostly because it's cognate to Latin homō. Also guma is an n stem, its genitive is gumins and the accusative is guman.
With so much evidence that the *n disappears. Must we say that this is an PIE process? Or just a very common thing to happen in individual languages?
It is not usual for Sanskrit to lose the word final n if it is not in front of a long vowel, for Latin, I have no idea, Gothic, sadly, takes pleasure in taking off the last letter of almost any word.
It's actually quite difficult to determine when this shift then took place. But I prefer to go for the 'easier' solution, that it happened in one language, and then split into several languages retaining this development.
Assuming an isogloss after the split-up is also all but preferable. We like to see isoglosses between Greek and Sanskrit, not Sanskrit Latin and Gothic.
Another theory, which I'm not at all fond off, but that likes to pop up in articles every now and then is the 'sandhi' hypothesis.
This hypothesis claims that words ending in resonants usually had two variants, one in the form that it was followed by a vowel, and the other in the form when it was followed by a consonant. This for exmaple tries to explain the difference between the thematic Primary and Secondary endings.
*-om is the secondary ending and (according to these theorists) *-ō is the primary ending.
*ō would be the result of a pre-consonantal variant of the parent form *om, losing the m, with compensatory lengthening.
While *-om would be the pre-vocalic variant of the form *om, retaining the m, and of course, not undergoing compensatory lengething.
From there onwards, people can conclude that PIE was a SVO language, because a sentence final V would never have Sandhi variation (since nothing follows, usually). And though this idea is somewhat elegant (it allows to explain the medio-passive 3pl ending variation of *-ēr *-er *-r) I prefer to keep it simpler and going for the reconstruction of *-oh₂ for the primary ending rather than *ō.
This Sandhi idea could then also be stretched to the n-stems to explain why some languages retain it, and others don't.
Nevertheless, in my opinion, there's more readily available less exotic explanations for all these variations which have my preference.
So to conclude, my suspicion is that the disappearance of a word final *n before long vowels did indeed occur before the Indo-European languages split up, and where this *n does appear in one of its daughter languages, this is due to analogy from the other case forms.
Comments
If word-final *-n had disappeared at the PIE stage, how would Hittite laman "name" have ever arisen?
Okay, I guess I still have to wonder how you deal with *ḱwōn "dog (nom.sg.)." And then how do you deal with h₂eḱmōn "stone"? Ultimately, I'd think that regardless of anything, there exists a stage of IE or Pre-IE where *n would be present in the nominative singular since it's clearly present elsewhere. Denasalization is a very common sound change and can easily occur in two different branches seperately.
As for the 1ps personal ending, I've come to accept that *-ō < *-oh₂) is borrowed from the perfective endings via the old subjunctive (which later turned into the famous thematic present as per Jasanoff's views). I mentioned how this might have come about semantically on my blog: Thoughts on the early Indo-European subjunctive 1ps ending.
I absolutely agree. But I thought it'd be entertaining to show some of those Sandhi denasalisation theories that you run into every now and then. There's some elegance to the solution, it's sadly just not the ideal solution ;)
As for *ḱuōn, In the Germanic branch this is of course not relevant, we find hunds in Gothic. Which is a derivation of one of the other root vocalism + d suffix.
In Sanskrit the nominative singular is exactly as you'd expect: śvā श्वा
And Latin has canis, which is an i-stem derrivation from an oblique case stem I guess (Really should teach myself Latin at some point).
Of course at some point, the *n must have been there.
*ḱuon-s > *ḱuōn > *ḱuō.
Whether that last shift occured in the seperate branches or very Late Indo-European is indeed hard to tell. Luckily for the greater scheme of things, it's not one of the more important issues. For reconstructions further back than late IE, we're going to need the *on-s < *ána-sa reconstruction anyway, with or without that n being there in a later stage.
But, I'm fairly confident that hara-s simply represents an Anatolian loss of *n. At least I always assumed that was what was going on, seeing as the acc. pl. has -us, rather than -uns or something like that.
But the u found in the acc. pl. is already problematic, so maybe it's not the best of examples. Most other examples of this 'law' PIE *Vns > Hitt. VS are more difficult to find. But Kloekhorst mentions one in his fantastic Hittite Etymological Dictionary:
*ḱuéns-ti > Hitt. /kuáStˢi/, ku-ṷa-aš-zi ‘he kisses’
Which I think is pretty convincing.
I'd say that the word final *n may have disappeared before all languages split up, except for Hittite, which split up quite a bit earlier. There's been enough time afterwards for the rest of the PIE languages to have common developments to which Hittite was excluded.
After all, after Hittite split of, PIE had the chance to completely reconfigure its verbal system, so why not get rid of a word final n?
Realise that Szemerenyi's law had not yet taken effect in Hittite. Think of the Hittite *nt-stems that end in -an-za /ant-s/ for example.
There's no contradiction then right? When you place the disappearance of word final *n in between the split of Hittite and the rest of PIE, and the split of all other IE languages.
Phoenix: "I'd say that the word final *n may have disappeared before all languages split up, except for Hittite, which split up quite a bit earlier."
Alright, but nowdays, I can't imagine how it's worth continuing the outdated definition of "Proto-Indo-European" as default which excludes Anatolian. This is why I like to relegate post-Anatolian Indo-European to the terms IE-2 or Post-IE. to make it clear that this is *not* Proto-Indo-European proper. So you're speaking of a "Post-IE stage" in my books. Let's move on...
Phoenix: "Realise that Szemerenyi's law had not yet taken effect in Hittite. Think of the Hittite *nt-stems that end in -an-za /ant-s/ for example."
I assume that your definition of Szemerenyi's law is radically different from my revisions. To me, Szemerenyi's Law refers in part to vocalic lengthening caused by Syncope and its Clipping exception at the end of the Mid IE period. Then there's the later matter of the disappearance of nominative *-s after resonants which Jens Rasmussen believes involved a former word-final phoneme *z (and I now am compelled to agree with that). However, *nt-stems have nothing to do with these two Pre-IE rules and the *-s should indeed have survived into PIE. So you must be referring to yet a third phenomenon which occurred much later in *Post-IE* (involving the lenition of word-final dental-plus-sibilant clusters?). Are you by chance conflating seperate events in this way?
If you're trying to suggest that the simplification of *-Vns > *-V:n was concurrent with the change of *-Vnts > *-V:ns in PIE/Post-IE, my objection would be that all instances of *-V:R already existed in PIE before Anatolian had seperated (see Fortson, Indo-European Language And Culture (2004), p.112 concerning kissaras "hand (nom.sg.)").
Now, it's your turn. Tag! You're it! :)
I take it your reasoning is: Since kessar proves that r-stams used to have no *-s suffix the *s must have disappeared, and have been added later. But that would mean all proof of phonetic length would've been lost.
From that point onwards, you could say all nominative /S/ that were added after resonants are later analogies.
Argh, you've done it again, that all works pretty well. And it is indeed true that keS-r is the older form.
It would imply that word final long vowels *a: and *e: did shorten back to *a and *e though.
I still see a little problem though. The common *ḗr < *ér-s ending is actually not found in hittite at all. Like keS-r's form seems to come from simply *r. Saying that this lengthening and drop of *s must have taken place from that example only seems rather difficult.
Hmm, I feel that I'm having some trouble explaining what I mean. Maybe I should start a new blog post again. I see your point, I'm not completely sure whether I agree with it yet. Stay on the lookout for another blogpost on this subject ;-)
Phoenix: "I still see a little problem though. The common *ḗr < *ér-s ending is actually not found in hittite at all."
Perhaps, but how then do you explain collective neuters like *wédōr "waters" and *kʷetwṓr "four (inan.)" considering that Szemerenyi's Law (i.e. the combined disappearance and lengthening of a word-final phoneme) seems to also affect the expected word-final collective morpheme *-h₂ in these cases? The plot thickens...