That nasal *aN again
I was wondering about these words in *-aN- I keep coming up with. The thing that annoys me is that if one ascribes it to a substrate one would have to posit a whole language based on words with only one vowel which occurs only in open syllables to boot. So why not include into PIE, or rather into its predecessor PPIE.
Since the PIE ablaut vowel is PPIE *a, the latter language will have had four vowels, *i, *a, *u, *aN. And the rules are, wrt stress
PPIE ´-a-|-á:-|-a-´ -> PIE ´-o-|-é-|-Ø-´
PPIE ´-i-|-í:-|-i-´ -> ´-i-|-éI-|-i-´ -> PIE ´-oI-|-éI-|-i-´
PPIE ´-u-|-ú:-|-u-´ -> ´-u-|-óU-|-u-´ -> PIE ´-oU-|-éU-|-u-´
PPIE ´-aN-|-áN-|-aN-´-> ´-aN-|-á-/-ó:#|-aN-´ ->
suppose that PPIE imported a suffix -áN# for small and unimportant things, by some the unusual suffix stress was corrected *´-aN#.
At some given time words with that suffix would have two variants *-ó: and *´-aN. Since the language already had an accusative in -m, the variants were reshaped to *-á: and *´-am and restricted to nom. and acc. respectively; voilà, we have a thematic feminine declension with *-a-.
On the other hand, some words in *´-aN# would have been incorporated into the system by having case suffixes added on directly, giving *-ó:, *-anV´-, later *-o:, *-inV- etc; voilà, n-stems.
This would explain the mix-up of feminine and n-stems in Germanic.
Something similar would explain the alternation in Latin between -o: and -am in 1sg of verbs (from *-áN and *´-aN, respectively).