To get things rolling again: some comments on a new solution to the issue of (some of) the “secondary” e-stems in Finnic. This was presented by Ante Aikio a few weeks ago, in a seminar held by the Finno-Ugrian Society on the 100th anniversary of Uralists Aulis J. Joki and Erkki Itkonen.
The idea is fairly simple and seems to solve several issues at once. I’ll probably link his handout here once it’s available online, but for now, a small summary.
- There never was a development *o-a > Finnic *a-e. Instead, it is the Samic reflex *oa-ē in these words that is an aberrant change in stem type.
- This reanalysis provides an answer on why words of the shape *a-ə have been strangely rare compared to *a-a, *ë-ə and *ë-a in previous reconstructions of Proto-Uralic.
- Mordvinic shows /u/ in words of this type, which, too, has been thought to suggest PU *o. No shift from *-a to *-ə is apparent here, though the only clear example against such a change seems to be *čašə > *čuž ‘barley’. (It also seems that an earlier geminate must be indeed posited in *tammə > *tumə ‘oak’.)
- Although not explicitly mentioned by Aikio: some cases of apparent development *o > *o in Mari can be now reanalyzed as falling under the already known *a > *o. I count at least *kole- ‘to die’, *komdəš ‘lid’, *šož ‘barley’, *omə ‘sleep’, *solə ‘intestine’.
- Samoyedic reflexes such as *kåə- ‘to die’, *ńåə- ‘to lick’ can also be derived better from original *ə-stems than *a-stems.
Now, this all comes at a cost though: with a special development of *a-ə in Samic and Mordvinic, but not of *ë-ə, it appears the lowering *ë > *a can no longer be considered an unifying feature of West Uralic! Indeed, the grouping is not supported by particularly many innovations at all (mainly this and *ðʲ → *ð) and its entire existence could well be questioned if this piece of evidence falls.
I think the situation can be salvaged though. An important hint is the existence of an exception class: *a remains *a before palatals. The previously known rare examples for *a-ə fall here. Aikio only mentions cases with *j and *x (!), but *aśkə- ‘to step’ can be added. (Reconstructing *aśka- when both Finnic *askel and Samoyedic *asəl point to *ə is rather suboptimal.) Given some other results in the recent years, we now know no less than four Uralic branches to feature a split development of *a before palatals: Samic, Mordvinic, Permic, and Samoyedic. (Edit, 2013-07-26: I’ve now also noticed some hints for a similar split in Mari.) The exact conditions differ somewhat (coda palatals only in Smy; it also appears *ń does not count in Samic) but the overall pattern is clear. Also, the contesting development in all four is a merger with *o!
I think this can be best understood in terms of an original Proto-Uralic allophonic split, between illabial *[a] in some kind of palatal contexts, and labial *[å] elsewhere. *[a] probably also was the value to occur as a stem vowel: while Samoyedic has *-å and the only other languages showing explicit evidence for a particular sound value all fall in the Western group, Smy still has *-la/-ĺa → *-lä/-jä which can be best understood starting from *a. Typology considerations support this as well: the highly reduced unstressed vocalism in Proto-Uralic is rather more likely to have featured the unmarked [a] than the marked [å].
We can now still apply a single West Uralic development *ë > *a — that is, to *[a], not [å]. This will leave *å as marginally phonemic, a situation which would then have been quickly resolved: in *a-stems by an assimilation *å-a > *a-a (which may even have occurred earlier); in *ə-stems, a merger *å > *o in Samic & Mordvinic, while Finnic went with *å > *a here too.
Reconstructing *[å] also helps to understand how the reflexes of PU *a so widely fall together with those of *o: not only conditionally in Samic & Mordvinic, but also in Permic and East Uralic. *a > *o or *o > *a would not be particularly unexpected per se (both developments are e.g. widely seen across Indo-European) but what makes this rather suspicious is how *ä and *e remain apart so well, despite /a/ vs. /o/ being a much more common vowel contrast cross-linguistically than /ä/ vs. /e/. The even more marked vowel contrast /å/ vs. /o/ can be easily expected to collapse though.