I find the development of *ü in Mansi fairly strange. This vowel is supposedly retained from Proto-Uralic in Proto-Mansi (e.g. *künčə → *künš “nail”) — but would after that only have been retained in Southern Mansi (/künš/), while Core Mansi would have lowered the vowel to *ä, together with labialization of any adjacent velar consonants (e.g. Eastern Mansi /kʷäš/).
There’s nothing a priori suspicious about this scheme. The issue is that the changes in vowel height here are completely backwards compared to other developments involving close vowels in Mansi.
- Proto-Uralic *i is reflected as open *ä, not retained as close *i. (E.g. *nimə → *näm “name”; *kiwə → *käw “stone”).
- Proto-Uralic *u is mostly reflected as open *a or *o, not retained as close *u. (E.g. *tulka → *towəl “feather”.) It does seem though that in *ə-stems when adjacent to a labial consonant, *u is reflected as *u, but this looks like a later assimilation rather than a retention. (E.g. *purə (→ ? *por-) → *pur- “to bite”.)
- Proto-Mansi *i (originating from PU *e) is retained, not lowered, in Core Mansi, and lowered, not retained, in Southern Mansi. (E.g. PMs *šim “heart” → Northern Mansi /sim/, but SMs /šäm/.)
- Proto-Mansi *u is also retained, not lowered, in Core Mansi, and lowered, not retained, in Southern Mansi. (E.g. “to bite” → NMs /pur-/, but SMs /por-/.)
I’ve thus hypothetized that perhaps *ü indeed was a part of the set of “open short vowels” in Proto-Mansi. A fully open front rounded vowel [ɶ] is vanishingly rare among the world’s languages, though. To realistically maintain *ü as a part of a series with *ä, I would have to assume that these were actually a short mid vowels, *e *ö. According to this idea, both Southern Mansi and Core Mansi would have later lowered *e to *ä. However, *ö would have undergone some changes before this. It would have lost its labialization in CMs, merging into *e; and in SMs, it would have been raised back to *ü, to create a more typologically plausible short vowel inventory, triangular *i *ü *u, *e *o, *a rather than “fork-shaped” *i *u, *e *ö *o, *a.
You may have noticed a flaw. I just said three paragraphs ago that Southern Mansi does not retain *i, but lowers this as well to *ä, and surely *ü, *o, *ä *a would be a very typologically implausible vowel inventory? (Also, eastern dialects of Khanty seem to have no problem with having a short /ö/ without any accompanying short ˣ/ü/, so clearly a raising *ö → *ü is not even very strongly expected.)
Well. There in fact is an /i/ in Southern Mansi, too. Thing is, although it frequently coincides with Core Mansi /i/ (which had previously fooled me into thinking it was actually retained), it does not directly continue Proto-Mansi *i! Apparently the development rather went like this:
- Dental consonants adjacent to *i are palatalized: *kit “2” → *kiť, *pil- “to fear” → *piĺ-
- *i merges into *ä: *šim “heart” → /šäm/, *jipəɣ “owl” → /jäpəw/; *kiť, *piĺ- → *käť, *päĺ-.
- Before palatalized consonants, *ä is (re)raised to *i: *käť → /kiť/, *päĺ- → /piĺ-/.
Step 3 occurs also before original palatal consonants, as seen from *päĺ “ear” → /piĺ/.
Importantly, in addition, adjacent to *w, *ä is raised + labialized to /ü/: e.g. *käw → /küw/ “stone”; *wit → *wäť → *wüť → /üť/ “water”.
(This scheme was brought to my attention by a reconstruction of Proto-Mansi by Mikhail Zhivlov. It may or may not have been proposed before him — hard to tell when your reading skills in Russian are entirely outsourced to Google Translate… Also, thanks to Ante Aikio for providing me with a copy of the work in the first place.)
So, my new idea: let’s do away with Proto-Mansi *ü entirely, postulate only *ä at this stage, and see if we can derive Southern Mansi /ü/ purely by raising next to labiovelars. (This will, obviously, require the related labialization *k → *kʷ to have occurred already by Proto-Mansi.)
Indeed, almost all cases of *ü → SMs /ü/ (that I can get together on a quick database scan, anyway) occur next to velar consonants.
- *kü- → /kün/ “outside”
- *küčə- → /küš-/ “to smoulder”
- *küjɜ- → /küj-/ “to follow”
- *külä → /kül/ “house”
- *künčə → /künš/ “nail”
- *ńükə- → /ńüw-/ “to rip”
- *pükkɜ → /pükńii/ “navel”
- *sükśə → /tüks/ “autumn”
- *tüŋkɜ → /tüŋ/ “moss”
- *ükə → /ük/ “1”
- Ugric *jükkɜr → /jükäär/ “root”
- Ugric *künɜ → /küń-keeś/ “light”
- Ugric *ɬüɣ → /tüw/ “3PS pronoun”
- Ugric *süŋkɜ → /süŋ/ “wedge”
Better yet however: there are cases where PU *ü does not have any adjoining velar consonant, and which show SMs /ä/!
- sülə “bosom” → /täl/
- nüðə “handle” → /näl/
- wülä “over” → /äl/
These reflexes have previously been considered irregular. Predictive power: acquired. 😀
The net increase in regularity only amounts to +1 though, as two cases of /ü/ appearing without an adjacent velar also come up. Still in the positive (even before adding in the points from not having to assume against-the-current vowel developments), but still with some room for improvement…
- *śüðə → /süĺii/ “coal”
- *śülkə- → /süĺk-/ “to spit”
I wonder if both of these featuring the consonant environment s_ĺ is relevant. Was there perhaps, briefly, an *sʷ in Proto-Mansi?!
For completeness, I’ll also note some items that show *ü → *ä despite an adjacent velar.
- *kütkə- → /kät-/ “to bind”
- *šilkɜ- → /täwl-/ “to fly”
- Ugric *tüɣɜtɜ → /täwt/ “fire”
The traditional account and my new model both fare equally badly with these. Perhaps the last one should be rather reconstructed as *teɣätɜ though: both Hungarian tűz and Khanty *töɣət could be derived from this form as well.