A reply I was writing in the comments to the previous post on laterals and palatals was getting long, and started to require tying in some related issues. So, after some further expansion, y’all’re getting a full post on approaches to reconstructing the PU “dental spirant” and “palatalized dental spirant”.
I take it as a fact that these were two distinct Proto-Uralic segments, not allophones of anything else. (Honti’s paper I was reviewing in “part one” has a full argument for the skeptic.) Beyond this clarity ends, however, and we enter a battlefield of competing proposals.
Notation
To keep track of things in a neutral manner, I shall first set up a temporary notation for these segments: *d₁ for traditional [ð] = Honti’s [ɬ], and *d₂ for trad. [ðʲ] = Honti’s [ɬʲ]. This is quite close to the notation *d, *ď that I have seen used by the more cautious followers of the traditional view. Yet I find simple number subscripts preferrable: these do not imply a palatalized/unpalatalized relationship between the two, which is not a thing that should be taken for granted. This also allows talking about cases where it is not clear which consonant occurred originally: I will use the symbol *dₓ for these.
The Contestants
In addition to the two previous reconstructions, the comment from reader David M. that prompted this post brings up a third:
I’d be interested to learn what you think of the Moscow School proposal that PU *ð, *ðʲ were [rʲ], [lʲ] (I forgot which is supposed to be which)? Unfortunately I don’t have a reference, all I’ve seen is secondary or tertiary literature that takes this for granted.
I don’t think there is any “Moscow School” that seriously proposes those values? I too have only seen the reconstruction only ever covered fairly tangentially. Muscovite scholars doing current work on Proto-Uralic (at minimum, M. Zhivlov and K. Reshetnikov in their recent “Studies in Uralic Vocalism” papers) seem to accept the *ð *ðʲ reconstruction traditionally supported in Finland. As far as I can tell, the *rʲ *lʲ proposal is instead a stray idea going back to Leiden. Frederik Kortlandt explicitly states in The Indo-Uralic verb (2001):
I reconstruct Proto-Uralic palatalized /r′/ and /l′/ instead of Sammallahti’s spirants /d/ and /d′/ because they pattern like resonants and are reflected as *r and *j in Samoyedic and as *l and *l′ in Finno-Ugric (cf. Sammallahti 1988: 485, 511f., 518, 532), cf. also the variation between Proto-Finno-Permic *śülki and Proto-Ugric *sül′ki ‘saliva’.
This is apparently the entire argument he offers on the topic… and it doesn’t help that he skips over the development of both of these consonants to *ð in West Uralic. (A sadly typical example of a poor level of rigor shown in studies on “long-range” language relationships.) But I actually find this idea deserving of a closer look anyway.
The proposals don’t end here, though. Fourth is an idea mainly explored by D. Abondolo: that *d₁ and *d₂ would be the results of a pre-Uralic split from *n and *ń, respectivly. [1] This idea results from assuming an original fortis : lenis distinction among consonants in the medial position, followed by generalization of the contrast to other positions in various ways. Several further assumptions are required to end up with the PU “fortis” or “lenis” consonants in the correct positions. TBH I find this framework close to ludicrous, but if I’m entertaining Kortlandt’s suggestion, I might as well include this too.
It is quite unclear what this idea implies for the PU values of the consonants though. A straightforward length distinction *n = [nː] versus *d₁ = [n] does not seem plausible, given how the “fortis ex-allophones” (i.e. the PU plain nasals) are found in a wide variety of positions, including word-initially. The lenis members being nasalized approximants such as [ɹ̃], [ȷ̃] or [l̃], [l̃ʲ] might be possible. These also could have been denasalized entirely by PU, leaving traditional [ð], [ðʲ] — or something else.
Fifth: A. Bomhard has suggested the values *d₁ = [tɬ], *d₂ = [tɬʲ], though without any detailed argument on the matter at all. [2] This, I think, works as a further example on how the exact identity of these consonants has been left so wide open by mainstream Uralistics that outsiders will happily offer their opinions on the matter.
Finally, a “five-and-a-halfth” proposal comes from Janhunen, who accepts *d₁ as [ð], but suggests that a pure palatal spirant [ʝ] might be a more plausible value for *d₂. [3]
Given such a wide variety of views, it should be evident that we are not bound to any of them. Even entirely new solutions might arise in a closer analysis.
Distribution
Kortlandt’s “pattern as resonants” claim will be our first step to understanding these consonants. “Pattern” here does not refer to any phonological processes (very few have been reconstructed for Proto-Uralic, yet), but to the phonotactics of these consonants.
Point one: in the CVCV(C(V)) roots of Proto-Uralic, there is distributional asymmetry. The voiceless obstruents *p *t *č *ć *k *s *š *ś all prefer the initial position (some more, some less strongly), while C₂ is usually a sonorant consonant, either of the “dental spirants”, or the “laryngeal”/”velar spirant” *x. [4] This does not necessarily translate to an aversion of initial sonorants though — *w- is particularly frequent — so this is not yet an argument for the “spirants” being liquids, as much as for them at least not being voiceless. Thus the only suggestions that take a hit here are Honti and Bomhard’s ideas of reconstructing *(t)ɬ *(t)ɬʲ. This might, at first, also sound like evidence in favor of Abondolo’s internal reconstruction — if there wasn’t actually a good number of CVNV roots! [5]
Looking slightly closer, we find that there is pretty much no reliable evidence for word-initial *d₁-. The same holds for initial *r-, as well as *x- and *ŋ-.
I mentioned *d₁äpd₁ä “spleen” before, reconstructed by Sammallahti as the ancestor of Samic *ðāpðē (as well as Mari *lep, Permic *lop, Hungarian lép, Khanty *leepətnə), but this raises some red flags: this would simultaneously be the only known PU root featuring initial *d₁, and *d₁ preceded by an obstruent. Quite the accident, when the consonant was evidently highly rare! Also even despite being fortified by Scandinavian loans, the entire lexicon of Proto-Samic (Samic being the only Uralic branch where *d₁ is not merged with a more popular consonant such as *l or *t) only contains five examples of *ð-, of which three go back to *d₂ anyway. Perhaps it would be better to assume only one original oddity here and to e.g. reconstruct PU *läpd₁ä, followed by an assimilation *l-ð → *ð-ð in Samic. Other long-range consonant assimilations such as *s-ś → *ś-ś (→ *ć-ć) seem to also have occurred in the subfamily. [6] It has even been furthermore proposed that the root might be simply *läppɜ and that a derivational *-ðä has been appended in Samic, but AFAIK no such derivational affix exists.
Initial *l-, then, is not particularly abundant, but not rare either; and though the evidence for initial *d₂- is scarce, there are two widespread roots that seem reliable: *d₂ümä “glue”, and *d₂ëmə “bird cherry”, [7] as well as a few other weaker etyma. Kortlandt’s reconstruction thus seems to hold up so far: the distribution of *d₁ resembles that of *r, while the distribution of *d₂ resembles that of *l. The traditional reconstruction also fares alright. The other proposals seem weaker.
Point two: consonant clusters. Not a lot of evidence is available here, but most of it turns out to support a reconstruction along the lines of Kortlandt’s.
- The cluster *wd₁ is well-establish’d: *käwd₁ə “rope”, *täwd₁ə “full” are widespread and fairly well-behaving roots. A possible third example is *lewd₁ä- “to find”. [8] The other instances of PU coda *w seem to generally involve a following sonorant: there are examples of *wj, *wl, *wn, *wŋ. The same holds for *j, with good examples of *jw, *jm, *jŋ being found, perhaps also *jr and *jj. Examples of glide + obstruent are however rare to nonexistent (possibly: *kowsə “spruce”?) In this sense, “diphthongal” roots resemble more the general picture of CVCV roots than that of CVCCV roots. In the latter type, medial obstruents are quite common, with *nt, *ŋk, *lk being some the most frequent clusters overall.
- The single other case where a spirant might have occurred as the 2nd part of a consonant cluster is the “spleen” root, which still is anomalous at best. This potentially works as evidence in favor of a voiceless fricative reconstruction; clusters such as *ks *pś are well-supported. But this could just as well be compared to some other roots which point to unusual clusters such as *kupla “bubble”, *śüklä “wart”.
- The cluster *d₂w is also recurring: *käd₂wä “female animal, esp. weasel” [9] is well-attested, and in addition *kod₂wa “time” (Samic, Finnic, Mari) and *pad₂wə “tinder” (Finnic, Samoyedic) can be set up. As with *wC clusters, the only other examples of *Cw involve an adjacent sonorant: *jw, *lw, *rw. And again, comparision with *j suggests that this was a general phonotactic limitation for glides: examples are known of *lj, *rj, *wj, *jj.
- Two cases of *d₂k seem reliable: *tud₂ka “tip”, *wad₂kə “river bend”. Also, *d₁k could have occurred in *śod₁ka “duck”, but to reiterate a previous suggestion of mine, *-ka here could also be a suffix added in West Uralic. — At any rate, this doesn’t tell much, since pretty much any PU consonant other than the glides *w, *j could precede *k. However, phonetically, it seems somewhat implausible for these clusters to have actually been articulated as a voiced spirant + voiceless stop.
- A spirant + nasal cluster may occur in *śedₓmä “kidney”, only reflected in Samic and Khanty. Examples are known of *l/*r + nasal; sibilant + nasal or nasal + nasal did not occur.
- Various other theoretically possible cluster types are absent, but this doesn’t help in deciding between the main hypotheses on the table: no PU clusters are known that would combine the liquids, sibilants, and spirants in any way with each other (**ll, **sl, **śś, **rd…)
Point three: morphology. Here the evidence is even thinner, particularly if compared to the rich declensional and derivational systems of Proto-Indo-European. (My impression is that Proto-Uralic seems to have been a rather more analytical language than any of its descendants.) Regardless: the spirants play no part here, being entirely restricted to word roots. This could indicate simply their overall rarity, but perhaps also them having been, at some pre-Uralic date, derived from more basic consonants or consonant clusters in some way. No argument for the manner of the “spirant”s’ articulation can be made here, but this is one further point to suggest they were rather marked consonants.
In summary, the phonotactic structure of Proto-Uralic turns out to be quite compatible with sonorant values for the “spirants”. By contrast these do not seem to show much commonality with the sibilants *s, *ć, *ś, *č, *š at all. I take this as evidence against Honti’s lateral fricatives *ɬ, *ɬʲ (or Bomhard’s affricates). It also strikes me as a point against the traditional reconstruction with dental fricatives — perhaps dental approximants [ð̞], [ð̞ʲ] would work better? Kortlandt’s idea holds up perhaps surprizingly well: the root-medial-focused general distribution and the variety of reconstructed clusters would both make perfect phonetical and phonological sense if the “spirants” were some flavor of non-standard sonorant consonants. All the other proposals have certain difficulties.
Do notice however that I have touched hardly any actual data in this post — this has been more of a synchronic analysis of Proto-Uralic. Actual comparision between the Uralic languages will be a different task yet, one which will highlight different strengths and weaknesses between the different reconstruction proposals. To be continued, then!
[1] The latest formulation of this, I think, being Abondolo, Daniel (1996): Traces of pre-proto-Uralic nasals and nasal prosodies. In: Finnisch-Ugrische Mitteilungen 18/19, pp. 9-18.
[2] Bomhard, Allan (2008-2013): Reconstructing Proto-Nostratic. Actually no more than two examples of these Uralic consonants appear among the hundreds of Nostratic etymological comparisions proposed. The first is *d₂ümä “glue” ~ Semitic *ṣ́-m-d “to join” (where *ṣ́ = [tɬʼ]) ~ PIE *gem- “to join”, the second *d₂okka- “to push” (a fairly questionable root) ~ Arabic ḍakka- “to press” (in Arabic *ṣ́ → ḍ) ~ Dravidian *dūkk- “to push”. I’d say that his reconstruction essentially does not find support from the external relationships proposed, and it should therefore be evaluated on its own terms.
[3] Tangentially mentioned in Janhunen, Juha (2007): The primary laryngeal in Uralic and beyond. In: Mémoires de la Société Finno-Ougrienne , pp. 203-227.
[4] The exact identity of *x is perhaps even less obvious than that of *d₁ and *d₂, on level with the notorious Indo-European laryngeals, but Janhunen’s paper ↑ takes some good steps towards understanding this.
[5] E.g. *kuma “turned”, *kämä “hard”, *lumə “snow”, *lämə “broth”, *enä “big”, *puna “hair”, *menə- “to go”, *panə- “to put”, *sënə “vein”, *kuńa “to blink”, *mińä “daughter-in-law”, *ëńə “tame”, *aŋa- “to open”, *jäŋə “ice”, *piŋə “tooth”, *suŋə “summer”… This number of course gets smaller if we insist that only roots with Samoyedic reflexes can be considered Proto-Uralic, which I think mainly shows how artificially limiting the data taken into consideration can distort our conclusions. — An internal reconstruction with *d₁ *d₂ related to something like *t *ć would perhaps fare better, and would go well in line with the suggestion from Janhunen that *x was originally the lenis counterpart or allophone of *k. But; cart, horse, before, etc.
[6] E.g. *sükśə → *śikśə → *ćëkćë “autumn”; *sVnśa- → *śanśa- → *ćōnćē- “to stand”. Unpalatalized initials are indicated here by e.g. Hungarian ősz, Mansi *toonć-.
[7] It may or may not be a coincidence that both of these have a medial consonant *-m-. Abondolo thinks it isn’t, but I find this data too flimsy to base conclusions on. Particularly when there are also roots such as *nimə “name” or *nüd₁ə “handle, shaft” about, for which a pre-Uralic nasal dissimilation would seem to predict **d₁imə and **d₁ünə.
[8] This is reconstructed on the basis of Finnic *lewtä- and Hungarian lel. However, these could also be independant derivatives *lew-tä-, *lew-lə-, from *lewə- “to hit”. I tend to be skeptical of PU roots supposedly only found in Finnic and Hungarian, which are surprizingly frequent given that Hungarian is the second-most lexically innovative branch of Uralic (after Samoyedic). Many examples have been later shown to be IE loans in either or both. And here too IE loan origin has also been suggested for the Finnic word. — It’s amusing how both of the more secure examples of *wd₁ have the vowel framework *ä-ə, but almost certainly this is coincidental.
[9] A root with surprizingly specific semantics, but words such as Inari Sami käđfi “female weasel”, Northern Khanty /kej/ “female fur animal”, and Mator /kejbe/ “mare” are in perfect correspondence to each other. Hungarian hölgy “dame”; (archaic) “ermine” seems to confirm the specific semantics in Samic, though its general shape is rather puzzling. PU *käd₂wä would at first glance seem to predict ˣkégy — but, the same unexpected ö and l also pop up in *wad₂kə → völgy “valley”, and *tud₂ka → archaic tölgy “breast, udder”, so there seems to have been some kind of a special umlaut development of vowels preceding older *ĺɣ. In the cluster itself, palatality appears to have been “metathesized” from the 1st member to the 2nd: *ĺɣ → *lj → lgy.
Freelance reconstruction 
…Indeed, that seems to have been half-remembered nonsense. I have since learned that Illich-Svitych accepted at least PU *-ð- and derived it from Proto-Nostratic *-d-. I think what happened is that I read Kortlandt’s paper (which I’m pretty sure I did at some point), remembered his reconstruction of [rʲ] and [lʲ], and then ascribed it to the Moscow School, because they seem to have a certain tendency to reconstruct these sounds in other protolanguages. :-]
Now that’s interesting, because – speaking of the Moscow School – *r and *ŕ [rʲ] were absent from initial position in Proto-Altaic, while *l and *ĺ [lʲ] were uncommon but present. I can only speculate whether this has phylogenetic, areal or universal reasons, but in any of these cases it supports what you’re saying.
Both of these look about equally onomatopoietic to me.
I’ve read… wait, “2008-2013”? I’ve read two volumes that are dated 2008 and don’t look like more was supposed to follow. ~:-| Anyway, in that book, Bomhard generally doesn’t spend a lot of space on explanations. He makes the intriguing proposal that PIE *h2 lowered *u and *i to *o and *e in the non-Anatolian branch, but only provides one example for each. Being the last glottalist, he needs to disagree with the standard explanation of the *b in *pib- “drink” as a reduplicated *piph3-, where the voiced h3 turned *p into *b; without even mentioning this, he postulates *piph1-, where *h1 was [ʔ], so that the cluster [pʔ] came to be interpreted as the ejective he needs! Why *h1 and not *h3? Not a word of explanation. :-/
Those are actually quite straightforward. *h1 must have been something that disappears easily (or loses phonemic status easily) and doesn’t easily modify adjacent vowels, so most likely [ʔ] and/or [h]. *h2 turned *e into *a, disappeared less easily than *h1, and its Anatolian descendants were written first with cuneiform signs for [χ], then (in Lycian) with Greek letters for velar plosives – [χ] and [ħ] are pretty much the only options.* It may well have turned into [h] in the non-Anatolian branch, though (it is one of the sources of one of the Balto-Slavic tones, pretty much the only native source of the Indo-Iranian voiceless aspirates, also contributed to the voiced ones, and apparently had the same effect on /j/ in Greek as *h1 did). *h3 is the only somewhat tricky one. It turned preceding *p into *b in at least two cases (the second is Celtic *abon- “river” from *(h2)ap-h3on-), so it’s generally accepted as voiced. It turned *e into *o, so many people have assumed it must have been labialized (Rasmussen’s [ɣʷ]) – but that doesn’t work. Not only hasn’t it merged with *w in any descendant, which is the very first thing I’d expect [ɣʷ] to do as soon as someone looks the other way, but the other labialized consonants (*kʷ, *gʷ, *gʰʷ, *w) didn’t have any effect on *e! Otherwise, someone famous (I think A. L. Sihler) wrote on a Wikipedia talk page, “the queen’s wedding would be the quoon’s wadding!” I offer the very letter O, apparently from Phoenician [ʕɒjin] < [ʕɑjin], as evidence that this assumption isn’t necessary. So, the simplest assumption seems to be that *h3 was the voiced counterpart of *h2, meaning [ʁ] or perhaps [ʕ].
* Also, proto-Nostratic *q (probably aspirated) seems to be one of its ancestors.
That reminds me of one funny word: Proto-Altaic *ĺi̯ŏ́mo-ŋa “name”. :-) (Overfraught with diacritics just to make clear that pitch and length can only be reconstructed for the diphthong, not for the other two vowels.)
On one hand, a ban on initial rhotics is not a rare thing to find. I’m not sure if this has a phonetic basis (i.e. *r- having a tendency to change to something else) or an evolutionary one (i.e. *r tending to arise by medial lenition). With *d₁ it’s probably the 2nd at least.
On the other, the existence of at least areal connections between Uralic and the various “Altaic” families somewhere in the prehistory is fairly obvious — they have far too much in common typologically. I don’t think there’s any tighter relation though. There are occasional perfect lexical matches (the best-known example might be Uralic *kälə ~ Mongolian *kele “language”), but they’re just too scarce to make up even a suggestive argument for relatedness. Loanwords, perhaps, but I don’t think this possibility has been studied…
Yes. He has released a slightly updated version earlier this year. You might have to be in contact with him to have seen it, it’s not in print.
That might depend on how you define “glottalist”. For a recent example, Martin Kümmel in The consonants and vowels of PIE (2012) ends up supporting a “voiced glottalic” reconstruction.
At this stage of research on PIE, perhaps, but there’s a lot of history, and just looking at Sanskrit or Gothic they’d be trickier to notice. (Plus with Uralic there is the lack of any sources older than the 1st half of the 2nd millennium CE.)
Too bad that the supposed *ĺ- here is apparently not actually attested anywhere, because otherwise this might make a decent explanation for some *l-initial “name” words in Uralic: Mordvinic *ĺem, Mari *lʏm. As it is we have to indeed make an ad hoc appeal to nasal dissimilation to explain these.
That might get even better: Proto-Mongolic *kele “tongue”, *kele- “say” is from Proto-Altaic *kʻi̯ăli /kʰi̯ali/, which has also yielded Proto-Turkic *Kele (no distinction between /g/ and /k/ word-initially before front vowels or something), Proto-Tungusic *xilŋü “tongue” and apparently Middle Korean kằró- “say” (where ă was /ʌ~ɐ/).
But then, nobody has proposed in a long time that Uralic and Altaic are sister-groups, just that they’re both Northern Eurasiatic (as is IE).
Oh, thanks for that link! I can’t count that, however, because Kümmel only reconstructs this for some pre-PIE stage, explicitly not for PIE itself. Proper glottalists thought that the glottalization was lost independently in different IE branches.
Indeed: there’s *j in Turkic, *d in Mongolic, and *n in Tungusic, Korean and Japonic. But that’s normal. Starostin et al. (2003: 64–65):
“Initial *ĺ- is reconstructed in a small but significant number of cases where all languages reflect *l- but Mong. has a reflex typical for *d-. In one case — *ĺi̯ŏ́mo(ŋa) — even the TM forms have not preserved traces of *l- (probably because of very early nasal assimilation *ĺi̯ŏmo(ŋa) > *ńi̯ŏmo(ŋa)), but the Mongolian reflex *d- cannot be explained in any other way.
One could reconstruct something like a voiced lateral affricate here, but we assume that this correspondence is in fact a match for the widely attested word-medial PA *-ĺ- (see below), whose initial reflex was hitherto unknown.”
Same for Hittite laman “name”. Maybe it’s repeated nasal assimilation instead of dissimilation.
It looks like that he actually leaves the question open (p. 304):
I don’t think it can be, tons of Uralic words with the sequence *lVm- (e.g. *lämä “scab”, *lämə “broth”, *lëmtɜ “lowland”, *lëmpɜ “bog”, *lomə “snow”) show no such assimilation going on anywhere. Although it is interesting to notice that *nimə is the only example preceded by *i in particular.
ETA: if we started from *n in particular, one apparent parallel does come up: *ńiGənə “lime bast” ~ Mordvinic *ĺeŋgɜ. “*G” here could have been any of *j, *x, *ŋ. If this was lost, and a nominal suffix *-ka were appended, we could predict Mo. *ńeŋgɜ. However, for this word Mari has *ni; *-n- was regularly lost near front vowels, but this is particular to Mari, so a sound law *n → *l / _iN would have to be a late Volgaic areal development. This then runs into the problem that *i was lowered to *e in Mordvinic, which no more allows any kind of a phonetically well-defined conditioning…
Thanks.
They’re still both unrounded front vowels…
Thanks for an interesting post. The issue of the “dental spirants” has bothered me for long. I’ve got no solutions to offer, though.
It is an important point that there is no a priori reason to regard *d2 the palatalized counterpart of *d1. At least from a phonetic perspective, there is nothing particularly implausible as reconstructing the latter as a voiced dental spirant *δ. Such a spirant is not really an exotic sound, even though typologically relatively rare – and all the attested reflexes can be derived from this value via phonetically perfectly natural sound change.
However, the idea of a “palatalized dental spirant” is a wholly different matter. While it is articulatorily possible to palatalize a dental spirant, I don’t know a single language that would possess such a phoneme. In general, spirant contrasts involving only the presence vs. the absence of a coarticulation (such as palatalization) seem to be near non-existent.
As regards the actual reflexes of *d1 and *d2, it is also interesting to ponder why several branches show a difference in reflexes in preconsonantal vs. intervocalic position.
Lule Saami (and some other West Saami languages):
*d(1/2) > *r /_C
*d(1/2) > *t /C_
*d(1/2) > *d /V_V
Mordvin:
*d(1/2) > l /_C
*d(1/2) > *d /V_V
Permic:
*d1 > l /_C
*d1 > Ø /V_V
Does this point to an original allophonic difference in intervocalic vs. preconsonantal position?
Then a small note on the following point:
“However, phonetically, it seems somewhat implausible for these clusters to have actually been articulated as a voiced spirant + voiceless stop.”
I don’t quite see why this would be implausible; such a cluster is completely normal in e.g. North Saami. In general a cluster onsisting of a voiced spirant followed by an unvoiced stop is no more difficult to produce than any other cluster of the type sonorant + unvoiced stop (cf. Finnish lk, rk etc.).
It’s difficult to find examples of palatalized dental consonants of any kind too. Still, this was (fleetingly) found in Old Irish, at minimum. Alongside palatalized versions of all other consonants of the language too, though, so I would not count this as a precedent.
It could. Sammallahti’s reconstruction of Proto-Samic, at least already includes this feature (*θ- *-θ vs. *-ð-).
The Mordvinic evidence for this is rather thin: only *śulgə “goldeneye” speaks for lateralization, and there is also possible counterevidence: Erzya /totka/ “rack for drying furs”, which to me seems like it could be a reflex of *tud₂ka.
Your mention that *d₁ → ∅ / V_V in Permic sounds quite interesting however — the duality is noted frequently, but this is the first time I see any conditioning given! I count 4 items this would cover: *ad₁ə → *wɔɔ “year”, *kud₁a- → *kï(j)- “to plait”, *wid₁ɜ- → *vi(j)- “to kill”, *wäd₁əm (?) → *veejəm “marrow”. Only the 2nd and the 4th could be explainable as Mari loans even in theory. — It’s curious that no more than a preceding glide appears to suffice to condition *l: *käwd₁ə → *kɔl “rope”, *täwd₁ə “full” → *dɔl “all”, *nejd₁ɜ → *nïl “girl”, and perhaps similarly *owd₁əmə → *wëën(m) “curtain”, *sowd₁ə → *sool “tree bark” (the last one only reflected in Permic + Mordv. *sud). Also *oon(m) “sleep” would have to have been shortened from *ad₁ə-ma to pre-Permic *ad₁ma before this change.
#śVd₁äm(ə) → #śVlem “heart” looks like an exception to this. Unless this, too, had a complex vowel nucleus in the 1st syllable that might be also responsible for the highly various vowel reflexes. *-iw- could work for yielding Finnic *ü, vs. rather more *i-like reflexes elsewhere… though I notice this is becoming a suspiciously high amount of the cluster *-wd₁- in here by now. — There’s also *vɔl “horse”, but the suggested western cognates meaning “cow, heifer” (Mordv. *vedraš ~ Estonian *vedis) point to *e and could be of unrelated origin (probably Indo-European loans from the root of e.g. Germanic *wiθruz “lamb”). Which would leave this + its Mari and Khanty cognates free to make up a distinct root *wälɜ “draft animal, mount”.
I do not actually know what phonetic reasons (if any!) underpin the close-to-worldwide prohibition of mixed-voicing obstruent clusters, but regardless languages usually do not contain such clusters, e.g. /sd/ or /zk/. But, also, I suppose nothing would prohibit allophonic devoicing, such as /ðk/ = [θk] (similar to the occurrence of [pt] = /bt/ in otherwise p-less languages like Arabic and Nganasan).
For the record, neither do I.
Gaelic, Russian, West Caucasian languages, and that’s pretty much it except for /x/-/xʷ/.
If by “dental” you mean “alveolar”, they’re widespread in Slavic. But not beyond – even in West Caucasian many of the theoretical possibilities are absent!
Synchronic voicing assimilation, usually regressive, is very common. The one thing I know that can override this, northern German syllable-final fortition*, yields cases opposite to your hypothetical examples: I’ve heard Sydney as [ˈzɪtniː] and Simbabwe as [zɪmˈbapvɘ] with a really remarkable [pv]. And even so, northern German has progressive voicing assimilation (in such things as /sb/ across morpheme boundaries).
* Usually described as word-final devoicing, but that misses important parts of the picture.
I don’t. There seems to be a stark difference between dentals and alveolars in their palatalization tendencies.
In languages where the coronal consonants are dental, we usually find that velars are the first consonants to palatalize and assibilate. This is frequent in pre-modern IE varieties: Pre-Indo-Iranian, Proto-Greek, Vulgar Latin, Proto-Slavic, Old English; other examples include Votic, and *g → j in Arabic might count as well. But in languages that have alveolars instead, it is these that are the first in line, as seen in Modern English, Korean, Japanese, and many Chadic languages. Mandarin works as an example of both of these, with q j x resulting from palatalization of the velars k g h and the alveolars c z s, but not of the dentals t d.
Assibilation of dentals is possible (the Finnic change *ti → *ci), but this does not seem to require a palatalized stage.
Fascinating, but I’m still not sure what exactly you mean by “dental”. Some use “dental” for what others call “laminal alveolar” or “laminal denti-alveolar”, like Romance, Slavic and not-too-northern-German /t d s/ and English /s/, while “alveolar” is restricted to apical alveolars like English and northernmost German /t d/ (and often the retracted /s/ we’ve been talking about, though Wikipedia says that one is laminal in Greek, just retracted). Others restrict “dental” to apico-dentals as found all over India (and distinct from apico-alveolars in many Dravidian and some Australian languages). Yet others seem to restrict it to [θ ð] (though you obviously don’t).
I’ve never noticed a difference in the place of articulation of Mandarin t d vs. c z s. I’ll try to listen at the next few opportunities.
In Permic *wd1 > *l seems, indeed, incontestable. However, the cluster *jd1 for *nïl ‘girl’ is a problematic reconstruction. In Saami there’s no trace of *d(1); the Proto-Saami form was clearly *niejtë (from PSaa *niejdë one would expect South Saami *nïejre and Inari Saami *nieiđâ). So we rather have to reconstruct Finno-Saamic *näjti (note that both Saami and Võro point to *ä in this word). If the comparison with Permic is correct, we rather seem to have a development *jt > *jd > *l there. The vowel correspondence would be regular, though (PU *ä > PPerm *ï before sonorants in Uralic *i-stems).
Btw, in North Saami many dialects have changed original *tk to /θk/, thus creating a phonological opposition between /θk/ and /δk/. I guess this might be typologically rare. In any case, in all dialects /δ/ in clusters such as /δk/ is pronounced fully voiced.
There seem to be more than one PU root behind these “girl, woman” words actually, and some words may also have been simply assigned to the wrong lexemes. A contrast can be found in Ob-Ugric: *näj(ə) is suggested by Mansi *nääj, Khanty *nääj “lady”, while *nej(ə) is suggested by Ms *nee, Kh *nii(ŋ) “woman”. To the former type we can add Samoyedic *nä-(tɜkkɜ), to the latter Erzya /ńi/. Samic *nisōn “lady” is probably also to be connected to either one. If we were to assign Permic to the one with *ä and (most of) Finnic to the one with *e, this would also leave little grounds to reconstruct *d₁ here. The Võro word could be explained as a contamination *neito + *näi (although in principle it could also be an inverse contamination *näito + *nei in Core Finnic + Livonian).
Also then there’s the problem that the rest of the family points to a bare root *nVjə anyway, and neither *-tV nor *-d₁V is a known nominal suffix. If reconstructing *d₁ here were abandoned, perhaps at least Finnic *neito could be instead explained as an ad hoc diminutive formed after the example of *nato “sister-in-law” (for Samic this seems to not work as well)
(Finally, there’s the isolated back-vocalic Finnic root *naa- in naaras “female” / nainen “woman”, which should perhaps be explained as a separate Indo-European loan as per Koivulehto.)
I don’t think there is a reason to resort to contamination or any other special explanation to account Võro -äi- in näio ‘girl’, because there is actually a bunch of words showing -äi- in Võro vs. -ei- in the rest of Finnic (this has been pointed out to me by Petri Kallio): the cognates of heisi, veitsi, neitsyt and seitsemän also have Võro -äi-. Hence, there obviously was a change *äi > *ei, similar to the already well-known change *ai > *ei (in heinä, reikä, etc.). In the case of seitsemän ‘seven’ there is also extra-Finnic evidence for original *ä (Saami *čiečëm and Samoyed *säjtwǝ ‘seven’).
He has very recently pointed this out to me as well, and this indeed seems to argue against contamination. It remains unobvious to me how exactly this correspondence should be explained though; against a supposed simple change *ä → *e / _j, about as many counterexamples would remain as well. A good Uralic pedigree exists at least for äijä, äimä, päivä, täi. It kind of appears as if we have two different types of *j here; an “umlauting” one vs. a “neutral” one? considering that the same situation holds for the supposed change *aj → *ëj (not *ej!), with counterexamples such as aivot, kaikua, painaa, taipua/taittaa, vaimo.
But it does not stop there: Võro also has käütämä “to fasten” (cf. *kewci “rope”) which appears to be a similar retention from PU *käwd₁ə — except now before *w! And yet we also have the perfect rhyme *täwd₁ə → *täwci “full” with no raising. Very puzzling.
Incidentally, concerning the supposed distinction of the roots in Ob-Ugric: the correspondence Ms *ää ~ Kh *ää is largely irregular in origin, as in the best-quality inherited vocabulary we usually find Ms *ää ~ Kh *ee and Ms *ee ~ Kh *ää instead (which, granted, seems phonetically paradoxical at first glance, but the latter should probably be explained by an ablaut from *ii to *ää in Khanty). A possible alternate hypothesis might be that there are only two inherited words in this bundle, which developed distinct meanings in proto-Mansi and proto-Khanty, followed both of these being loaned from one to the other?
The changes *ai > *ei (*ëi) and *äi > *ei are indeed puzzling because they only seem to have affected certain words and not others, without any obvious conditioning factors. But at least in principle this could just be an instance of irregular change, which are a fact of life after all – there are many other examples of tendentially conditioned changes that only affected a limited number of words (e.g., the change *o > *ë adjacent to labials in some Saami languages). But if so, then “Gulf of Finland Finnic” must have been a rather uniform language form at the time when they took place, because otherwise we would not expect an irregular change to have affected exactly the same words everywhere. The development *äi > *ei in Livonian is not necessarily related and could have occurred independently.
The case of Võro käüttä- is intriguing, but I’m not sure it’s related to the change *äi > *ei.
The Khanty ~ Mansi correspondences look paradoxical in Honti’s reconstruction, but I think this reconstruction is just wrong here. I don’t really see a good alternative to Helimski and Zhivlov’s solution according to which Proto-Khanty quantity was preserved in Surgut and not in Vakh-Vasjugan. Then PKh *ä must be reconstructed instead of Honti’s *ee/*öö (and likewise, PKh *a must be reconstructed instead of Honti’s *oo). The main argument is that Proto-Khanty ablaut makes no sense in Honti’s reconstruction, but behaves in a uniform way in Zhivlov’s reconstruction: PKh *ä (= Honti *ee/*öö) ablauts to PKh *e in the U-grade and *i in the I-grade. Honti’s system does not include *e, but this must be reconstructed to cases where East Khanty shows reflexes of *ä (= Honti’s *ee/*öö) but Irtysh and North Khanty show reflexes of *i. (Btw, this solution also makes the distinction between Honti’s *ee and *öö needless; the development is PKh *ä > VVj ö adjacent to velars and e elsewhere, whereas PKh *e always gave VVj e and never ö.) In this way we get an ablaut pattern *ä ~ *e (U-grade) ~ *i (I-grade), which is completely analogous to *ää ~ *ee (U-grade) ~ *ii (I-grade).
In this solution the front and back vowels also behave in an identical way with regard to ablaut: in Zhivlov’s reconstruction we have *a ~ *o (U-grade) ~ *ï (I-grade), which completely analogous to *ä ~ *e (U-grade) ~ *i (I-grade).
Regarding Khanty *ää, in turn, I don’t think it can be explaned as ablauted from *ii. The relation goes the other way round – *ii is ablauted from *ää and represents the I-grade (the corresponding U-grade is *ee; Honti fails to make a distinction between *ii and *ee).
…If it depends on the next vowel, it’s umlaut, not ablaut.
True – historically Khanty ablaut arose from umlaut, but the conditioning vowels in the next syllable later mostly merged into schwa or got lost in much the same way as in Germanic. This was postulated by Helimski in his paper “Ablaut als Umlaut im Ostjakischen” in 2001; the paper can be found here:
Click to access 2.189.pdf
Thanks, I finally read the paper (as far as the mojibake allows it). :-)
Upon closer checking, the Livonian change seems indeed likely to be unrelated; *äj develops to Livonian /ei/ also in those cases that remain in Core Finnic, while usually mid vowels in this position (sonorant-closed syllables) difthongize to Livonian /ie ~ je/ and /uo ~ wo/: e.g. kuoi “moth”, kieuž “rope”.
I’m not sure if reconstructing PKh *ä and *a in place of Honti’s *ee and *oo actually makes things phonetically easier: the problem in this model becomes how do these get to /ee/ and /oo/ in the dialects other than Surgut without clashing into *ää and *aa? Zhivlov’s explanation that *ä → *e → *ee seems difficult to motivate (or to tie in with the labialization to *öö), and appears to require that the lowering of the close vowels *i *ï *ü *u to *ä *a *ö *o would have occurred separately in all the Khanty dialects. Since one of the major sources of PKh *aa is PU *ë, and at least this vowel could in principle have still been at *ëë during this stage of development, but no similar option seems to be possible among the front vowels, nor for PKh *aa where appearing to derive from original PU *a or *o. Assuming that the A-grades are not original but rather secondarily lowered would seem to work.
Zhivlov’s reconstruction of the original ablaut system of course makes good sense, but I don’t think it can be ruled out that it represents a pre-Proto-Khanty stage. For that matter, I’d need to study the language further to form an opinion on if the “U-grade” can be considered completely separate from the I-grade, or if, as Honti thinks, it is explainable as a secondary split from it. One possibility I suspect in particular is that Zhivlov’s *e and *o might represent loanwords that have diffused between the Khanty dialects at a later date: these words seem to disproportionally often lack cognates beyond Mansi, if even there.
It seems rather clear that the U-grade must be systematically distinguished from the I-grade. Even East Khanty retains the two separate in the case of one vowel *aa ~ *oo (U-grade) ~ *ïï (I-grade). In other cases the I- and U-grades have merged in East Khanty, but they’re quite systematically preserved distinct in South and North Khanty.
It seems very difficult not to regard the A-grade as primary. For the I-grade and U-grade we can postulate triggers, which in many cases seem to be preserved: long vowels in the second syllable can be reconstructed as umlaut triggers to Proto-Khanty, whereas short vowels seem to have invariably merged into *ǝ / zero. In contrast to the I- and U-grades, the A-grade seems to be triggered by nothing.
The U-grade is in most cases triggered by a long vowel in the second-syllable, but because long vowels also act as triggers of the I-grade, there must have been a merger of two types of triggers after the umlaut first arose. Perhaps the U-grade was triggered by a long labial vowel, and afterwards labial vowels in unstressed syllables were illabialized.
There are a few cases where Zhivlov’s *e (the U-grade of *ä) appears in a word with a Uralic etymology. The Uralic antecendent has PU *ä or *i, both of which normally yield PKh *ä, which then surfaces as *e in the U-grade:
PU *käsä ‘dew’ > *kelää
PU *jäkšV- ‘cold’ > *jeglii
PU *käri- ‘bind, wrap’ > *kerää ‘bunch’
PU *särä ‘fibrous object’ > *Lerääj ‘thin root’ (cf. underived *Lär ‘root’)
PU *pistä- ‘sting’ > *pestii ‘sharp’ (note that the following *t blocked the change *s > *L).
I could agree on considering it the default grade, but it might regardless have a phonetic trigger: the original Proto-Uralic stem vowel. In plenty of cases of PKh *ii ~ *ee ~ *ää, *ää seems to occur in original *ä-stems, while *ii ~ *ee occurs in original *ə-stems:
PU *kenśä- → PKh *käänć- “to become thin”
EU *kettä → PKh *käät “2”
PU *ńälmä → PKh *ńääləm “tongue”
PU *śepä → PKh *sääpəL “neck”
PU #śänä (?) → PKh *säänəɣ “mushroom”
PU *tärmä (?) → PKh *täärəm “strong” (probably parallel Indo-Iranian loans, cf. Sanskrit dharma)
PU *pišä- → PKh *pääL- “to cook”
—but:
PU *ed₁ə (*ed₁əw, *ed₁u?) → PKh *eeL “front, fore”
PU *jeśə → PKh *iis “soul”
PU *läktə- → PKh *liiɣət- (EKh /lüüɣət-/) “to leave”
PU *lewlə → PKh *liiL “soul, breath”
PU *peksə → PKh *piiɣəL “rope”
PU *mäkə (?) → PKh *miiɣ (EKh *müüɣ) “hill”
PU *näkə- → PKh *nii- “to see”
PU *ńärə “nose” → PKh *niir “promontory”
PU *śäkśə (?) → PKh *siiɣəs “gull”
PU *sewə- → PKh *ɬee- “to eat”
Also a candidate for an original *o-stem might be PU *werəw/*wero → WKh *weer “debt”.
Some clear counterexamples can be found too, though the overall pattern seems clear:
PU *jelmä → PKh *jeeləm “world, weather”
PU *kerə → PKh *käär “bark”
PU *näjə → PKh *nääj “lady”
PU *wänśə- → PKh *wäänć- “to cut”
The development would then be similar to that in Samic, where inherited *ä-ə and *a become *ie and *uo, but *e-a and *o-a become *ea and *oa.
Interesting. Are there further examples of this though? Examples of PU *st seem very rate to come by in the first place.
Regarding PKh *pestii ‘sharp’ < *pis-tä-, I should have added that the underived root is preserved in PKh *päl- 'sting, prick, stab, cut'. This is compared to PMs *piil'- by Honti, but apparently erroneously: both the vowel and the consonant correspondence is irregular.
I don't know other cases of Uralic *st preserved in Khanty. Nevertheless, the sound correspondence is paralleled by PKh *Lïstə- 'soak', apparently a phonologically obscured causative of *Lïl-aa- ‘get damp’. And then there's PKh *wïstV 'green', which looks like some kind of derivative of *wiša 'green'.
That reminds me.
German Leim “white wood glue”, Lehm “loam” – could be cognate with each other if the h in Lehm is not etymological and if Lehm is from a northern or central dialect; English oa usually corresponds to German ei, and loam is sticky. But wait. What about slime? The s- is mobile, as shown by Latin limus “mud”… and now I’m completely confused. :-)
While a change from [tɬ] to [k] is fairly commonplace (a bunch of languages in the Caucasus has done it), Bomhard’s correspondence of a Semitic ejective to an IE voiced plosive requires the glottalic theory. The Moscow School instead correlates Semitic/Afroasiatic/Nostratic ejectives and IE voiceless plosives with great success.