The Great Proto-Indo-European Thread's Sequel

[KathTheDragon]

Actually, "salt" is easy to explain from a root *sh₂el-. The nom.sg. gives *sh₂éls > *sh₂áls > *sh₂ā́l = "*sā́l" by Szemerényi's law, while the accusative is *sh₂élm̥ > *sh₂álm̥ = "*sálm̥", and the genitive is *sh₂lés = "*səlés" which would generally be reflected as "*salés".

Going back to *o for a moment, yes Howl, you are confusing long vowels. Nothing about the evidence for *o being longer than *e forces it to be the same length as *ō - if it did, why then would *ō be reflected as long even in closed syllables? No, the evidence best supports a reconstruction of *o as half-long. And before you object to that on theoretical grounds, well, this system didn't survive very long after it was created, most branches merging half-long *o with *å < coloured *e, Indo-Iranian showing the split development of Brugmann's law (which to my knowledge can be a very early sound change), and Anatolian merging *o into *ō (before or after lenition).

To my knowledge, there is not a single case of *ō that cannot be explained through Szemerényi's law, Stang's law, or (in the case of *h₂ōwyom) Vrddhi. Similarly, many cases of *ē disappear through the same mechanisms, while the *ē of Narten stems (note that the concept of Narten roots is flawed and ultimately wrong) can be explained through old reduplication (at least in the case of Narten presents - I'm not sure if there are any truly convincing examples of Narten nouns). I'm not sure why adding *i *u to PIE's inventory is problematic for this theory, since it makes the vowel system typologically more natural. As for *ī *ū, well, "*sūs" did have a laryngeal, otherwise *suHiHnom is hard to explain. I'm not sure what's so inexplicable about its inflection - *súHs *suHés > *sūs *suwes/*swes is really straightforward.

[Tropylium]

Kümmel adduces not only Brugmann's law as evidence for *o being "stronger" than *e, but also *o becoming Proto-Tocharian *e (to merge with PIE *ē) in contrast to *e merging with *i and *u in PToch *ä, as well as certain effects in Anatolian. Specifically, in Luwian, we have *éC(C)V > aCCV, while *óC(C)V > āCV, indicating that *ó counted as a long vowel for the purposes of Čop's law, while in Hittite *ó caused lenition of at least *h₂, and maybe generally if Melchert is wrong that it didn't. Kümmel goes on to explain that this assymmetry, while not proving that it developed from an original length contrast, receives the most natural explanation if it did.

Yes, we've been over this before: there's quite a bit of evidence for *o being "stronger", but this does not necessarily imply lenght. The Tocharian case for example would be easily derivable also from *e *ɔ with a height difference. For that matter, this suggests not just *e being closer than *o, but even closer than *ē? Which seems to be going on also in Germanic (*e *ē > *e *ā in NWG, *e *ē > i ē in Gothic) and partly Anatolian (*e *ē > *e *ē, but *eh₁ > *ǣ) …

From what I read there's actually still a fair bit of disagreement over what the regular reflexes of *o *ō *a *ā are in Tocharian, which makes it a bit hard to give any overall reanalysis however. (I recently saw Kortlandt propose that all long vowels split according to syllable closure: > *e *a *o in closed syllables, > *a *o *u in open ones.) The only thing that seems clear is that *e ends up, at some point in pre-Tocharian, as a weaker vowel than the entire set *o *a *ē *ā *ō.

Nothing about the evidence for *o being longer than *e forces it to be the same length as *ō - if it did, why then would *ō be reflected as long even in closed syllables? No, the evidence best supports a reconstruction of *o as half-long.

If not for the fact that "half-lenght" is an ad hoc feature that is not evidenced anywhere: as you say, all descendants either point to a normal short vowel or to a normal long vowel. By this approach we could be also e.g. solving Grassmann's Law by suggesting that *DheDh roots should be reconstructed with "half-aspirated" initials, or with a bit more similarity, solving æ-tensing in English dialects by reconstructing Early Modern English a as half-long /aˑ/.

And before you object to that on theoretical grounds, well, this system didn't survive very long after it was created

"It was an unstable system" seems to me like a general-purpose excuse to reconstruct something that is typologically implausible.

The biggest problem is IMO not in this three-length system's later decay, or even in its synchronic phonology, it's in getting to it in the first place. I do not know of any precedents of a language where old long vowels, or worse yet just one single old long vowel, end(s) up as "half-long" in contrast to new long vowels arising via compensatory lengthening. In languages like Estonian or Proto-Germanic or Proto-Samic, overlong vowels arise by the additional compensatory etc. lengthening of the already existing long vowels, not of the short vowels.

If one really wants to hang on to an *ā as the source of *o and "Brugmannian" *ā, then I think the best option will be that proposed by Robert Woodhouse: he dates Brugmann's Law as entirely earlier than the rise of new lengthened-grade *ē and *ō, and reinterprets it as conditional shortening rather than lengthening, unrelated to unconditional *ā > *ō > *o in "Proto-European". The same might work also for Anatolian (though there is the difference that *h₃e > *e₃ > *o is lengthened in Hittite, unlike in II).

This would have other costs to it, e.g. it essentially means that Indo-Iranian is the first non-Anatolian branch to split off, and all "Satemic" or "Indo-Greek" similarities have to be either later areal features, archaisms, or unrelated.

(Tangentially, I also suspect that the aforementioned overlong vowels in Proto-Germanic could be probably dispensed with some similar reshuffling of relative chronology.)

I do still think that the thing that's "missing from people's toolboxes" is however the fact that lenght contrasts can develop from height contrasts, e.g. /ʌ ɑ/ > /ɑ ɑː/ in Australian English, *ë *a > /a aː/ in Southern Sami, /eː iː/ > /iː i/ in western Hungarian.

[WeepingElf]

Fair. This concurs with my opinion that at an early stage, PIE *e was [æ] and PIE *o was [ɑ], and the greater "strength" of the latter was a matter of openness rather than length. This difference in quality is also what one would expect from a pitch accent causing ablaut - [æ] is a "brighter" sound than [ɑ], of course, and pitch accents are IMHO more likely to change vowel qualities like this than to cause lengthening of vowels, and unaccented vowels at that.

That said, a stage where apophonic *o was phonetically longer cannot be ruled out, of course, but this length difference has to be kept separate from the "classical" PIE long vowels (either lengthened grade or lengthened by a following laryngeal). Perhaps Kath, who seems to know the current literature on PIE much better than me, knows something I am not aware of which makes the assumption of a long early PIE *o more forceful than I can see, though.

[KathTheDragon]

Eh, Kümmel's article, which I've linked already, is the only paper I know of that explicitly discusses the question. Most other references I've seen to the vowel system are just people assuming (incorrectly) that *o is the unaccented variant of *e and saying nothing more.

[Howl]

However, there are still reasons to assume *o was long at some point in the history of PIE. The non-coloring of *o by laryngeals creates an obvious parallel with Eichner's law (non coloring of *ē). But this does not apply when you assume early PIE did not have *ē or that Eichner's law is invalid. Gąsiorowski combines this with an *a/*ā ablaut in his presentation Another_long_grade_Non-canonical_ablaut_involving_PIE_%C4%81. And there is another parallel. If *o was long, then both the perfect (*o) and the -s aorist (*ē) had a long grade. So we have two past-oriented aspects with a long grade.

If you do insist on using Brugmann's law as an archaism, the best solution is probably to split the secondary long vowels. So PIE *o is then /o:/ and PIE *ō is /o:ə/. This avoids the awkward reconstruction of overlong vowels.

[KathTheDragon]

The perfect was not "past-oriented", it was a tenseless stative.

[WeepingElf]

Eh, Kümmel's article, which I've linked already, is the only paper I know of that explicitly discusses the question. Most other references I've seen to the vowel system are just people assuming (incorrectly) that *o is the unaccented variant of *e and saying nothing more.

OK. I wouldn't say it can't be - there is no real problem with that hypothesis; only that I feel that it is speculative.

[KathTheDragon]

It can't just be the unaccented counterpart of *e because it's so frequently stressed without any room for it to be analogical. E.g. *dóru ~ *dréws shows no signs of having ever had any other ablaut.

[Howl]

The perfect was not "past-oriented", it was a tenseless stative.

This is not a fact, but at most a convincing assumption. Even in the earliest Ancient Greek and Vedic Sankrit the perfect denoted a state that was the result of a PAST action.

[jal]

the perfect denoted a state that was the result of a PAST action.

Can a state ever be the result of something different than a past action? Can't see most states developing from a current or future action...


JAL

[Howl]

the perfect denoted a state that was the result of a PAST action.

Can a state ever be the result of something different than a past action? Can't see most states developing from a current or future action...

Then please tell, what action makes someone young? Some biomedical researchers want to know.

More linguistically, the perfect of PIE *weyd 'to see' was *woyd and it did not mean 'having eyesight' or 'being visible'. It meant 'having seen' and thus 'to know'. This is what I mean with past-oriented or past action.

[KathTheDragon]

Given that the default assumption in IEistics is that the stative was a pure stative, insisting otherwise seems counter-intuitive. I'll see if I can find an article discussing the stative in detail, however.

Edit: I can't find any such articles, just lots of papers with a much broader scope that state fairly categorically that the stative was just a stative. Sihler's "New Comparative Grammar of Greek and Latin" goes as far as stating categorically that the stative wasn't "(present) state resulting from previous action or experience", but that this notion is a common misconception that is unfortunately still repeated.

[Frislander]

Also given how *woyd would develop semantically in many IE languages, meaning something like "having sight of" might actually be more reasonable than "having seen", because certainly in Germanic at least, the reflexes of this verb are typically specifically used to denote knowledge of skills or facts rather than people (German wissen vs. kennen), which is definitely more of a continuing state of knowledge rather than the more perfect-like semantics of having met someone before the time of speaking; "I know him" is approximately equivalent to "I have seen him before", but "I know how to do this" is more equivalent to "I see how to do this".

[WeepingElf]

I have a question on something else, but also related to the possibility of an early stage of PIE having been an active-stative language.

I seem to remember reading somewhere, perhaps in a handbook or a linguistics encyclopedia, that several older iE languages avoided neuter transitive subjects by passivizing the clause, i. e. you didn't say "The stone smashed the pot" but "The pot was smashed by the stone"; but I can't remember where I have found that. Can anybody point me to a reference for that?

[WeepingElf]

On the vowels issue: I don't insist on pre-ablaut PIE having just one open vowel, as I have maintained so far. It may easily have had two, perhaps *æ and *ɑ, perhaps *a and *ə, though this would make the GVC more complex, and I can't say anything yet about which vowel existed where. As for Aquan (and my Hesperic conlangs), this could simply have merged the two (if it isn't just a third daughter of Early PIE besides Late PIE and Anatolian where the three PIE non-high vowels merged, Indo-Iranian-style, into one, anyway).

[nimic_vostru]

The biggest problem is IMO not in this three-length system's later decay, or even in its synchronic phonology, it's in getting to it in the first place. I do not know of any precedents of a language where old long vowels, or worse yet just one single old long vowel, end(s) up as "half-long" in contrast to new long vowels arising via compensatory lengthening. In languages like Estonian or Proto-Germanic or Proto-Samic, overlong vowels arise by the additional compensatory etc. lengthening of the already existing long vowels, not of the short vowels.

Didn't the Sanskrit phenomenon of pluti (trimoric vowels) not stem from short vowels usually, wholly bypassing normal long vowels? That's what I've been lead to believe, at least.

[Tropylium]

I don't know where the story comes from that PIE has few plain velars. But when I eyeball the data in LIV2, I get a different impression.

Look at their distribution too. E.g. the median distribution for roots beginning with clear *g- is 3 subgroups, none is found in more than 6, none have reflexes in Anatolian, and only one in Tocharian. For *ǵ it's median 4, maximum 12 (*ǵenh₁- 'to beget' and *ǵneh₃- 'to know'); for *gʷ, median still 3, maximum 11 (*gʷyeh₃- 'to live').

The only root-initial plain velar found in ≥10 subgroups is *(s)ker- 'to cut', where both zero-grade *kr- and s-mobile *sker- would be expected to trigger *ḱ > *k; the only root-final plain velars in ≥10 subgroups are *lewk- 'to be light' and *yewg- 'to join', which can be by neutralization from *-wkʷ-, *-wgʷ-.

[mèþru]

velars and uvulars IMO

[KathTheDragon]

What about cases like *ḱenk- "hang"? The plain velar isn't derivable from *ḱ due to the root contraints, and I don't see any non-ad-hoc way to derive it from *kʷ.

[Howl]

I don't know where the story comes from that PIE has few plain velars. But when I eyeball the data in LIV2, I get a different impression.

Look at their distribution too. E.g. the median distribution for roots beginning with clear *g- is 3 subgroups, none is found in more than 6, none have reflexes in Anatolian, and only one in Tocharian. For *ǵ it's median 4, maximum 12 (*ǵenh₁- 'to beget' and *ǵneh₃- 'to know'); for *gʷ, median still 3, maximum 11 (*gʷyeh₃- 'to live').

The only root-initial plain velar found in ≥10 subgroups is *(s)ker- 'to cut', where both zero-grade *kr- and s-mobile *sker- would be expected to trigger *ḱ > *k; the only root-final plain velars in ≥10 subgroups are *lewk- 'to be light' and *yewg- 'to join', which can be by neutralization from *-wkʷ-, *-wgʷ-.

For the root *kes 'to comb, to scrape' I can find reflexes in all branches except Tocharian and Armenian.
For the root *gleyH 'to smear, to stick' I can find reflexes in all branches except Tocharian. This includes Lydian kλida 'earth'.

Also, I think (s)ker 'to cut' actually comes from kʷer 'to cut' with neutralization from the s-mobile.