The Great Proto-Indo-European Thread's Sequel

[Vijay]

This talk of satemization has me wondering: Is anyone familiar with a language called Bangani, spoken in Garhwal up in the Himalayas?

See Van Driem 1996: in which the alleged centum lexicon in Bangani is analyzed to just be a result of semantic or formal misinterpretation: the words either don't mean what Zoller claims they do, or he did not give accurate reports of their phonetic structure (often both problems exist).

Which Zoller then responded to here, so it's a controversial issue and each is basically accusing the other of lying and misinformation.

[zompist]

Well pronunciation of Sanskrit is well preserved due to folks like Pāṇini and religious tradition.

I'm pretty sure that when the same argument would be used for Latin, its pronunciation would be reconstructed quite differently than we currently do.

We don't have anything comparable to Pāṇini for Latin or Greek. Or really any European language until the 1800s. Pāṇini had an understanding of phonetics that was simply not present in Europe; the Romans and Greeks didn't have the vocabulary to be precise about how their languages were pronounced, even things like voicing, aspiration, and place of articulation.

That said, I'm skeptical that the oral tradition was much better for Sanskrit than for Latin. Were the brahmins even reading Pāṇini?

(Reading Vijay's story: I'd be really surprised if the final -a in yajña was not spoken 2500 years ago. The whole point of sandhi was to notate things like that...)

[Howl]

I'm more inclined to follow a theory where the palatovelars represent an innovation of the satem languages, though clearly relatively early, and probably any pre-satem dialect(s) may have co-existed during the time of PIE itself, but I am not convinced that the system currently reconstructed in a single PIE dialect for long.

The three sets of dorsals in PIE have separate reflexes before front vowels in Albanian (the, ke, se) and Luwian (ze, ke, kue). So I have no doubt whatsoever that (1) this system is correct and (2) this system existed for a long time in PIE.

Actually, Latin pronunciation was written about, but there wasn't much of an effort to stick to the original pronunciations as Romance dialects evolved. Hindus, on the other hand, felt that it was important to preserve the original pronunciation instead of reading texts with the same pronunciation as the local languages.

The Hindus probably kept an old pronunciation at the time when there was no script. However after a script became available there was not much need to learn the verses by memory and the old pronunciation would also fade, just like it did for Latin and Ancient Greek.

[KathTheDragon]

So I read this article a few days ago on the topic of ergativity in PIE, in particular the application of typology to constrain the possible reconstructions. It concludes that PIE showed a split alignment (but not necessarily traditional split-ergative), showing accusative marking in at least personal pronouns, and most likely all animate (common gender) nouns and pronouns as well, while all other nominals showed most likely "neutral" alignment (i.e. all of S, A, and O marked the same), with genuine ergative alignment being improbable. A 2-1 system for PIE does best explain all the reflexes and modifications, including the development of ergativity in Hittite.

The major consequence of this analysis is that the account of the sigmatic nominative being a generalisation from the thematic nominative *-os, itself a development of the "ergative" *-os, which otherwise gave the genitive, is virtually invalidated (only virtually since the reconstruction with all nouns showing ergative alignment cannot be completely ruled out, even though it is the least satisfactory reconstruction).

[mae]

-

[Nortaneous]

Probably diffused from what became Balto-Slavic - cf. American English ae-breaking. (Sound changes tend to lose environmental complexity when they diffuse, and BSl seems to have the most complex conditioning.) Another possibility is widespread interdialectal loaning, of course...

[hwhatting]

[The Hindus probably kept an old pronunciation at the time when there was no script. However after a script became available there was not much need to learn the verses by memory and the old pronunciation would also fade, just like it did for Latin and Ancient Greek.

Actually, oral transmission of the Rigveda continued parallel to the written transmission. Wrong pronunciation, like all errors in the ritual, was supposed to have dire consequences, so Brahmins learnt to chant the hymns exactly like their predecessors. I've seen this mechanism being called (half-jokingly) "the Vedic tape recorder" by Western scholars of Sanscrit. So the situation is exceptional and not at all comparable to the tradition of other sacred languages.

[Tropylium]

You conveniently neglect the root constraint against identical consonants, according to which *ḱenḱ- is impossible.

This "constraint" already has a dozen or so counterexamples (LIV has *dʰewdʰ-, ¹*h₁reh₁-, ²*h₁reh₁-, *h₂elh₂-, *h₂reh₂g-, *h₂sews-, *prep-, *ses-, *skek- plus, if you count neo-roots extracted from reduplicated perfects, *h₁eh₁s-, *tetḱ-) — it's only a statistical tendency, not any kind of a real phonological rule.

All fine-tuned root constraints probably should be ignored anyway when dealing with actual reconstruction, to avoid circular logic. (Good robust ones like "PIE verb roots are monosyllabic" would be a different matter, though.)

So let's look at your statistics:
…
And the rest, like the maximum number of 'sub-groups', are outliers. And data gaps do matter for those.

Discounting everything except the median is an obviously superficial way to deal with statistical analysis.

I do have to reiterate that data gaps do not introduce any major distortions as long as a sample is representative, e.g. "just the verbs". This is elementary scientific practice: you cannot expect to have "full" datasets and must be able to work with only samples.

But then there is still Albanian kezë 'woman’s head-dress, bonnet, hair-net' which is directly cognate to Balto-Slavic *kasā́ˀ 'braided hair'

What's a "direct cognate"? And why would it be one? To me it seems you're drawing unjustified conclusions a bit too quick here.

Both words seem to come from PIE *koseh₂ 'hairdo' [A concise historical grammar of the Albanian language, page 61]. So it is not even a form that is restricted to one branch. Which is what I guess Tropylium means with 'one-off nominalizations'.

Yes. I mentioned the 'hair' > 'ox' case only as an example of an unreliable one-off nominalization. Obviously it is not intended as conclusive proof that this root would be absent in Albanian entirely.

[Tropylium]

I'm more inclined to follow a theory where the palatovelars represent an innovation of the satem languages, though clearly relatively early, and probably any pre-satem dialect(s) may have co-existed during the time of PIE itself, but I am not convinced that the system currently reconstructed in a single PIE dialect for long.

The three sets of dorsals in PIE have separate reflexes before front vowels in Albanian (the, ke, se) (…)

How many examples of *Ke > /Ke/ (or /Kje/ etc.) in Albanian are there really? I looked over Orel's overview and interestingly all of his examples of the retention of plain velars occur before *a, *o or consonants (not even before *u). Wikipedia has just one example before *e (*gʰed-n-ye- > gjej), plus one root-final example (*h₁rewg- > regj) for which they give a thematic *-e-, but where I would not be sure that that particular formation is what is being continued.

[KathTheDragon]

You conveniently neglect the root constraint against identical consonants, according to which *ḱenḱ- is impossible.

This "constraint" already has a dozen or so counterexamples (LIV has *dʰewdʰ-, ¹*h₁reh₁-, ²*h₁reh₁-, *h₂elh₂-, *h₂reh₂g-, *h₂sews-, *prep-, *ses-, *skek- plus, if you count neo-roots extracted from reduplicated perfects, *h₁eh₁s-, *tetḱ-) — it's only a statistical tendency, not any kind of a real phonological rule.

We can immediately remove *h₂reh₂g- from your list, as LIV rather has *h₂reh₁g-. *skek- isn't probative as we already know that *sK neutralises the velar, and we can't know whether this is relevant to the alleged constraint. As for *dʰewdʰ-, the Greek verb is apparently an invention of grammarians, and the only other reflexes are nominal, so this root isn't relevant either (afaik root constraints are only strong for verbal roots - nominal roots show many many more exceptions). As for *prep-, the first stop is only *p if we connect OIr. richt - we could alternatively connect the family of Lat. corpus, and reconstruct *kʷrep-. The remaining five are genuine exceptions, but it's now highly significant that they all contain fricatives. Make of this what you will, but I don't think this is a good enough argument to go ignoring the thousand other roots that do not show repeated consonants.

[Tropylium]

Well yeah, I already agreed that my 2nd suggestion seems more likely, I'm really only protesting you claiming that *ḱ…ḱ would be outright impossible (even in pre-PIE?)

[Nortaneous]

I'm more inclined to follow a theory where the palatovelars represent an innovation of the satem languages, though clearly relatively early, and probably any pre-satem dialect(s) may have co-existed during the time of PIE itself, but I am not convinced that the system currently reconstructed in a single PIE dialect for long.

The three sets of dorsals in PIE have separate reflexes before front vowels in Albanian (the, ke, se) (…)

How many examples of *Ke > /Ke/ (or /Kje/ etc.) in Albanian are there really? I looked over Orel's overview and interestingly all of his examples of the retention of plain velars occur before *a, *o or consonants (not even before *u). Wikipedia has just one example before *e (*gʰed-n-ye- > gjej), plus one root-final example (*h₁rewg- > regj) for which they give a thematic *-e-, but where I would not be sure that that particular formation is what is being continued.

Wiktionary has:
- qere 'baldness' < *keres- ~ Middle Irish carrach 'mangy'
- qerthull 'winch, circle' < *ker-ḱ- ~ Lat. circus
- qall 'mock' < *key-h2- (but the regular outcome of *ey is /i/)
- qel 'bring' < *kel ~ Gk. κέλλω, Lat. celer
- qikël 'point, spike' ~ Lith. kekulas
- qirem 'hoarsen' < PIE *(s)Ker- ~ Lat. crocio
- qit 'hit, throw' < PIE *keyd- ~ Eng. hit (but this root is usually reconstructed as *kh2eyd)
- gjej 'find' < *ghed-n-ye/o- ~ Eng. get
- gjer 'dormouse' < *geli- ~ Gk. γαλέη, Lat. glīs (these correspondences don't look regular - maybe instead from a substrate?)
And probably a few more. I stopped looking.

[Tropylium]

We discussed the possibility of a *h₂ʷ in PIE a few times in the old thread. I now have a summary write-up of this on my blog, including some new typological arguments.

I dropped a line to the Suter guy too; he reports that his paper was rejected by several dedicated IEist journals and that reviewers were generally very negative at his attempt to approach laryngeal theory by comparative evidence rather than the usual heavy-duty internal reconstruction.

[KathTheDragon]

That's... unnerving.

[Howl]

Suter's ideas about IE zero-grade in Hittite are also very interesting. His ideas maybe somewhat out of the mainstream and in need of more work but he definitely deserves to be heard.

[WeepingElf]

I concur with the idea that PIE *h₂ and *h₃ were uvulars, the paper mentioned by Tropylium in his blog post is interesting and convincing; but I am not yet sold to the *h₂^w thing. Also, I still feel as if *h₃ was a labialized *h₂. But that's just what I think ATM.

[KathTheDragon]

Surprising nobody, I'm very on board with *h₂ʷ.

[WeepingElf]

Hmmm ... what does tell you that this is not just *h₂w? And even if it is a unitary phoneme, what then is *h₃, which for most purpose behaves like a labialized *h₂?

[KathTheDragon]

I believe Tropylium has summarised the evidence for preferring *h₂ʷ to *h₂w quite plainly - go read his blog post. As for *h₃, I (yet again) contest that we can pin it down to being labialised *h₂. Your only argument is that its colouring product either merges with *a or ends up rounded, against my argument that it cannot have been labialised since Anatolian uniformly reflects it as *ḫ when it is reflected as a fricative at all. Moreover, as argued by Trop in another blog post, Anatolian points to something more like *ɐ *ɑ than *a *o, which supports my reconstruction of *h₂ *h₃ as velar and uvular respectively (damn this board's font for not distinguishing x and χ)

[WeepingElf]

I don't insist on *h3 being a labialized *h2. It hasn't escaped me that few scholars reconstruct it that way. My main reason to assume that *h3 was a labialized *h2 was the Greek triple reflex, which clearly shows that in that language, the o-colouring effect of *h3 was still productive at a time when PIE *o almost certainly already was rounded, but it may just be that whatever *h3 was, it was associated with rounding, or the process that added rounding to *o in Late PIE also added rounding to *h3. One possibility was that *h3 was pharyngeal (while *h2 was uvular), and the feature [+pharyngeal] changed into [+rounded], moving *h3 from [ħ] to [χʷ] and *o from [ɑ] to [ɒ] in one sweep. But how much phonetic sense does that make?