The Great Proto-Indo-European Thread's Sequel

[WeepingElf]

Coincidental tweet by Marijn van Putten:

https://twitter.com/PhDniX/status/1622662636948054017

Plot twist: Marijn is actually reading us?

Interesting tweet and discussion. Marijn is the owner of this blog. It never occurred to me that Verner's Law could be a case of this, but indeed, the PIE accent responsible for it is usually considered a pitch accent.

I prefer explanations that draw on distinctions found in Uralic and other Mitian languages

Isn't that somewhat circular reasoning though? Because this makes internal reconstruction of pre-PIE look more like other "Mitian" languages and validate the hypothesis even more.

I understand your objection. We don't know whether the Mitian macrofamily is real or not, even if this is perhaps the best explanation for the morphological resemblances between all those languages, so we cannot draw conclusions from that.

Besides, even if Mitian is firmly established to have existed 10 000 years ago, we still don't know much about typology in the region back then. Prosodic features are gained and lost even more readily than average, too.

Yes - such features are very volatile. And we know nothing about that "(Trans)Caucasian substratum", so anything is possible - but then, substratum theories are probably overrated. In Romance historical linguistics, where they were all the rage a century ago, they have fallen out of fashion by now, mainly because we now know much more about the pre-Roman languages of western Europe, and they turned out not to match Romance isoglosses well.

[abahot]

Question -- Proto-Indo-European is reconstructed with an extensive and very complicated system of nominal ablaut in many athematic nouns, but I've also read that only faint traces of this system appear in daughter languages, and only the earliest ones at that. Why do we reconstruct such a developed system of noun ablaut, and how do we know the full extent of the patterns when only bare remnants of it remain anywhere?

[Richard W]

I prefer explanations that draw on distinctions found in Uralic and other Mitian languages

Isn't that somewhat circular reasoning though? Because this makes internal reconstruction of pre-PIE look more like other "Mitian" languages and validate the hypothesis even more.

But isn't the commonly preferred hypothesis one of areal convergence? Wouldn't that have a similar effect?

[WeepingElf]

Question -- Proto-Indo-European is reconstructed with an extensive and very complicated system of nominal ablaut in many athematic nouns, but I've also read that only faint traces of this system appear in daughter languages, and only the earliest ones at that. Why do we reconstruct such a developed system of noun ablaut, and how do we know the full extent of the patterns when only bare remnants of it remain anywhere?

I think there are more than enough traces in the older languages such as Ancient Greek (which hasn't really changed much in its vowel system) or Sanskrit (which has neutralized qualitative ablaut but preserved quantitative ablaut well) to justify this reconstruction. Historical linguistics is a highly developed discipline which has grown a lot since its inception 200-odd years ago.

I prefer explanations that draw on distinctions found in Uralic and other Mitian languages

Isn't that somewhat circular reasoning though? Because this makes internal reconstruction of pre-PIE look more like other "Mitian" languages and validate the hypothesis even more.

But isn't the commonly preferred hypothesis one of areal convergence? Wouldn't that have a similar effect?

This would mean that pronouns and inflectional endings spread from language to language in a large area - but pronouns and inflectional endings are unlikely Wanderwörter. Wanderwörter are usually words pertaining to cultural, technological, religious etc. innovations, things that spread across a large area where the languages do not have words for them, such that the words spread with the things. I wouldn't say that such a spread of grammatical morphemes across eight families in a vast area stretching from Eastern Europe all the way to Bering Strait was impossible, but that doesn't seem likely to me. And it seems as if genetics has thrown up evidence in favour of a population fanning out over roughly the same area from somewhere near Lake Baykal at the end of he last ice age, but of course, genes and languages do not always travel together, so one has to be careful here.

[hwhatting]

Question -- Proto-Indo-European is reconstructed with an extensive and very complicated system of nominal ablaut in many athematic nouns, but I've also read that only faint traces of this system appear in daughter languages, and only the earliest ones at that. Why do we reconstruct such a developed system of noun ablaut, and how do we know the full extent of the patterns when only bare remnants of it remain anywhere?

I think there are more than enough traces in the older languages such as Ancient Greek (which hasn't really changed much in its vowel system) or Sanskrit (which has neutralized qualitative ablaut but preserved quantitative ablaut well) to justify this reconstruction. Historical linguistics is a highly developed discipline which has grown a lot since its inception 200-odd years ago.

To add to this, we have far more than "traces". In the languages with the oldest record, the ablaut paradigms in Sanscrit and Greek are extensive, Anatolian also shows them, and the languages which show less ablaut tend to be the ones where the written record starts later. But even there, words with ablaut have been regularised or replaced by derivations based on different ablaut grades, e.g. in Germanic among the heteroclitics English sun (oblique stem with suffix -n- and zero grade) vs. Scandinavian sol (nominative stem with suffix -l- and full grade). So both older languages having more of the system and the variation of derivations in the younger languages point to an extensive system that was later abandoned.

[abahot]

Alright, that was helpful. Thank you, you two.

[abahot]

Could we explain the *D^h set of consonants as only having been breathy voiced because of an areal innovation in dialects along the southern edge of the PIE dialect continuum? That is, in Italic, Hellenic, Phrygian, Armenian, and Indo-Iranian.

One thing that I find rather noteworthy is that this overlaps almost perfectly with the language families where PIE *o is actually attested as being rounded (as was brought up on some thread on the old board years and years ago) -- Celtic, Italic, Hellenic, Phrygian, and Armenian.

[WeepingElf]

The academic Indo-Europeanists have very good reasons to reconstruct Proto-Indo-European the way they do. Of course, the current model is not the final word, but it is the result of more than 200 years of work by hundreds of scholars, so they can't all be barking up the wrong tree all the time.

[abahot]

The academic Indo-Europeanists have very good reasons to reconstruct Proto-Indo-European the way they do. Of course, the current model is not the final word, but it is the result of more than 200 years of work by hundreds of scholars, so they can't all be barking up the wrong tree all the time.

Isn't literally this entire thread speculating about PIE?

[WeepingElf]

The academic Indo-Europeanists have very good reasons to reconstruct Proto-Indo-European the way they do. Of course, the current model is not the final word, but it is the result of more than 200 years of work by hundreds of scholars, so they can't all be barking up the wrong tree all the time.

Isn't literally this entire thread speculating about PIE?

It is. There are still many things we don't know about PIE, especially what happened in earlier stages. But that doesn't mean we should question what scholars have already found out about it. Of course, the accepted scholarly opinion may change in the future in light of new evidence (best example: the acceptance of the laryngeal theory, which turned a mess of six ablaut series into a neat system with just one), but we should stand on earlier scholars' shoulders and not tread on their feet, or so the saying goes.

[abahot]

I'm honestly really not sure why this discussion is any different than ones that this thread has had in the past. Saying something along the lines of "laryngeals didn't exist" is ridiculous, but wondering about the realizations of reconstructed sounds is not. Especially because discussions of this nature, and even of the very same question (the realization of the *D^h series) have come up before.

[WeepingElf]

OK - I apologize for the misunderstandings.

Indeed, the phonation types of the stops are currently one of the hottest topics in IE comparative linguistics; just about everybody admits that the standard model is so for mainly historical reasons and relies more on Sanskrit than it ought to, and that it accounts for the reflexes quite well but has its problems.

Let's take stock of the facts:

1. PIE clearly had three types of stops, or consonants reflected as stops in most IE languages, traditionally called "voiceless", "voiced" and "voiced aspirated", usually written *T, *D and *Dh. Models with fewer types cannot account for the facts, and there is no good evidence for the existence of a fourth type, after the Sanskrit voiceless aspirated stops have turned out to be secondary (*T + laryngeal).

2. Of these three types, *T is clearly the most frequent, and reflected by plain voiceless stops everywhere except in Germanic (where they become voiceless fricatives, sometimes secondarily voiced by Verner's Law, except in some clusters) and Armenian (where they are normally aspirated, but in some contexts, yet various different things). They are therefore usually reconstructed as plain voiceless stops.

3. *D is the least common type, though not very much less common than *Dh, and reflected as plain voiced stops except in Germanic and Armenian which both have voiceless stops here. They are therefore usually reconstructed as plain voiced stops.

4. *Dh is clearly distinct from *D, though they merge with *D in most languages. Of the languages where they don't, Indo-Aryan (but not Iranian, which belongs to the many languages with a *D/*Dh merger) has breathy-voiced stops, also some Armenian dialects, and there is some evidence for such reflexes in Old Armenian; Greek has voiceless aspirated stops; Italic has fricatives that are voiceless initially and voiced medially; Germanic has voiced sounds which are in some contexts stops and in others fricatives. They are therefore usually reconstructed as "voiced aspirated" stops.

5. The resulting system is typologically rare; apparently, the only known language with such a system is Kelabit, an obscure Austronesian language somewhere in the jungles of Borneo. This, however, is a rather weak argument against the traditional reconstruction, because rare systems occur, and the PIE system apparently was unstable as every language has changed it (yes, even the Indo-Aryan languages did, by adding a fourth type, voiceless aspirated stops, thereby changing it into a neat 2x2 grid).

6. There was something like an assimilation rule in place which evidently affected *T and *Dh but not *D, which probably means that they once had a common feature at the exclusion of *D.

7. There was something like a dissimilation rule in place which affected the *D set, which probably means that they either had a highly marked realization (such as ejectives) or, to the contrary, were something weak and volatile (such as voiced fricatives).

While facts #1-#4 speak for the traditional reconstruction, facts #5-#7 speak against it. The question is which alternative reconstruction can make sense of all these facts with the least effort. Meanwhile, it is not necessary that #6 and #7 are satisfied in Late PIE; the assimilation and dissimilation rules may have been effective in an earlier period when the system was different from that at the time of breakup.

[bradrn]

Incidentally, thank you for writing the first actually intelligible overview of the PIE stop system that I’ve found so far.

[WeepingElf]

BPJ, a very knowledgeable and insightful member of the CONLANG mailing list, brought a proposal by James Clackson to my attention according to which

*T = plain voiceless
*D = stiff-voiced
*Dh = slack-voiced

As slack voice is (according to Wikipedia) closer to voicelessness than stiff voice, it is conceivable that *T and *Dh formed a class at the exclusion of *D, but I am not sure whether such a difference in degree has such an effect.

[bradrn]

BPJ, a very knowledgeable and insightful member of the CONLANG mailing list, brought a proposal by James Clackson to my attention according to which

*T = plain voiceless
*D = stiff-voiced
*Dh = slack-voiced

As slack voice is (according to Wikipedia) closer to voicelessness than stiff voice, it is conceivable that *T and *Dh formed a class at the exclusion of *D, but I am not sure whether such a difference in degree has such an effect.

This is basically Javanese plus an extra series. Hard to know whether any of these series would group together, but I’ve been wondering about this kind of system for a while.

[Moose-tache]

As slack voice is (according to Wikipedia) closer to voicelessness than stiff voice...

Uh... I don't know about this. Stiff and creaky voices involve glottal/laryngeal restriction, and slack and breathy voices involve glottal/laryngeal relaxation, so they tend to pattern more with voiceless and voiced, respectively. The Austroasiatic languages give examples of creaky or breathy merging with voiceless or voiced (respectively), as well as creaky or breathy emerging from voiceless or voiced (respectively), but I'm not sure about any examples of a phonation distinction and a voicing distinction lining up the other way around. Wikipedia mentions that stiff voice is often transcribed with voiced glyphs, but this is probably because IPA doesn't have neat representations of stiff and slack the way it does for creaky and breathy. I think this tells us more about transcription and less about phonology.

With that in mind, I think a better way to think about the phonation theory is this:
T: stiff/creaky
D: none
Dh: slack/breathy

Remember that roots can be T...D, D...T, T...T, Dh...D, D...Dh, and Dh...Dh, but not T...Dh, Dh...T, or D...D. This can be reworded as:
each root has one phonation: stiff/creaky or slack/breathy. It may also include consonants which are neutral (i.e. D).
This explains the prohibition against D...D roots: they have no phonation. It also explains the prohibition against T...Dh and Dh...T roots: they have too many phonations. You might say "Well, what about roots with mobile s? They can retain the form sT...Dh!" I would submit that these are not actually stiff/creaky, but simply a form of D. The change T...Dh > Dh...Dh is matched by the change T...Dh > D...Dh after mobile s, which in turn experiences secondary devoicing without taking on any phonation distinction. When the daughter languages subsequently lose the phonation distinction, these sT clusters seemlessly merge with the voiceless plosives (this also dovetails neatly with Germanic, where sT does not yield fricatives).

This yields the following developments.
A) Italic, Hellenic, Indo-Iranian: The slack/breathy distinction survives as aspiration or fricative...ization, and the stiff/creaky - neutral distinction is reworked as voiceless - voiced, pretty reasonable considering that stiff/creaky phonation restricts the ability to provide voicing to a consonant.
B) Germanic, Armenian: The stiff/creaky series becomes more defined by voicelessness, again not surprising, but this voicelessness is exaggerated as aspiration and eventually fricativization, to better distinguish this series from the neutral series, which is then free to fulfill its destiny as the most unmarked consonant imaginable: plain voiceless. The breathy distinction was then irrelevant, but in the new voicing paradigm, slack/breathy easily shifts to being defined by the feature +voiced.
C) Celtic, Balto-Slavic, Albanian: The stiff/creaky and slack/breathy series are reanalyzed as voiceless and voiced, with neutral falling into the latter. I'm not sure how Winter's Law fits into this, but then Winter's Law isn't obviously intuitive with the traditional system, either. Albanian's development of D often yields affricates, which could be the result of the neutral series being insufficiently distinct, and developing an off-glide (off-fricative?).

Here are some possible objections.
A) "But in Armenian, sT is aspirated. It doesn't develop the same way as s+D." Yeah, that's Armenian. I don't know what you want from me.
B) "But in Indic, T+H gives voiceless aspirates, while D+H does not. Why would stiff/creaky and neutral yield different results under the influence of laryngeals?" I believe that the emergence of voiceless aspirates in Indo-Aryan is a fairly late development. The ancient Indians were famously accomplished grammarians, and the T D Dh system offended them just as much as it does modern linguists.
C) "Can we circle back to Winter's Law?" Winter's Law isn't universal in Balto-Slavic, and the situation is still subject to fierce debate. One possibility is that in coda position, neutral consonants were seen as lower weight than other consonants. I don't know, but like I said I'm not sure there is an obvious explanation in other theories, either. Glottalic theory might explain the rising tone (which I'm not 100% certain is even what's going on with acutes), but it can't explain why coda ejectives would cause lengthening.

[Nortaneous]

4. *Dh is clearly distinct from *D, though they merge with *D in most languages. Of the languages where they don't, Indo-Aryan (but not Iranian, which belongs to the many languages with a *D/*Dh merger) has breathy-voiced stops, also some Armenian dialects, and there is some evidence for such reflexes in Old Armenian; Greek has voiceless aspirated stops; Italic has fricatives that are voiceless initially and voiced medially; Germanic has voiced sounds which are in some contexts stops and in others fricatives.

Complicated by Winter's Law in Balto-Slavic and Lachmann's Law in Latin (thus reconstructable for Italic), in which the *D series under certain conditions (preceding a *T-series consonant in Latin and in closed syllables[?] in Balto-Slavic) does something indistinguishable from excreting a preceding *h1. Since the Balto-Slavic acute (which arose from laryngeals and Winter's Law) is realized in Latvian as vowel glottalization, Winter's Law is often (e.g. by Kortlandt) reconstructed for Proto-Balto-Slavic as glottalization and used as evidence for *D having some glottalic character (i.e. */ˀD/ contrasting with *Dh = */D/), in addition to a number of cases of seeming *d ~ *h1 interchange (**dwi-déḱm̥(t) > *h₁wi(h₁)ḱm̥tih₁, **dḱm̥tóm > *h₁ḱm̥tóm, although there are other explanations for ἑκατόν) and the fact that, even though they can't be ejectives, they sure pattern like them. (Is *D-like patterning attested for other types of stop?)

If they were implosives, we'd expect them to become liquids somewhere (at least by SEA precedent - is this common in other regions with implosives?), but *b is missing and PIE has some typologically unusual clusters involving *m and *w... (but cf. the types of syllable structure you get in Tibetic)

5. The resulting system is typologically rare; apparently, the only known language with such a system is Kelabit, an obscure Austronesian language somewhere in the jungles of Borneo.

That's a common misconception - Kelabit does not have this! It instead has a DTh series, developed by partial devoicing of voiced geminates. (I assume these are similar to the mixed-voicing clusters in Khoisan? But in Khoisan - which as an areal grouping is IMO fairly phonologically similar to the clusterous SEA one, highly compressed and onset-heavy, tho of course the word structure is totally different - they're constrained to the left edge, not typical geminate position.)

[bradrn]

Complicated by Winter's Law in Balto-Slavic and Lachmann's Law in Latin (thus reconstructable for Italic), in which the *D series under certain conditions (preceding a *T-series consonant in Latin and in closed syllables[?] in Balto-Slavic) does something indistinguishable from excreting a preceding *h1. Since the Balto-Slavic acute (which arose from laryngeals and Winter's Law) is realized in Latvian as vowel glottalization, Winter's Law is often (e.g. by Kortlandt) reconstructed for Proto-Balto-Slavic as glottalization and used as evidence for *D having some glottalic character (i.e. */ˀD/ contrasting with *Dh = */D/), in addition to a number of cases of seeming *d ~ *h1 interchange (**dwi-déḱm̥(t) > *h₁wi(h₁)ḱm̥tih₁, **dḱm̥tóm > *h₁ḱm̥tóm, although there are other explanations for ἑκατόν) and the fact that, even though they can't be ejectives, they sure pattern like them. (Is *D-like patterning attested for other types of stop?)

Isn’t this just the glottalic theory?

If they were implosives, we'd expect them to become liquids somewhere (at least by SEA precedent - is this common in other regions with implosives?)

I don’t think I’ve ever heard of implosives becoming liquids… do you have any examples?

[WeepingElf]

As slack voice is (according to Wikipedia) closer to voicelessness than stiff voice...

Uh... I don't know about this. Stiff and creaky voices involve glottal/laryngeal restriction, and slack and breathy voices involve glottal/laryngeal relaxation, so they tend to pattern more with voiceless and voiced, respectively. The Austroasiatic languages give examples of creaky or breathy merging with voiceless or voiced (respectively), as well as creaky or breathy emerging from voiceless or voiced (respectively), but I'm not sure about any examples of a phonation distinction and a voicing distinction lining up the other way around. Wikipedia mentions that stiff voice is often transcribed with voiced glyphs, but this is probably because IPA doesn't have neat representations of stiff and slack the way it does for creaky and breathy. I think this tells us more about transcription and less about phonology.

Well, I am not an expert on this, and I haven't really understood the relevant Wikipedia pages, and the table at the bottom of those pages according to which slack voice was closer to voicelessness than stiff voice may just be wrong.

With that in mind, I think a better way to think about the phonation theory is this:
T: stiff/creaky
D: none
Dh: slack/breathy

Remember that roots can be T...D, D...T, T...T, Dh...D, D...Dh, and Dh...Dh, but not T...Dh, Dh...T, or D...D. This can be reworded as:
each root has one phonation: stiff/creaky or slack/breathy. It may also include consonants which are neutral (i.e. D).
This explains the prohibition against D...D roots: they have no phonation. It also explains the prohibition against T...Dh and Dh...T roots: they have too many phonations. You might say "Well, what about roots with mobile s? They can retain the form sT...Dh!" I would submit that these are not actually stiff/creaky, but simply a form of D. The change T...Dh > Dh...Dh is matched by the change T...Dh > D...Dh after mobile s, which in turn experiences secondary devoicing without taking on any phonation distinction. When the daughter languages subsequently lose the phonation distinction, these sT clusters seemlessly merge with the voiceless plosives (this also dovetails neatly with Germanic, where sT does not yield fricatives).

A nice and interesting theory, but your system in a way stands on its head: the least marked stops (the neutral ones) are the least common, and why then is *b so rare? Yet, I can't say that you were wrong. It is at least an interesting possibility to consider.

This yields the following developments.
A) Italic, Hellenic, Indo-Iranian: The slack/breathy distinction survives as aspiration or fricative...ization, and the stiff/creaky - neutral distinction is reworked as voiceless - voiced, pretty reasonable considering that stiff/creaky phonation restricts the ability to provide voicing to a consonant.
B) Germanic, Armenian: The stiff/creaky series becomes more defined by voicelessness, again not surprising, but this voicelessness is exaggerated as aspiration and eventually fricativization, to better distinguish this series from the neutral series, which is then free to fulfill its destiny as the most unmarked consonant imaginable: plain voiceless. The breathy distinction was then irrelevant, but in the new voicing paradigm, slack/breathy easily shifts to being defined by the feature +voiced.
C) Celtic, Balto-Slavic, Albanian: The stiff/creaky and slack/breathy series are reanalyzed as voiceless and voiced, with neutral falling into the latter. I'm not sure how Winter's Law fits into this, but then Winter's Law isn't obviously intuitive with the traditional system, either. Albanian's development of D often yields affricates, which could be the result of the neutral series being insufficiently distinct, and developing an off-glide (off-fricative?).

Here are some possible objections.
A) "But in Armenian, sT is aspirated. It doesn't develop the same way as s+D." Yeah, that's Armenian. I don't know what you want from me.
B) "But in Indic, T+H gives voiceless aspirates, while D+H does not. Why would stiff/creaky and neutral yield different results under the influence of laryngeals?" I believe that the emergence of voiceless aspirates in Indo-Aryan is a fairly late development. The ancient Indians were famously accomplished grammarians, and the T D Dh system offended them just as much as it does modern linguists.
C) "Can we circle back to Winter's Law?" Winter's Law isn't universal in Balto-Slavic, and the situation is still subject to fierce debate. One possibility is that in coda position, neutral consonants were seen as lower weight than other consonants. I don't know, but like I said I'm not sure there is an obvious explanation in other theories, either. Glottalic theory might explain the rising tone (which I'm not 100% certain is even what's going on with acutes), but it can't explain why coda ejectives would cause lengthening.

Fair. Yet, I feel that this is more complicated than my "aspiration theory" (to recall: *T is aspirated, everything else as in the standard model), and it involves phonation types that are otherwise unknown in Europe (but then we don't know what kind of "phonetic landscape" was in place in Eastern Europe 5,000 years ago!). But maybe my "aspiration theory" is problematic in other respects, and of course, I may be wrong.

[Nortaneous]

Complicated by Winter's Law in Balto-Slavic and Lachmann's Law in Latin (thus reconstructable for Italic), in which the *D series under certain conditions (preceding a *T-series consonant in Latin and in closed syllables[?] in Balto-Slavic) does something indistinguishable from excreting a preceding *h1. Since the Balto-Slavic acute (which arose from laryngeals and Winter's Law) is realized in Latvian as vowel glottalization, Winter's Law is often (e.g. by Kortlandt) reconstructed for Proto-Balto-Slavic as glottalization and used as evidence for *D having some glottalic character (i.e. */ˀD/ contrasting with *Dh = */D/), in addition to a number of cases of seeming *d ~ *h1 interchange (**dwi-déḱm̥(t) > *h₁wi(h₁)ḱm̥tih₁, **dḱm̥tóm > *h₁ḱm̥tóm, although there are other explanations for ἑκατόν) and the fact that, even though they can't be ejectives, they sure pattern like them. (Is *D-like patterning attested for other types of stop?)

Isn’t this just the glottalic theory?

No, the original glottalic theory was that they were ejectives. Proposals for *D as a voiced glottalized series of some sort came much later, as "Cao Bang theory" - Cao Bang being a Tai variety that shifted *ˀD *D > D Dh, with the idea being that Greek, Armenian, and Indic did the same. (What happens in Nuristani? Is the Iranian *D/*Dh merger secondary within IIr?)

I don’t think I’ve ever heard of implosives becoming liquids… do you have any examples?

Implosives became nasals in Vietnamese and ɗ~ˀl interchange is attested in Tsou.