The Great Proto-Indo-European Thread

[2+3 clusivity]

Isn't the Hittite plural copula's root in <as-> just /h1s-/? Am I missing something? 2Nd person plural is odd, granted, but it seems like leveling.

[KathTheDragon]

Isn't the Hittite plural copula's root in <as-> just /h1s-/? Am I missing something? 2Nd person plural is odd, granted, but it seems like leveling.

Yes, but Jouna rejects the modern laryngeal theory, so I am avoiding presupposing it in pulling apart his arguments.

[Sleinad Flar]

A survey of my reference material turned up one Old English verb with a preterite even remotely approaching the one you gave, hēawan 'strike; chop', with a preterite hēow. However, no Tocharian śauw- turned up when I looked through IE verbs from the root. Please cite words properly in the future. How am I supposed to reply to you when I don't even know what you're talking about?

I think he's referring to the form śauwā-, the ToB causative preterite of kau- 'kill'. See http://www.win.tue.nl/~aeb/natlang/ie/tochB.html (with some ctrl-F). Palatalization is a productive way to form causative preterites in Tocharian, so that's probably why the form doesn't show up in IE dictionaries.

[KathTheDragon]

A survey of my reference material turned up one Old English verb with a preterite even remotely approaching the one you gave, hēawan 'strike; chop', with a preterite hēow. However, no Tocharian śauw- turned up when I looked through IE verbs from the root. Please cite words properly in the future. How am I supposed to reply to you when I don't even know what you're talking about?

I think he's referring to the form śauwā-, the ToB causative preterite of kau- 'kill'. See http://www.win.tue.nl/~aeb/natlang/ie/tochB.html (with some ctrl-F). Palatalization is a productive way to form causative preterites in Tocharian, so that's probably why the form doesn't show up in IE dictionaries.

Ah, yes, I see it now. I don't know that much about Tocharian, since I have little more than references in my other sources.

[Tropylium]

On the other hand there are several factors why PIE *h : ɦ was ultimately opted, viz.
1. The traditional notation for the series bh dh gh is wrong (still so in IPA). The proper notation is actually *bɦ dɦ gɦ (with a voiced glottal fricative) which makes them in terms of PIE as clusters b+ɦ, etc. with *ɦ.

This reconstruction seems entirely wrong to me. As you know, the traditional reconstruction of voiced aspirates rests on a single, non-primary branch — Indic. And as you also note, secondary voiced aspirates from clearly reconstructible *D+h₂ behave identically. Instead of projecting voiced aspirates backwards, I suspect the solution should be that voiced aspirates are an innovation in Indic.

This also does away with the typological issue of the existence of voiced aspirates without voiceless ones. After all, voiceless aspirates were innovated in Indic as well. (Perhaps also in Iranian; but I find the reanalysis I've recently seen suggested, with e.g. *pH > *f being equal to *pC > *fC and *H > ∅, quite attractive.)

The question, then, is how to distinguish the two voiced series. The "compromise glottalic" theory as supported by e.g. Martin Kümmel, which posits plain voiced stops vs. implosives, seems the most likely to me. This allows combining the numerous possible traces of glottalization in the traditional *D series as uncovered by Kortlandt (implosives in Sindhi, Winter's Law in Balto-Slavic, vestjysk stød in Danish, etc.) and the fact that most IE languages indicate original voicing. (It's also quite compatible with deriving at least some cases of the traditional *D series from some sort of interaction with laryngeals.)

Your examples of the analysis of Dʰ as Dɦ (p. 400) strike me as not very strong at all, e.g.
1) the Germanic words under 'krümmen, biegen' do not strictly require Late PIE *a. They are derivable from *o, as indicated by Armenian, and the comparative method indeed requires that we do so. So there is no independent confirmation for a laryngeal, and your reconstruction of *gʰαVgʰα- = ? *gɦαVgɦα- instead of *gʰVgʰ- is ad hoc.
2) the Celtic words under 'strong' derive regularly from the zero grade *l̥, as also indicated by Indic, and do not require Late PIE *a either.

I know you suggest that syllabic sonorants simply merged with plain sonorants outside of II, instead of developing svarabhakti vowels, but this argument appears to again rest on two fallacious grounds: your unconstrained application of PIE ablaut (therefore positing, in violation of Occam's Razor, unnecessary ablaut alternants for data that can be traced back to single proto-forms); and your "Fortunatov's Law II", which consists of inserting unwarranted ad hoc laryngeals into roots and claiming that they produce Indic retroflexion.

As an aside, I'd be interested in hearing with how exactly do you deal with the developments *R̥ > *uR in Germanic, > *iR in Balto-Slavic, since your otherwise overpowered ablaut system still does not seem capable of generating *i or *u from nothing. I doubt you can show the existence of "variant" roots indicating *i or *u elsewhere for every single instance. Digging up a single case like Germanic *murθ- : Indo-Iranian *mur-, or Balto-Slavic *mir- : Late Avestan mirya- (whose regular derivation from earlier mərya- you dismiss!) strikes me as mere confirmation bias: uncritical cherrypicking as opposed to systematic scholarship.

But back to your reanalysis of the voiced aspirates. Your argumentation seems to again rest on your unnatural and unacceptable conviction that your *α would have to always co-occur with a laryngeal. You call it a conclusion, but you only reach this conclusion via repeated appeals to circular logic. It seem that your thesis is completely drenched in loops of this sort.

Here the argument seems to be: 1) *ɦα or *αɦ, never plain *α, must be reconstructed because the two always co-occurred; 2) you argue for an *a-grade, which you think implies earlier *α; this then implies a laryngeal. 3) Since a laryngeal occurred, we can analyze *dʰ apart into *dɦ. 4) This thus confirms that a laryngeal indeed co-occurred with *α.
And we are simply back where we started, not at any kind of a conclusion soundly built from premises.

[KathTheDragon]

As you know, the traditional reconstruction of voiced aspirates rests on a single, non-primary branch — Indic.

Remember that Indo-European studies have been plagued by a centricism on Sanskrit pretty much since the field's inception. It's not going to disappear overnight.

[WeepingElf]

On the other hand there are several factors why PIE *h : ɦ was ultimately opted, viz.
1. The traditional notation for the series bh dh gh is wrong (still so in IPA). The proper notation is actually *bɦ dɦ gɦ (with a voiced glottal fricative) which makes them in terms of PIE as clusters b+ɦ, etc. with *ɦ.

This reconstruction seems entirely wrong to me. As you know, the traditional reconstruction of voiced aspirates rests on a single, non-primary branch — Indic.

Yes, this is a weakness of the traditional model, attributable, as Kath has pointed out, on the historical Sanskrit-centrism of PIE studies. The early 19th-century IEists assumed that Sanskrit was PIE, based on the notion that the Rig Veda was 5,000 years old. When this turned out to be untenable, they moved their reconstruction away from Sanskrit only as much as was necessary to account for the facts. And because reconstructing the stop grade reflected as breathy-voiced in Indic was not in direct contradiction against the facts, these phonemes remained in place, regardless of their rarity outside Indic.

That doesn't mean that these phonemes could not have been breathy-voiced stops in PIE, but if positive evidence comes from only one branch and even its closest relatives within IE, the Iranian languages, don't show it, the matter becomes doubtful. (Greek, at least, points at aspiration.) For instance, assuming that they were voiced fricatives instead, works IMHO quite well - but then, they are reflected as such again in only one branch, Germanic.

And as you also note, secondary voiced aspirates from clearly reconstructible *D+h₂ behave identically. Instead of projecting voiced aspirates backwards, I suspect the solution should be that voiced aspirates are an innovation in Indic.

Yep. Though, as I wrote above, Greek does point at aspiration, and the majority of branches (in fact, all but Greek and partly Italic) point at voicing, so voiced aspirates are not implausible here. But as I wrote above, typologically less uncommon voiced fricatives work just as well.

This also does away with the typological issue of the existence of voiced aspirates without voiceless ones. After all, voiceless aspirates were innovated in Indic as well. (Perhaps also in Iranian; but I find the reanalysis I've recently seen suggested, with e.g. *pH > *f being equal to *pC > *fC and *H > ∅, quite attractive.)

The question, then, is how to distinguish the two voiced series. The "compromise glottalic" theory as supported by e.g. Martin Kümmel, which posits plain voiced stops vs. implosives, seems the most likely to me. This allows combining the numerous possible traces of glottalization in the traditional *D series as uncovered by Kortlandt (implosives in Sindhi, Winter's Law in Balto-Slavic, vestjysk stød in Danish, etc.) and the fact that most IE languages indicate original voicing. (It's also quite compatible with deriving at least some cases of the traditional *D series from some sort of interaction with laryngeals.)

This may have been the case, though I am not convinced yet. So PIE would have had a *T D' D instead of a *T D Dh system. While I have heard that implosives don't behave like ejectives with regard to the lack of a bilabial member and all that, the implosives may have been the outcome of pre-PIE ejectives. I have never seen the necessity of the glottalic theory to describe PIE as it was when the non-Anatolian IE branches parted from each other; it works perfectly well if one assumes the "glottalic system" to have been in place in some pre-stage.

Your examples of the analysis of Dʰ as Dɦ (p. 400) strike me as not very strong at all, e.g.
1) the Germanic words under 'krümmen, biegen' do not strictly require Late PIE *a. They are derivable from *o, as indicated by Armenian, and the comparative method indeed requires that we do so. So there is no independent confirmation for a laryngeal, and your reconstruction of *gʰαVgʰα- = ? *gɦαVgɦα- instead of *gʰVgʰ- is ad hoc.
2) the Celtic words under 'strong' derive regularly from the zero grade *l̥, as also indicated by Indic, and do not require Late PIE *a either.

I know you suggest that syllabic sonorants simply merged with plain sonorants outside of II, instead of developing svarabhakti vowels, but this argument appears to again rest on two fallacious grounds: your unconstrained application of PIE ablaut (therefore positing, in violation of Occam's Razor, unnecessary ablaut alternants for data that can be traced back to single proto-forms); and your "Fortunatov's Law II", which consists of inserting unwarranted ad hoc laryngeals into roots and claiming that they produce Indic retroflexion.

Ablaut is not a magical black box from which any vowel can be pulled as needed. It is a system of alternations governed by rules which are so complex that we don't understand them yet in their entirety, but no IEist worth his stripes doubts that those rules are there.

As an aside, I'd be interested in hearing with how exactly do you deal with the developments *R̥ > *uR in Germanic, > *iR in Balto-Slavic, since your otherwise overpowered ablaut system still does not seem capable of generating *i or *u from nothing. I doubt you can show the existence of "variant" roots indicating *i or *u elsewhere for every single instance. Digging up a single case like Germanic *murθ- : Indo-Iranian *mur-, or Balto-Slavic *mir- : Late Avestan mirya- (whose regular derivation from earlier mərya- you dismiss!) strikes me as mere confirmation bias: uncritical cherrypicking as opposed to systematic scholarship.

Right - there are no "u-grades" anywhere in PIE, and also no "i-grades" (though there is a class of reduplicated presents with */i/ in the reduplication syllable, but as no other forms have an "i-grade", this is better not ascribed to an exotic ablaut grade; instead, a morphological explanation is to be taken recourse to here).

But back to your reanalysis of the voiced aspirates. Your argumentation seems to again rest on your unnatural and unacceptable conviction that your *α would have to always co-occur with a laryngeal. You call it a conclusion, but you only reach this conclusion via repeated appeals to circular logic. It seem that your thesis is completely drenched in loops of this sort.

Here the argument seems to be: 1) *ɦα or *αɦ, never plain *α, must be reconstructed because the two always co-occurred; 2) you argue for an *a-grade, which you think implies earlier *α; this then implies a laryngeal. 3) Since a laryngeal occurred, we can analyze *dʰ apart into *dɦ. 4) This thus confirms that a laryngeal indeed co-occurred with *α.
And we are simply back where we started, not at any kind of a conclusion soundly built from premises.

Indeed, it is circular. And circular reasoning is not a trivial transgression; it is DEADLY.

[Pabappa]

Do you think a stop system consisting of plain voiceless, plain voiced, and voiced implosive could exist? That seems even more unlikely than the traditionalist voiceless/voiced/voiced-aspirate setup. Implosives could have existed at some point, but i doubt that there was ever a series of implosives consisting of just /ɗ ɠ ɠʷ/ or /ɗ ʄ ɠ ɠʷ/ (due to the lack of /b/ in most traditional reconstructions). So the implosive stage would have to be reconstructed for a stage further back, which would mean changing to a traditionalist voiced-aspirate setup for PIE proper. Also how would you explain PIE /b d g/ > Greek /pʰ tʰ kʰ/? Would voiced aspirates appear as an intermediate stage also here? The intermediate couldnt have been regular voiceless stops, since those already existed in the language and were not affected by the shift. If you have to use voiced aspirates twice, it seems more sensible to just put them in the common proto-language instead.

[vokzhen]

Do you think a stop system consisting of plain voiceless, plain voiced, and voiced implosive could exist? That seems even more unlikely than the traditionalist voiceless/voiced/voiced-aspirate setup. Implosives could have existed at some point, but i doubt that there was ever a series of implosives consisting of just /ɗ ɠ ɠʷ/ or /ɗ ʄ ɠ ɠʷ/ (due to the lack of /b/ in most traditional reconstructions). So the implosive stage would have to be reconstructed for a stage further back, which would mean changing to a traditionalist voiced-aspirate setup for PIE proper. Also how would you explain PIE /b d g/ > Greek /pʰ tʰ kʰ/? Would voiced aspirates appear as an intermediate stage also here? The intermediate couldnt have been regular voiceless stops, since those already existed in the language and were not affected by the shift. If you have to use voiced aspirates twice, it seems more sensible to just put them in the common proto-language instead.

At a guess, the theory is that a voiceless-voiced-implosive system becomes voiceless-voiced-glottalized, with the glottalized pushing the voiced series in some branches to breathy, and merging with the voiced in others. Such a system is fairly well-attested, but still unstable enough to split different ways in different languages. Lack of a labial implosive would be due to changing into /w/ or /m/, which is both cross-linguistically common and supported by PIE's odd distribution of *m *w *b (KathAveara has made several posts about this in the thread).

That doesn't mean that these phonemes could not have been breathy-voiced stops in PIE, but if positive evidence comes from only one branch and even its closest relatives within IE, the Iranian languages, don't show it, the matter becomes doubtful. (Greek, at least, points at aspiration.) For instance, assuming that they were voiced fricatives instead, works IMHO quite well - but then, they are reflected as such again in only one branch, Germanic.
[snip]
Yep. Though, as I wrote above, Greek does point at aspiration, and the majority of branches (in fact, all but Greek and partly Italic) point at voicing, so voiced aspirates are not implausible here. But as I wrote above, typologically less uncommon voiced fricatives work just as well.

Do they, though? The normal path of development is voiced stop > voiced fricative. I know of no instances of them becoming breathy, as they do in Indic and possibly in pre-Greek. And while they can be restored in part from fricative back to stop, it's rarely without loss: the voiced velar vocalized extensively in the history of Germanic, for example, and the labial remained a fricative in many positions and they can easily disappear entirely from some positions, as happened extensively in Western Romance.

[WeepingElf]

Do you think a stop system consisting of plain voiceless, plain voiced, and voiced implosive could exist? That seems even more unlikely than the traditionalist voiceless/voiced/voiced-aspirate setup. Implosives could have existed at some point, but i doubt that there was ever a series of implosives consisting of just /ɗ ɠ ɠʷ/ or /ɗ ʄ ɠ ɠʷ/ (due to the lack of /b/ in most traditional reconstructions). So the implosive stage would have to be reconstructed for a stage further back, which would mean changing to a traditionalist voiced-aspirate setup for PIE proper. Also how would you explain PIE /b d g/ > Greek /pʰ tʰ kʰ/? Would voiced aspirates appear as an intermediate stage also here? The intermediate couldnt have been regular voiceless stops, since those already existed in the language and were not affected by the shift. If you have to use voiced aspirates twice, it seems more sensible to just put them in the common proto-language instead.

At a guess, the theory is that a voiceless-voiced-implosive system becomes voiceless-voiced-glottalized, with the glottalized pushing the voiced series in some branches to breathy, and merging with the voiced in others. Such a system is fairly well-attested, but still unstable enough to split different ways in different languages. Lack of a labial implosive would be due to changing into /w/ or /m/, which is both cross-linguistically common and supported by PIE's odd distribution of *m *w *b (KathAveara has made several posts about this in the thread).

I am not at all sold by the implosive theory, nor by Kortlandt's list of alleged points of evidence. The implosive system would be a workable intermediate stage between the "classical" glottalic system and the traditional PIE one:

*T T' D > *T D' D > *T D Dh

Another possible pathway, though, and perhaps a more plausible one, would be:

*T T' D > *Th T D > *T D Dh

Especially if the Pre-PIE *T series was phonetically aspirated, as in Georgian.

A better understanding of Anatolian phonology would be helpful in finding out what happened. We know that the *D and *Dh grades merged in Hittite while *T stayed distinct, but AFAIK the jury is still out whether the difference was one of voicing or one of aspiration. I have seen arguments in favour of the latter (they wrote "/t/" as <tt> and "/d/" as <t>, rather than writing "/t/" as <t> and "/d/" as <d>; I think that was the matter), but I am no expert on this matter.

The question remains, though, whether a glottalist pre-stage (I think we can forget about glottalism in Late PIE) is warranted at all. As KathAveara and others have said, the paucity of *b may be due to a merger with *w, which in PIE occurs in environments where one would expect a stop, such as initial *wr-. The root structure constraints, however, seem to point at the *D grade having been a highly marked one in some stage.

That doesn't mean that these phonemes could not have been breathy-voiced stops in PIE, but if positive evidence comes from only one branch and even its closest relatives within IE, the Iranian languages, don't show it, the matter becomes doubtful. (Greek, at least, points at aspiration.) For instance, assuming that they were voiced fricatives instead, works IMHO quite well - but then, they are reflected as such again in only one branch, Germanic.
[snip]
Yep. Though, as I wrote above, Greek does point at aspiration, and the majority of branches (in fact, all but Greek and partly Italic) point at voicing, so voiced aspirates are not implausible here. But as I wrote above, typologically less uncommon voiced fricatives work just as well.

Do they, though? The normal path of development is voiced stop > voiced fricative. I know of no instances of them becoming breathy, as they do in Indic and possibly in pre-Greek. And while they can be restored in part from fricative back to stop, it's rarely without loss: the voiced velar vocalized extensively in the history of Germanic, for example, and the labial remained a fricative in many positions and they can easily disappear entirely from some positions, as happened extensively in Western Romance.

Fine. Certainly, breathy-voiced stops explain things best in terms of observed features. Greek points at aspiration; most branches point at voicing. This makes breathy-voiced stops most plausible. There is indeed no known precedent for *Ð > *Dh, but then, breathy-voiced stops are rare enough outside Indic to make typological generalizations difficult.

[KathTheDragon]

FWIW, Germanic shows at least two instances of the development voiced fricative > voiced stop. The first is some time after the application of Verner's Law, when the voiced fricatives produced by it and Grimm's Law because stops after homorganic sonorants and word-initially (except in the case of *g, to judge from Old English). The second instance is the total change of *d from [ð] to [d] in West Germanic (it is clear that Gothic did not participate in this change).

[Tropylium]

Do you think a stop system consisting of plain voiceless, plain voiced, and voiced implosive could exist? That seems even more unlikely than the traditionalist voiceless/voiced/voiced-aspirate setup.

Voiceless/voiced/implosive is in fact relatively commonly found. Martin Kümmel has the following quote in Typology and reconstruction: The consonants and vowels of Proto-Indo-European:

1) According to Hopper (1973: 141) “a typologically plausible triple stop system should have only one voiced series”. Is this true?
No. Even if most types of systems with two voiced series are quite rare, one such type is rather frequent: about 16% of the three-series languages counted by Maddieson (1984: 28f.) exhibit a system of voiceless stops opposed to plain voiced explosives and voiced implosive stops. In this sample, implosives were not distinguished from laryngealized voiced stops (see Clements & Rialland 2005: 19; Hamann & Fuchs 2008: 104f.). Recent research has provided evidence that “implosives” should rather be defined as “non-explosive” or “non-obstruent” stops – ingressive airstream being only a secondary feature (see Stewart 1989: 231ff.; Clements & Osu 2002; 2005; Clements & Rialland 2005: 17ff.) – and that “laryngealized voiced stops” are “non-explosives” with distinctive glottalization and less voicing, something like [ˀɗ̥] (Clements & Rialland 2005: 19ff.). But even if this distinction is made, the type remains rather frequent (if only in Africa).

…Although I believe there is another option which is quite common as well (I'd have to check Maddieson's paper to see if he covers it): the voiceless/voiced oral/voiced prenasalized system (P : B : ⁿB). But this does not seem to fit PIE very well at all, for one because both voiced series can occur before a nasal.

What seems like a bigger problem is that implosive systems are indeed usually front-heavy. /ɠ/ is a lot rarer than /ɓ/ /ɗ/, and I don't know if I've ever even seen a language with /ɠʷ/.

Of course, presuming an even earlier system with ejectives would again help here…

Also how would you explain PIE /b d g/ > Greek /pʰ tʰ kʰ/? Would voiced aspirates appear as an intermediate stage also here?

That seems necessary, yes. Possibly the same applies for Italic as well. A shorter option there would be *b *d *g > β ð ɣ, as in Germanic, with later devoicing word-initially. But alternately the development of aspirates and their subsequent devoicing could have been an Italic-Venetic-Paleobalkan-Greek areal feature.

If you have to use voiced aspirates twice, it seems more sensible to just put them in the common proto-language instead.

Assuming the shift *D > *Dʰ in two areas is indeed not completely optimal. But this seems like it would have an inter-systemic motivation from the existence of the "second" voiced series. If we instead consider the *Dʰ : *D system to have already been in place in late PIE, we will have to instead assume the shift *Dʰ > *D in at least five groups (Anatolian, Celtic, Germanic, Iranian+Nuristani, Balto-Slavic+Albanian; possibly also Armenian, though some dialects do have voiced aspirates), an interesting contrast to Indic, where the *Dʰ series seems to be quite stable.

Compromises between these could be sketched, e.g. the "Cao Bang shift" (as Michael Weiss seems to have named it) *D : *Ɗ > *Dʰ : *D being a common Indo-Greek-Armenian innovation, which was then areally also acquired by Italic. In this case only Iranian+Nuristani and (parts of?) Armenian would have undergone a back-development *Dʰ > *D.

But there will probably be further ideas about this yet, at some point by someone.

As one possible clue, Weiss notes in the presentation linked above that of the PIE root structure constraints, the rarity of *DeD is actually less weird than then commonality of *DʰeDʰ. Perhaps Jouna is indeed on the right track in supposing that the *D/*Dʰ contrast is actually a split of an earlier single series? But if so, it seems it would probably have targetted all voiced stops within a root.

(I'm rambling by now. Let's call it a post, here.)

[WeepingElf]

Do you think a stop system consisting of plain voiceless, plain voiced, and voiced implosive could exist? That seems even more unlikely than the traditionalist voiceless/voiced/voiced-aspirate setup.

Voiceless/voiced/implosive is in fact relatively commonly found.

Yes - they aren't really that rare, especially in Africa, they are common. (Now, of course, PIE wasn't spoken in Africa.)

Martin Kümmel has the following quote in Typology and reconstruction: The consonants and vowels of Proto-Indo-European:

1) According to Hopper (1973: 141) “a typologically plausible triple stop system should have only one voiced series”. Is this true?
No. Even if most types of systems with two voiced series are quite rare, one such type is rather frequent: about 16% of the three-series languages counted by Maddieson (1984: 28f.) exhibit a system of voiceless stops opposed to plain voiced explosives and voiced implosive stops. In this sample, implosives were not distinguished from laryngealized voiced stops (see Clements & Rialland 2005: 19; Hamann & Fuchs 2008: 104f.). Recent research has provided evidence that “implosives” should rather be defined as “non-explosive” or “non-obstruent” stops – ingressive airstream being only a secondary feature (see Stewart 1989: 231ff.; Clements & Osu 2002; 2005; Clements & Rialland 2005: 17ff.) – and that “laryngealized voiced stops” are “non-explosives” with distinctive glottalization and less voicing, something like [ˀɗ̥] (Clements & Rialland 2005: 19ff.). But even if this distinction is made, the type remains rather frequent (if only in Africa).

…Although I believe there is another option which is quite common as well (I'd have to check Maddieson's paper to see if he covers it): the voiceless/voiced oral/voiced prenasalized system (P : B : ⁿB). But this does not seem to fit PIE very well at all, for one because both voiced series can occur before a nasal.

It indeed doesn't fit PIE.

What seems like a bigger problem is that implosive systems are indeed usually front-heavy. /ɠ/ is a lot rarer than /ɓ/ /ɗ/, and I don't know if I've ever even seen a language with /ɠʷ/.

Of course, presuming an even earlier system with ejectives would again help here…

Yep. Implosive systems are usually front-heavy, while the PIE *D grade is back-heavy. So, implosives do not fit well here (other than as a transitional stage between ejectives, which tend to be back-heavy, and the plain voiced stops we see in Late PIE).

Also how would you explain PIE /b d g/ > Greek /pʰ tʰ kʰ/? Would voiced aspirates appear as an intermediate stage also here?

That seems necessary, yes. Possibly the same applies for Italic as well. A shorter option there would be *b *d *g > β ð ɣ, as in Germanic, with later devoicing word-initially. But alternately the development of aspirates and their subsequent devoicing could have been an Italic-Venetic-Paleobalkan-Greek areal feature.

Maybe. The Greek development seems to require voiced aspriates - which may have been still there in Greek's closest relative, Ancient Macedonian! At least, I seem to remember reading somewhere that the AM reflexes of PIE *Dh are sometimes spelled β δ γ, sometimes φ θ χ - which would make sense if they still had been voiced aspirates. So we would have them in both Greco-Macedonian and Indic (and, apparently, at least some dialects of Armenian) - and probably at least in the Greco-Aryan dialect area in Late PIE.

If you have to use voiced aspirates twice, it seems more sensible to just put them in the common proto-language instead.

Assuming the shift *D > *Dʰ in two areas is indeed not completely optimal. But this seems like it would have an inter-systemic motivation from the existence of the "second" voiced series. If we instead consider the *Dʰ : *D system to have already been in place in late PIE, we will have to instead assume the shift *Dʰ > *D in at least five groups (Anatolian, Celtic, Germanic, Iranian+Nuristani, Balto-Slavic+Albanian; possibly also Armenian, though some dialects do have voiced aspirates), an interesting contrast to Indic, where the *Dʰ series seems to be quite stable.

True. *Dʰ > *D would have to have happened in many branches independently. And as you say, the breathy-voiced stops are quite stable in Indic - but Indic also has voiceless aspirates, creating a neat 2x2 system. Late PIE did not have voiceless aspirates, making the whole thing less stable.

The assumption of voiced fricatives instead of the voiced aspirates would give Late PIE a sort of 2x2 system, too (there were voiceless fricatives, at least *s, and probably also the laryngeals) which would have collapsed after laryngeal loss.

Compromises between these could be sketched, e.g. the "Cao Bang shift" (as Michael Weiss seems to have named it) *D : *Ɗ > *Dʰ : *D being a common Indo-Greek-Armenian innovation, which was then areally also acquired by Italic. In this case only Iranian+Nuristani and (parts of?) Armenian would have undergone a back-development *Dʰ > *D.

So Cao Bang did it. May thus have happened in Late PIE, or in the Greco-Aryan dialect area thereof.

But there will probably be further ideas about this yet, at some point by someone.

As one possible clue, Weiss notes in the presentation linked above that of the PIE root structure constraints, the rarity of *DeD is actually less weird than then commonality of *DʰeDʰ. Perhaps Jouna is indeed on the right track in supposing that the *D/*Dʰ contrast is actually a split of an earlier single series? But if so, it seems it would probably have targetted all voiced stops within a root.

(I'm rambling by now. Let's call it a post, here.)

Hmm. A split could have happened somewhere. After all, the best candidate for the closest living kin of IE - Uralic - has just one grade of stops. Of course, Uralic could have gone the Tocharian way and collapsed them.

[WeepingElf]

An aftherthought:

Perhaps the "voiced aspirates" were realized differently in Late PIE dialects: breathy-voiced stops in Greco-Aryan and voiced fricatives in NW IE. Greek and Indic, as well as Armenian and perhaps Macedonian, point at the former; Germanic and Italic might point at the latter.

[Salmoneus]

Italic doesn't have to point at fricatives. After all, most 'voiced aspirates' end up as stops in Latin, and even in Sabellic they end up as voiced stops after fricatives or nasals.

[Pabappa]

Do you think a stop system consisting of plain voiceless, plain voiced, and voiced implosive could exist? That seems even more unlikely than the traditionalist voiceless/voiced/voiced-aspirate setup.

Voiceless/voiced/implosive is in fact relatively commonly found.

Yes - they aren't really that rare, especially in Africa, they are common. (Now, of course, PIE wasn't spoken in Africa.)

Well, I meant voiceless/voiced/implosive specifically without a fourth or fifth series in addition. The Cao Bang slideshow shows that such systems do seem to exist, but they're very rare, and probably unstable. Most languages in Africa and SE Asia either have more than three series, or have three but one of them has only a few members.

If you have to use voiced aspirates twice, it seems more sensible to just put them in the common proto-language instead.

Assuming the shift *D > *Dʰ in two areas is indeed not completely optimal. But this seems like it would have an inter-systemic motivation from the existence of the "second" voiced series. If we instead consider the *Dʰ : *D system to have already been in place in late PIE, we will have to instead assume the shift *Dʰ > *D in at least five groups (Anatolian, Celtic, Germanic, Iranian+Nuristani, Balto-Slavic+Albanian; possibly also Armenian, though some dialects do have voiced aspirates), an interesting contrast to Indic, where the *Dʰ series seems to be quite stable.

True. *Dʰ > *D would have to have happened in many branches independently. And as you say, the breathy-voiced stops are quite stable in Indic - but Indic also has voiceless aspirates, creating a neat 2x2 system. Late PIE did not have voiceless aspirates, making the whole thing less stable.

The assumption of voiced fricatives instead of the voiced aspirates would give Late PIE a sort of 2x2 system, too (there were voiceless fricatives, at least *s, and probably also the laryngeals) which would have collapsed after laryngeal loss.

Well, I think five "dog bites man" stories is still more plausible than two "man bites dog". That is, *Dʰ > *D is more likely to happen many times than *D > *Dʰ is to happen just twice, considering how much digging had to be done by linguists to find even a single example of it that was not part of a wider shift involving voiceless stops.

Also, I cant find it on the page for some reason, but I agree that b > w (or ɓ > w) is quite possibly the explanation for the lack of traditional /b/ in PIE. I had forgotten about that. It is strange that it didnt affect *bʰ or *p though. I realize that sound change just screams "Bilabial fricative!!" but I think that whatever the case may have been in the deepest past that it's fairly certain that by PIE proper and even a ways back the three stop series were indeed stops in all positions.

Also, this is just a general comment, not directed specifically at any argument here or elsewhere, but I really wish PIE linguists could just get over their addiction to unconditional sound shifts, such as "all voiced stops become aspirated unconditionally in all positions". Shifts like that just dont happen very often in the real world. Apparently they happen at least sometimes, such as Cao Bang, but it doesnt have to be the only explanation or even the most popular explanation. Particularly given the PIE root structure constraints mentioned above, wouldnt it be more likely that there were originally just two stop series, and then there was a conditional shift that created a third one from certain uses of the other two? Rather than just mirroring back the PIE root structure constraints for a thousand years in absolutely unchanged form. I assume Cao Bang is related to Vietnamese, after all, which would mean it's a predominantly monosyllabic language, which would mean that there arent a lot of different environments to arise to affect sounds differently. PIE was certainly not like that at all. Armenian has had some PIE-like shifts, but it seems to have covered over the gaps in its sound system a lot better than the hypothetical pre-PIE did.

[Tropylium]

Do you think a stop system consisting of plain voiceless, plain voiced, and voiced implosive could exist? That seems even more unlikely than the traditionalist voiceless/voiced/voiced-aspirate setup.

Voiceless/voiced/implosive is in fact relatively commonly found.

Yes - they aren't really that rare, especially in Africa, they are common. (Now, of course, PIE wasn't spoken in Africa.)

Well, I meant voiceless/voiced/implosive specifically without a fourth or fifth series in addition. The Cao Bang slideshow shows that such systems do seem to exist, but they're very rare, and probably unstable. Most languages in Africa and SE Asia either have more than three series, or have three but one of them has only a few members.

Kümmel reports specifically that implosives are relatively common even among three-phonation systems — in contrast to the *P *Bʰ *B system that is barely attestable. I don't think four-series systems are especially common in Africa, either. So on a quest for a least worst solution, the implosive theory has at least a clear starting edge.

Judging by how almost no two IE branches have similar treatment of the stops, it seems like we shouldn't be looking for an especially stable solution anyway. A look at other language families is fairly illustrating:
• Proto-Sino-Tibetan is reconstructed with a *Pʰ *P *B system, which can still be directly attested as widely as e.g. Burmese, Classical Tibetan, and Cantonese Wu.
• Proto-Austronesian or at least Proto-Malayo-Polynesian is reconstructed with a *P *B system, which can be directly attested in all sorts of descendants (though several Oceanic languages do something funny).
• Proto-NE-Caucasian (also Proto-North Caucasian, if you accept that) is reconstructed with a *Pʼ *P *B system, and this survives in most descendants (leaving the issue of "tense"/"geminate" stops aside).
• Proto-Athabaskan had a *Pʼ *P *B system as well, supplemented by voiced and voiceless fricatives, which again survives in most Athabaskan languages.

(Having a good diachronic model of Proto-Niger-Congo and its development would be interesting here, but I don't think that has been worked out yet.)

If you have to use voiced aspirates twice, it seems more sensible to just put them in the common proto-language instead.

Assuming the shift *D > *Dʰ in two areas is indeed not completely optimal. But this seems like it would have an inter-systemic motivation from the existence of the "second" voiced series. If we instead consider the *Dʰ : *D system to have already been in place in late PIE, we will have to instead assume the shift *Dʰ > *D in at least five groups (Anatolian, Celtic, Germanic, Iranian+Nuristani, Balto-Slavic+Albanian; possibly also Armenian, though some dialects do have voiced aspirates), an interesting contrast to Indic, where the *Dʰ series seems to be quite stable.

True. *Dʰ > *D would have to have happened in many branches independently. And as you say, the breathy-voiced stops are quite stable in Indic - but Indic also has voiceless aspirates, creating a neat 2x2 system. Late PIE did not have voiceless aspirates, making the whole thing less stable.

The assumption of voiced fricatives instead of the voiced aspirates would give Late PIE a sort of 2x2 system, too (there were voiceless fricatives, at least *s, and probably also the laryngeals) which would have collapsed after laryngeal loss.

Well, I think five "dog bites man" stories is still more plausible than two "man bites dog".

The other main reason I still find the standard system suspicious is that *Dʰ > *D does not seem like a "dog bites man" story. Do any precedents for an unconditional shift *Dʰ > *D exist?

Clearly enough the standard system would have been unstable. But as you can gather from the Cao Bang presentation, it seems that whenever voiced aspirates arise, they often continue on to voiceless aspirates. Aside from a couple of cases where this happened entirely on its own, this is also well-attested from languages that already have voiceless aspirates, such as in Mandarin, where the Middle Chinese *B series becomes sometimes plain voiceless, sometimes aspirated. But of the major IE groups, only Greek and probably Italic seem to have done this. It seems to me that we should expect this to have happened wider, if voiced aspirates once occurred everywhere in IE.

Another change that's somewhat clearly attested from at least dialects of Armenian and NW Indo-Aryan languages such as Punjabi is a development *BʰV > PV̤ > PV, i.e. the shift of breathiness to the vowel, and its possible later loss. In Punjabi the series actually rather splits, if a single paragraph in the Wikipedia article is to be believed: they become voiceless when initial, plain voiced when non-initial. But neither of these developments seems to have happened starting from the PIE system either (unless we count Tocharian and Anatolian, but they admit multiple explanations).

---

As for the back-heaviness problem, one additional thought: recent phonological research, of which Kümmel mentions some, seems to indicate that "implosive" is not a phonation of stops, but rather a manner of articulation of its own. (This also explains e.g. why implosive affricates or fricatives do not exist, in contrast to ejectives.) And just like how there are voiced, voiceless and glottalic (ex)plosives, implosives can similarly exist as voiced, voiceless or glottalized.

Now "front-heaviness" vs. "back-heaviness" of stop series is normally connected to voicing. So perhaps we should expect voiceless implosives to naturally tend towards back-heaviness?

But since voiceless implosives are way rarer than voiced ones, this seems to put us back to square one on the weird typology problem…

[Pole, the]

That is, *Dʰ > *D is more likely to happen many times than *D > *Dʰ is to happen just twice, considering how much digging had to be done by linguists to find even a single example of it that was not part of a wider shift involving voiceless stops.

Well, considering that we also have another voiced series that is glottalized, I can't find anything puzzling in that sound change.

I look at it that way: if we started from /t ɗ d/, with the difference between two of the segments as /ɗ/ being glottalized, it would be sensible to compensate by moving /d/ in the opposite direction — towards breathy sounds. Thus the difference being realized more as [ɗ] : [dʰ], and then the breathiness becoming phonemic in some dialects.

[zompist]

• Proto-Sino-Tibetan is reconstructed with a *Pʰ *P *B system, which can still be directly attested as widely as e.g. Burmese, Classical Tibetan, and Cantonese.

Cantonese has no voiced stops. Are you thinking of Wu?

[WeepingElf]

I look at it that way: if we started from /t ɗ d/, with the difference between two of the segments as /ɗ/ being glottalized, it would be sensible to compensate by moving /d/ in the opposite direction — towards breathy sounds. Thus the difference being realized more as [ɗ] : [dʰ], and then the breathiness becoming phonemic in some dialects.

That at least sounds plausible, at least to me.

[hwhatting]

Judging by how almost no two IE branches have similar treatment of the stops, it seems like we shouldn't be looking for an especially stable solution anyway.

A point which can be used, of course, to defend the traditional reconstructions (T D D^h, K^w K K') as well.

The other main reason I still find the standard system suspicious is that *Dʰ > *D does not seem like a "dog bites man" story. Do any precedents for an unconditional shift *Dʰ > *D exist?

Well, the answer to that question may have not much of a bearing on the issue, if it happens in a different system with different systemic pressures.

Clearly enough the standard system would have been unstable. But as you can gather from the Cao Bang presentation, it seems that whenever voiced aspirates arise, they often continue on to voiceless aspirates. Aside from a couple of cases where this happened entirely on its own, this is also well-attested from languages that already have voiceless aspirates, such as in Mandarin, where the Middle Chinese *B series becomes sometimes plain voiceless, sometimes aspirated. But of the major IE groups, only Greek and probably Italic seem to have done this. It seems to me that we should expect this to have happened wider, if voiced aspirates once occurred everywhere in IE.

It is, of course, also conceivable that the voiced aspirates only were a feature of a "core PIE" after Anatolian and Tocharian split off - after all, the two resp. one series that they show may atually continue not the classical T D D^h system, but its precursor, whatever that was.

[2+3 clusivity]

Which languages show oblique cases in /-bh-/ v. /-m-/? I used to have a source discussing it, but cannot find it.

[KathTheDragon]

Which languages show oblique cases in /-bh-/ v. /-m-/? I used to have a source discussing it, but cannot find it.

This is an excellent source on the oblique plural cases.

[Tropylium]

• Proto-Sino-Tibetan is reconstructed with a *Pʰ *P *B system, which can still be directly attested as widely as e.g. Burmese, Classical Tibetan, and Cantonese.

Cantonese has no voiced stops. Are you thinking of Wu?

Whoops. Yes, correct.

Judging by how almost no two IE branches have similar treatment of the stops, it seems like we shouldn't be looking for an especially stable solution anyway.

A point which can be used, of course, to defend the traditional reconstructions (T D D^h, K^w K K') as well.

Stability is not equivalent to rarity. It's rather a four-way split:
1) Things that arise easily but disintegrate rarely: common and stable. (E.g. /m/, /k/.)
2) Things that arise easily and disintegrate easily: unstable, but might be common if they arise more easily than disintegrate. (E.g. /θ/, /c/.)
3) Things that arise rarely and disintegrate rarely: stable, might be areally common if they disintegrate even more rarely than they arise. (E.g. /kp/, /kʼ/.)
4) Things that arise rarely but disintegrate easily: barely attestable. (E.g. /θʼ/, /cʷ/.)

Although so far, language typology has barely investigated "kinetics" at all, usually it's all "thermodynamics", with arguments based mostly on assumed compatibility with universal grammar.

It is, of course, also conceivable that the voiced aspirates only were a feature of a "core PIE" after Anatolian and Tocharian split off - after all, the two resp. one series that they show may atually continue not the classical T D D^h system, but its precursor, whatever that was.

If we granted that the pre-Indic/pre-Greek system with voiced aspirates evolved from something else, and granted that there exists a "core PIE" group that excludes Tocharian and Anatolian, it pretty much follows directly that in the absense of other evidence, the "Cao Bang shift" should be considered core PIE at earliest.

I don't know about the latter, though. I know computational analyses every once in a while confirm Anatolian and Tocharian as early splits, but are there any innovations that could be common to all IE languages except A/T? Other than assumed stop system changes like these.

(I generally treat computational phylogenies as not very useful, as long as they do not allow checking out what would be the defining innovations of each subgroup they generate.)

[Terra]

The root structure constraints, however, seem to point at the *D grade having been a highly marked one in some stage.

(1) How so?
(2) By this, do you mean that *T and *Dh are more similar to eachother than either is to *D? The pairs *kap, *ghabh (take) and *tem, *dhem (dark, dim) come to mind, and make me wonder the same thing. But, the problem is, I haven't come across any other pairs like this, and 2 pairs doesn't seem strong enough to to be called a trend.