The Great Proto-Indo-European Thread

[WeepingElf]

Wait a sec, I thought most Indo-Europeanists now think that *a was just an allophone of *e when adjacent to *h2 and the back velars? And that *o goes back to an earlier *a and *e goes back to an earlier *ə?

There's a lot who believe that *a was just an allophone of *e, but I don't know how many believe the rest of what you wrote.

I think TaylorS has the idea from Glen Gordon, who thinks that Etruscan is the closest known relative of IE, and uses an idiosyncratic internal reconstruction.

@Tropylium: Salmoneus has pointed out the weak spots of your argumentation; I agree with him pretty much on this matter. It is not impossible that your model is right (nor is it impossible that KathAvera's is right, or Glen's/TaylorS's); but I don't think it is better than the traditional model, which has the advantage of accounting for many IE branches (as opposed to just one branch - Tocharian - that is quite weird anyway) best, and posits a more typologically commonplace (and therefore more likely) vowel system. As I have said before, the traditional PIE vowel system would have been that of the latest stage before breakup, and it may have been the result of a shift from some other system. I agree with you that PIE *e and *o were quite open - at least [ɛ] and [ɔ] if not [æ] and [ɒ] - and may have slowly crawled upwards in the vowel triangle. But why can't *a just have been [a]?

[KathTheDragon]

To be honest, my system isn't meant as a replacement of the traditional system, but rather as an ancestor to it. In deriving it, I started from the traditional *e *a *o and worked backwards. The only adjustment my system calls for in the traditional model is that *o be unrounded and allophonically long, the latter of which explains Brugmann's law, and *ó > ā in Hittite.

[Tropylium]

the traditional model (…) has the advantage of accounting for many IE branches (as opposed to just one branch - Tocharian - that is quite weird anyway) best, and posits a more typologically commonplace (and therefore more likely) vowel system.

Whether it actually explains all branches best is precisely the question.

And hence another problem in this is that the proto-language is not a privileged entity whose typological optimization would be the single most important thing to accomplish. What has to be rendered typologically plausible is the entire scheme of linguistic evolution, including the other phonological systems assumed to have occurred at some point, and the sound changes connecting these systems. We should indeed reconstruct a more typologically odd proto-system if this cost can be offset by more plausible trajectories of phonological development.

I don't think this should be an especially complicated point? It's recognized by all mainstream Indo-Europeanists as soon as we start to discuss the stop system. *P *B *Bʰ is typologically odd, but has been considered the easiest point of departure to explain the reflexes across the family. The contesting *P *Pʼ *B is typologically common, but requires some rather complicated developments to reach anything that isn't Germanic or Standard Armenian.

For the record, since Anatolian and Tocharian might have the least standard-PIE-compatible vowel systems, and are usually considered the earliest divergences (i.e. they carry more weight than deep-seated daughters like Latin or Old Norse), one compromise position could be to reconstruct a different system for Proto-Indo-Hittite; and the *e *a *o system only in the ancestor of all non-Anatolian/Tocharian languages.

[KathTheDragon]

We must absolutely bear in mind that PIE as reconstructed is not a snapshot of the language at a single point of time, but a confluence of features old and new.

[Tropylium]

Anyway, before going into further detail… Have I linked you people yet to one of the main papers I have been building these ideas off of? Namely Martin Kümmel's Typology and reconstruction: The consonants and vowels of Indo-European.

The main improvements I see this bringing are:
1. It explains the typologically off-balance vowel system in traditional PIE, where open *a is very rare. I think it's quite widely accepted that *o must have been somewhere around /ɑ/ in at least some pre-PIE period.

Could it not equally have been /e/ that was once /a/?

There's a whole discussion to be had about that, sure, but for one, we can observe that vowel systems with /i e a u/ (and no /o/) are generally more common than /i a o u/ (and no /e/). If you want to construct a more detailed model of pre-PIE vocalism where ablaut goes back to *a ~ *o, go right ahead. We can then investigate if that provides any especial benefits.
Perhaps a "square" system with *e ~ *o < *æ ~ *ɑ might be preferrable yet. This is what I've seen posited for early Proto-Germanic these days as well.

2. It explains Brugmann's Law. In languages where particular vowels unconditionally gain length, this always proceeds starting from the tensest vowels in the system, which normally are the open (and non-reduced) vowels.

Why does that need to be explained? /o/ > either /a/ or /a:/. Your typological argument assumes (for the o-theories) /o/ > /a:/, and then presumably /o/ > /a/ as an independent unrelated change wherever the first one doesn't happen. Wouldn't it be just as likely, though, for unconditional /o/ > /a/, and then /a/ > /a:/ in certain circumstances? That is, fine, the lengthening is more likely to happen when it was /a/ - but how does that tell us that it STARTED as /a/ rather than just becoming /a/? After all, we know it either became or started as /a/ already, so what does this argument add?

The problem is that Brugmann's Law only applies to original *o, and not pre-vowel-collapse *a of any origin (*h₂e- and the like). They must have still been separate by the time of this change. All traditional formulations recognize this and posit *o > *ō and later > *ā, where the first step is left as phonetically mysterious. (We could assume only that Late PIE *a was not yet a low vowel, but then we end up with a LPIE vowel system /e ɜ o/ which is typologically absolutely terrible.)

The model sketched as Kümmel has *o retaining a degree of earlier length, as [oˑ], even after *ō has arisen, but I'm not sure if that is a particularly good assumption. Has anyone heard of a vowel system where some vowels come in short vs. long pairs, while others come in longish vs. extra-long pairs?

4. If trad. *a is instead reconstructed as a non-open vowel such as /ʌ/ or /ɜ/, then its merger into /a/ can be equated with the widespread resolution of syllabic laryngeals and resonants as *a(R), and no separate vowel lowering needs to be assumed.

What do you mean by *a? Do you mean *h₂e?

Yes, Late PIE *a, which comes mainly from *e next to *h₂, probably also loanwords (e.g. *tawros 'bull', which definitely should not be reconstructed as *th̥₂wros or some other monstrosity).

Anyway, I'm not sure I follow this bit. Syllabic laryngeals and resonants presumably resolved as *(R)@(R). That schwa then merged with /a/ in some languages, but with other vowels in other languages, and in some languages was coloured by the surrounding consonants first. If you're saying that the resolant of syllabics merged with *h₂e before the latter lowered to /a/, that leaves you with all the non-/a/ resolutions to explain (and if you aren't, then I don't follow what you're gaining from this explanation).

Syllabics do not have a single patterning of what vowel was inserted. Germanic does *R̥ > *uR but *H̥ > *a; Balto-Slavic does *R̥ > *iR ~ *uR but *H̥ > *a. So already in the standard model, the chronology comes out as:
1. Resolution of syllabic resonants in the various LPIE dialects with various vowels, some (but not all) of them *ə.
2. A widespread lowering of *ə to *a.

Indeed, this seems like a problem. Though I don't think the Uralic evidence unambiguously points to *o. The "pork" word has three reflexes: Finnic *porsas; Mordvinic *purćəs; Permic *pårś. The consonant skeletons in these do not quite match. If we assume that there were two loans, then *porćas versus *porśas might do. However it's been recently found out that also *parćəs (or, in traditional vowel notation *parćes or *parćis) would produce /u-ə/ in Mordvinic. There is evidence that the shift in some limited amount of environments occurred in Finnic as well; if this included after *p, then we'd be in the clear. Another option might be loaning from early Mordvinic.

When you've got evidence on one side of the argument and no evidence on the other, it seems tendentious to re-write the existing evidence by developing theories in another field, the field of another language family, just to make it easier to keep on holding the theory there's no direct evidence for.

Rewriting PIE is not at all the motivation for a reanalysis of Uralic loanwords. It has enough reasons of its own, such that Uralic sometimes turns up *o even for Indo-Iranian loans that never had PIE *o (say, LPIE *ēlā > PII *ārā > trad. PU *ora 'awl, thorn'), which has been a known problem. If seeming *o in western Uralic can be in inherited vocabulary a secondary development from *a under certain conditions, then it's only natural to ask if this could have happened in loanwords as well.

FWIW we just as well also "have positive evidence" that PIE *r̥ (!) was a rounded vowel, given e.g. Northern Sami /tʃoarvi/, Erzya /sʲuro/, Mari /ʃur/, Udmurt and Komi /ɕur/ 'horn' (all from PII *śr̥wa).

Is that a surprise, though? After all, syllabic /r/ DID produce rounded vowels in several known surviving IE language families (…)

This was a methodological point: it's not very reliable to gather some modern reflexes at face value if the question is to determine what the original sound value during loaning was. We have to first see what can be reconstructed to have existed around the time of loaning. Which per current research turns out to be *śarwə and not *śorwa.

Tocharian, incidentally, does something quite weird with PIE *o and *a:
• short *a > *a
• short *o > *e (via *ɵ? merges with *ē > *ʲe)
• long *ā > *o (likely via *oː)
• long *ō > *a (via ??)
The last pair of developments in particular looks very strange, and some kind of a contrived roundabout would have to be assumed. The second is a bit unusual as well.

The second bit isn't unusual at all, I don't think.

Depends on what you mean by "unusual". It is not too suspicious to be believed, no — but it's not an especially common shift in vowel systems either. My entire argument is about the difference between "possible" and "probable", after all. A question of phonetics, and not of the segment inventory.

It all looks reasonably straightforward, no roundabout needed. /o/ fronts, leaving a short gap at the back. /a:/ shortens and raises to fill that gap by becoming /o/.

No, that is not straightforward if you pause to think about the phonological interpretation.
— If you appeal to a need to refill a "short gap", then why did it get filled by long *ā and not short *a?
— Loss of length that starts among just the open vowels (while leaving, at least, *ō until later) is at best extremely rare.
— If you assume *ā > *a > *o, does this not indeed require positing some roundabout development *a > *X > *a to avoid merging these?

[Salmoneus]

There's a whole discussion to be had about that, sure, but for one, we can observe that vowel systems with /i e a u/ (and no /o/) are generally more common than /i a o u/ (and no /e/).

Who said PIE had /i/ and /u/? I thought they were normally considered /j/ and /w/? In any case, they don't seem like normal vowels in terms of behaviour.
I would agree that at least some /o/ was at some point unrounded, but that doesn't mean the system was /e a/ rather than, eg, /a @/.*

The problem is that Brugmann's Law only applies to original *o, and not pre-vowel-collapse *a of any origin (*h₂e- and the like). They must have still been separate by the time of this change. All traditional formulations recognize this and posit *o > *ō and later > *ā, where the first step is left as phonetically mysterious. (We could assume only that Late PIE *a was not yet a low vowel, but then we end up with a LPIE vowel system /e ɜ o/ which is typologically absolutely terrible.)

Not seeing the problem here. *h₂e presumably at first changed to something front or mid; *o presumably changed to something back. That's what you'd expect anyway, so suggesting an /e a o/ phase (with *o > *A: in some situations) seems unexceptional. Whether that's *o>*A>*A: or *o>*o:>*A: doesn't really seem interesting. [Note for instance lengthening of non-final open mid vowels, but not of *a (or rather, the loss of length for *a:/*a) in Romance.]

1. Resolution of syllabic resonants in the various LPIE dialects with various vowels, some (but not all) of them *ə.
2. A widespread lowering of *ə to *a.

Yes. That seems very reasonable. And like I say: if you're merging *ə with *h₂e before they shift to *a (or shifting the latter and having the former be *a all along) then you have to explain how the former ends up as other vowels in other languages. If not... then what are you gaining that's actually different?

Rewriting PIE is not at all the motivation for a reanalysis of Uralic loanwords. It has enough reasons of its own

Then let's leave that to Uralists...

This was a methodological point: it's not very reliable to gather some modern reflexes at face value if the question is to determine what the original sound value during loaning was.

Defeasible evidence is better than no evidence!

Depends on what you mean by "unusual". It is not too suspicious to be believed, no — but it's not an especially common shift in vowel systems either.

It's not that uncommon - and who said that every single daughterlang of PIE could only have 'especially common' shifts? After all, you propose changes that are just as unlikely! And Tocharian is odd in many ways.

No, that is not straightforward if you pause to think about the phonological interpretation.
— If you appeal to a need to refill a "short gap", then why did it get filled by long *ā and not short *a?

Because /a:/ is much more likely to become /o/ (/O/, etc) than /a/ is. /a:/ has a strong tendency to back, and backed vowels have a tendency to round. Examples: English (at least three times iirc), Irish, Swedish, etc.

— Loss of length that starts among just the open vowels (while leaving, at least, *ō until later) is at best extremely rare.

If by 'extremely rare' you mean 'commonplace', sure. Eg. every branch of Romance at least twice, maybe sometimes three times I think.

— If you assume *ā > *a > *o, does this not indeed require positing some roundabout development *a > *X > *a to avoid merging these?

I just assumed /a:/ > /O/. But in any case it's worth remembering that the vowel symbols aren't discrete possibilities, they're vague indications of a place in a 3d space, and changes aren't instant but are shifts over time. So it does actually happen that two vowels cross each other 'in motion', as it were. Vowels can come quite close to one another without merging, too.

[vokzhen]

There's a whole discussion to be had about that, sure, but for one, we can observe that vowel systems with /i e a u/ (and no /o/) are generally more common than /i a o u/ (and no /e/).

Who said PIE had /i/ and /u/? I thought they were normally considered /j/ and /w/? In any case, they don't seem like normal vowels in terms of behaviour.

There may be good theoretical reasons to have /j w/ and not /i u/, but I think this is one of those cases where phones are more important than phonemes (like arguing certain languages lack nasal phonemes, though they clearly have widespread nasal phones). Even if PIE didn't have /i u/, from my understanding it's assumed to have something close to by pretty much everyone (at least by the stage of PIE usually discussed), which is really what's going to be important when looking at the stability of the vowel system.

[TaylorS]

I think TaylorS has the idea from Glen Gordon, who thinks that Etruscan is the closest known relative of IE, and uses an idiosyncratic internal reconstruction.

@Tropylium: Salmoneus has pointed out the weak spots of your argumentation; I agree with him pretty much on this matter. It is not impossible that your model is right (nor is it impossible that KathAvera's is right, or Glen's/TaylorS's); but I don't think it is better than the traditional model, which has the advantage of accounting for many IE branches (as opposed to just one branch - Tocharian - that is quite weird anyway) best, and posits a more typologically commonplace (and therefore more likely) vowel system. As I have said before, the traditional PIE vowel system would have been that of the latest stage before breakup, and it may have been the result of a shift from some other system. I agree with you that PIE *e and *o were quite open - at least [ɛ] and [ɔ] if not [æ] and [ɒ] - and may have slowly crawled upwards in the vowel triangle. But why can't *a just have been [a]?

WHOOPS, my bad, I thought for some reason that was a common assumption, since the lack of a low vowel is typologically unheard of except for (IIRC) Kiowa-Apache.

IIRC the argument against *a is that most cases of it involve an adjacent back-velar and that the rest can be assumed to be cross-dialectal borrowings.

[TaylorS]

Anyway, before going into further detail… Have I linked you people yet to one of the main papers I have been building these ideas off of? Namely Martin Kümmel's Typology and reconstruction: The consonants and vowels of Indo-European.

This is awesome, thanks!

[WeepingElf]

Anyway, before going into further detail… Have I linked you people yet to one of the main papers I have been building these ideas off of? Namely Martin Kümmel's Typology and reconstruction: The consonants and vowels of Indo-European.

That is a very interesting paper, thank you! Also, I see no problems with it right now. It may well be that PIE *e once was /a/, and PIE *o once was /a:/. From there, the following trajectories are plausible:

*a (> *ə) > *æ > *ɛ = "Classical" PIE *e
*a: (> *a) > *ɒ > *ɔ = "Classical" PIE *o

The bits in parentheses are optional; Glen Gordon may be something of a crank, but that doesn't mean that he can't be right about this one. He is clearly less deranged than e.g. Octaviano.

The laryngeal colouring effects are clearly later than the rise of ablaut, and may have kicked in at the *æ *ɒ stage.

Let me add two further matters here, both involving PIE laryngeals and the hypothesis that they and the velar stop series split into three in Pre-PIE according to the colours of neighbouring vowels.

One concerns PIE laryngeals and their possible cognates in Uralic. Adam Hyllested has written an interesting paper in which he observes a correlation between PIE laryngeal colour and PU vowel colour. Basically, all PIE laryngeals correspond to PU *k in initial and PU *x in medial position. If the PU form has a front (unrounded) vowel, PIE has *h1; if PU has a low vowel, PIE has *h2; if PU has a rounded vowel, PIE has *h3. The correspondence is not perfect, though.

The second matter concerns frequencies of velar stops and laryngeals. Among the velar stops, the front velars (*ḱ ǵ ǵh) are the most frequent, but among the laryngeals, the back laryngeal *h2 is the most frequent. Why this discrepancy? Two ideas:

1. The frequency of *h2 may be a mirage. There are many laryngeals of unknown colour in PIE; basically, all laryngeals that have neither an e-grade vowel next to them nor are reflected as vowels in Greek remain unknown - many of them may have been *h1. Also, many instances of *h1 may have been missed as this laryngeal has no vowel-colouring effects and is accordingly hard to detect. There are quite a few forms where we know that a laryngeal was there only because there is an *a or *o where an *e would be otherwise expected. *h1 would remain undetected in such instances.

2. The antecedent of the laryngeals before the three-way split of the velar series may have been articulated farther back than the velar stops, perhaps *χ rather than *x, so the "default" POA for it was "back velar" while for the stops, it was "front velar".

EDIT: Added link to Hyllested's paper.

[2+3 clusivity]

Some Thoughts on the Structure of PIE Personal Pronouns and Related Forms.

While doing some PIE reconstruction stuff, I started working on the following. None of it is finished, polished, or particularly well thought through -- not sure if I will finish it. Anyway, TL;DR my musings -- and errors -- follow:

(1) Thematic Elements in Independent and Enclitic Personal Pronouns. First and Second Person PIE pronouns used two major thematic formants to distinguish person: (1) /*m-/ v. /*t-/ and (2) /*(-)u̯-/ v. /*(-)i̯-/. These formants were heavily utilized by daughter languages and often spread through the pronominal paradigms via analogy.

(1)(a) /*m-/. An /*m-/ formant occurs in non-nominative 1st sg forms. Further, this formant is echoed in the Athematic and copular 1st sg. verbal ending: /*-mi / and, perhaps, in the 1st plur /*-mes~mos/. As the verbal forms typically echo the oblique cases rather than the nominative, the latter suggests 1st plur obliques were formed in /*n̥s-/ < /**m̥s/. Beyond simple analogy, such a reconstruction might help explain remnant /*(-)m-/ in forms such as:

Armenian 1st nom plural: /mekʰ/ < /*m-es/, 1st acc plur /mez/ < /?**me-ǵʱ-V/, 1st gen plur /mer/ (C.f. Armenian singulars, which do not have /m-/)

1st plural enclitic Tocharian A: /-m/ and Tocharian B: /-me/ (non-enclitics generalized the nominative) assuming this form generalized out of the first personal to all other plurals

The /*n̥s-/ < /**m̥s/ change appears early and general after referencing Hittite, which shows oblique plurals in /ants-/. Similarly, Luvian forms show: < ānts-a >, <ants-antsa> (reduplicated?), etc. At least in Anatolian, the affricate /-ts-/ even more strongly suggests a POA assimilation such as /*ants-/ < /**m̥t(s)-/. Anatolian /-ts-/ v. general PIE /*-s-/ in these forms is itself rather curious.

(1)(b) /*t-/. A /*t-/ occurs in nominative and non-nominative 2nd singular forms. The non-nominative forms are generally unremarkable -- even Classical Armenian /kʰ-/ forms reflect < /?*tɣ(ʷ)-/ < /*tu̯-/. It should be stressed that the thematic correspondence between nominative and non-nominative /*t-/ is very unusual amongst PIE pronouns and demonstratives as generally reconstructed. Perhaps late formed nominative singular might have once been /**uH-/ before a /*t-/ was analogized onto it from the oblique forms.

The non-nominative /*t-/ is also reflected in athematic verbal singular /*-si/ (Beekes, Sihler, Fortson, Ringe), which appears to be < /**-ti/ given active Hittite: /-si/ and /-ti/, Tocharian A: /-t/, Tocharian B: /-t(o)/, and a variety of /*-t(H)-/ forms throughout PIE in non-present-active. Dual showed /*-t-/ forms; Sihler: /*-tH1es/, Beekes: /-tHVs/, Fortson: /*-to-/, Ringe: /*-tes/. Plural forms show: /*-te/ (Sihler, Ringe, Fortson) or /*-tH1e/ (Beekes).

Note also that both nominative and non-nominative /*t-/ frequently take /-(e)u̯-/ extensions – i.e. the other second person formant. The nominative may have been prefixed /**t-/ to earlier forms, either 2nd plur thematic /**-i-/ (Anatolian) or a thematic /**-u(H)-/ stem shared with the dual and plural: /**-uH-/ (as seen in other PIE languages).

(1)(c) /*u̯-/. The /*u̯-/ formant pervades the first person, particularly in the plural and dual but also extending into the singular in Anatolian and Tocharian. /*u̯-/ is most apparent in the nominative plural forms such as Tocharian B: /wes/ and Gothic /weis/ < /*u̯-ei-/ (Beekes, Sihler and M&A). Hittite forms suggest /*u̯-VHs/ paralleling 2nd plural nominative, while Tocharian forms appear enclitic in origin. In the dual, /*u̯-/ also characterizes the nominative in forms such as Tocharian B: /wene/ and Gothic: /wit/ < Sihler /*we-H1-/ but c.f. M&A /*nó-H1-/, the latter of which appears to be an oblique. Through analogy, Gothic obliques also oppose 1st dual /uŋ-/ v. an apparently later developed thematic 2nd dual /iŋ-/.

Note also, unlike most PIE verb forms, Hittite takes a 1st plural in /-ue-ni/ (/-ni/ is also suffixed to the otherwise normal 2nd plural /-te-ni/). The Hittite forms match well with reconstructions of the dual (showing thematic / u̯/!): Sihler & Ringe: /*-wos/, Beekes: /*u̯es /, and Fortson: /*-we-/. Perhaps the Proto-Anatolian plurals had a plural form in line with the dual /-w-/ (odd that these Anatolian forms seem to follow the nominative rather than the oblique formant, perhaps explaining the odd /-hi/ forms as parallel to ? /**H(e)-uḰ/ > /*a-uk/ -> /uk/.

Of particular interest are Hittite and Tocharian 1st Singular Nominative and Accusatives. Hittite Nominative/uk/ and accusative /am(m)uk/ v. the 2nd nominative /tsik(a)/ < /*tika/ (showing 2nd thematic /-i-/!) and the unremarkable/analogized 2nd accusative / tuk(a)/. The highly aberrant Tocharian Nominative/Accusative forms show:

Basic form: /*N(?)-uK-/, here N is similar or identical to the formant producing Enclitic A: /-ɲi/ and B: /-ɲə/), /*-u-/ is a 1st person formant, and /K-/ is a demonstrative ("I here"). It seems that an enclitic /*n-/ form was prefixed to or merged with the case-form of the non-enclitic first person pronouns. The /n-/ enclitic seems odd comparatively; however, it logically seems to arise by dissimilation from the general plural enclitic of Proto-Tocharian in /-m-/ -- itself, perhaps, picked from the 1st plural.

Tocharian A: Masc. /nɨʃ/ < (?via de-palatalizing dissimilation or by analogy with non-palatal demonstrative nominatives) /?*ɲə- u̯çɨ/ < /**ne+uḰV(s)/and Fem. /ɲuk/ [ɲəu̯k] < /*ɲə-u̯ko/ < /**ne+uḰā/ (/*-eu-/ would have produced /-o-/)

Tocharian B: /ɲəç, ɲɨç/ [?ɲəçə, ɲɨçə] < /?*ɲə́çə ~ *ɲəçə / < /**ne+uKe/
C.f. Toch. A. enc. /-ɲi/, M. 1st Sing Gen: /ɲi/, and F. 1st Sing. Gen: /nāɲi/; Toch B. enc. /-ɲə/, 1st Sing. Gen: /ɲi/

A discussion of /*u̯-/ would not be complete without considering 2nd person plural obliques in /*us-/ and enclitics in /*u̯oHs /. These forms have developed /*(-)u-/ as part of the pronoun stem. C.f. Sihler’s representative reconstruction: nominative: /*y-ūs, y-uHs/ v. accusative /*us-mé/, enclitic /*wōs/, reflexive /*s-u̯é/.

(1)(d) /*i̯-/. Similarly, the /*i̯-/ formant is found in the second person. It is most notable in the plural nominative reconstructed as: /*i̯-uH/ (Beekes), /*y-ūs, y-uHs/ (Sihler), and /*y-uHs/ (M&A) but C.f. also /*(u)swé/ (M&A). It appears in dual Tocharian B /y-ene/ and Gothic /y-ut/ and /i-ŋkʷ-/. As noted above, it appears in Hittite: /tsik(a)/.


(2) /*-Ḱ(-)/ extensions of PIE personal pronouns. Occurrences of /*-K(-)/ in PIE personal pronouns represent fossilized occurrences of the PIE demonstrative (?deictic adverb) /*ḱi/ (Beekes) -- perhaps /*ḱ(V)i/.

Background: 1st Sg Nom Personal Pronoun. PIE pronouns typically show a /*-Ḱ-/ formant in first person singular personal pronouns. This element has been reconstructed as follows:

Beekes PIE: /*H1eǵ-(oH/Hom)/;
Sihler PIE: /*eǵoH/;
M&A Pie: /H1eǵ-/.

Such reconstructions with /-ǵ-/ fit well for forms such as Gothic /ik/ < /*égʲ/ and, as traditionally reconstructed, Sanskrit /aham/ < /*H1eǵHom/ (Beekes) among other languages.

On the other hand, a /*-ǵ-/ reconstruction does not necessarily fit well with the following forms:

Early Hittite: /uk/, latter Hittite /am(m)uk/ < /?*uḱ/;

Classical Armenian: /es/ < /*eḱ/;

Arguably, Tocharian A: Masc. /nɨʃ/, Fem. /ɲuk/ and Tocharian B: /ɲəç, ɲɨç/ (note, in particular, the gender distinction in Tocharian A); and

More tentatively, Indo-Iranian forms could be reconstructed as /*HéḱH3Vm/

/*-Ḱ(-)/ Elsewhere.

First Person Singular: Classical Armenian: accusative /z-is/ < later /*?z-/ + PIE /*-ikʲ/; Tocharian A & B accusatives as mentioned above; Hittite: /am(m)u-k/ < /*H1m-ukʲ/. Also, Germanic forms such as Gothic accusative /mik/ < /*mék~mik/ (Ringe) < /méǵ/. Venetic forms as well?

First Person Dual: paradigmatically for non-nominative Germanic duals such as Gothic accusative /uŋkis/ < /*uŋk-/ (Ringe) < /?*n̥̄ǵ-/. C.f. Sihler. PIE: /*n̥H1-wé/ and Toch B:/wene/.

Second Person Singular: Hittite, Germanic, and maybe Venetic (though analogy seems more likely in that case).

Second Person Dual: paradigmatically for non-nominative Germanic duals such as Gothic accusative /iŋkʷis/ < /*iŋkʷ-/ (Ringe) < PIE Dual/Plural 2nd person formant /*i̯-/ + /?*-ŋǵ-/ by analogy with 1st person dual. C.f. Sihler. PIE: /*uH1-wé/ and Toch B:/yene/, which also shows /i̯-/ + analogy with the first dual.

[WeepingElf]

Nice stuff, 2+3!

Here is an interesting article on PIE pronouns.

[KathTheDragon]

And my thoughts on your thoughts. (In reference to Anatolian material, I'm using Kloekhorst's Hittite Etymological Dictionary.)

The /*n̥s-/ < /**m̥s/ change appears early and general after referencing Hittite, which shows oblique plurals in /ants-/. Similarly, Luvian forms show: < ānts-a >, <ants-antsa> (reduplicated?), etc. At least in Anatolian, the affricate /-ts-/ even more strongly suggests a POA assimilation such as /*ants-/ < /**m̥t(s)-/. Anatolian /-ts-/ v. general PIE /*-s-/ in these forms is itself rather curious.

Apparently, Hittite regularly changed sequences of /ms/ and /ns/ into /nts/ preceeding a vowel, except intervocalically, which leaves this preform ambiguous.

Perhaps late formed nominative singular might have once been /**uH-/ before a /*t-/ was analogized onto it from the oblique forms.

Note that the /i/ in the Anatolian nominative must be original, since there's no way to generate it otherwise.

The non-nominative /*t-/ is also reflected in athematic verbal singular /*-si/ (Beekes, Sihler, Fortson, Ringe), which appears to be < /**-ti/

How??

Hittite Nominative/uk/ and accusative /am(m)uk/ v. the 2nd nominative /tsik(a)/ < /*tika/ (showing 2nd thematic /-i-/!) and the unremarkable/analogized 2nd accusative / tuk(a)/.

The most likely explanation of these forms is that the 2s nom. i is original, and the rest of the paradigm possessed u/w. This then spread into the whole 1s paradigm via a nom-nominative form, probably the accusative. In the rest of IE, the u/w spread into the 2s nom. instead. Incidentally, this fixes the laryngeal in the 2s as h₁.

The highly aberrant Tocharian Nominative/Accusative forms show:

Basic form: /*N(?)-uK-/, here N is similar or identical to the formant producing Enclitic A: /-ɲi/ and B: /-ɲə/), /*-u-/ is a 1st person formant, and /K-/ is a demonstrative ("I here"). It seems that an enclitic /*n-/ form was prefixed to or merged with the case-form of the non-enclitic first person pronouns. The /n-/ enclitic seems odd comparatively; however, it logically seems to arise by dissimilation from the general plural enclitic of Proto-Tocharian in /-m-/ -- itself, perhaps, picked from the 1st plural.

If you assume a basic consonantism for the non-nominative 1s of h₁m-n- then the Tocharian form arises from the zero-grade of this (and also provides the Hittite form).On the other hand, a /*-ǵ-/ reconstruction does not necessarily fit well with the following forms:

Early Hittite: /uk/, latter Hittite /am(m)uk/ < /?*uḱ/;

This form is ambiguous, as Hittite doesn't distinguish between single and geminate stops (other than kʷ, apparently) in word-final position.

Arguably, Tocharian A: Masc. /nɨʃ/, Fem. /ɲuk/ and Tocharian B: /ɲəç, ɲɨç/ (note, in particular, the gender distinction in Tocharian A)

Even more ambiguous, since Tocharian merged all its obstruent series without exception.

First Person Dual: paradigmatically for non-nominative Germanic duals such as Gothic accusative /uŋkis/ < /*uŋk-/ (Ringe) < /?*n̥̄ǵ-/. C.f. Sihler. PIE: /*n̥H1-wé/

Or, with h₃ and Cowgill's Law, *n̥h₃-wé > ungwe > unkwe > unk, unkiz

[2+3 clusivity]

And my thoughts on your thoughts. . . . Note that the /i/ in the Anatolian nominative must be original, since there's no way to generate it otherwise. . . . The most likely explanation of these forms is that the 2s nom. i is original, and the rest of the paradigm possessed u/w. This then spread into the whole 1s paradigm via a nom-nominative form, probably the accusative. In the rest of IE, the u/w spread into the 2s nom. instead. Incidentally, this fixes the laryngeal in the 2s as h₁. . .

I think we are arguing to the same end -- but I am being less clear. I agree with you on the originality of /*-i-/ in the second singular nominative form, at least for Hittite.

Incidentally, this fixes the laryngeal in the 2s as h₁.

Wait, what does? Hittite /-k/ <--> PIE /*-h₁/?

If you assume a basic consonantism for the non-nominative 1s of h₁m-n- then the Tocharian form arises from the zero-grade of this (and also provides the Hittite form).

Hungh. What steps would you take?

Or, with h₃ and Cowgill's Law, *n̥h₃-wé > ungwe > unkwe > unk, unkiz

Was not aware of the second Cowgill's law. If correct, it's very interesting that the /k/ is cropping up out of *H3/?*H2.

[KathTheDragon]

No, no. The laryngeal lengthens the i in Hittite, and then the k comes from either the spread of the final consonant of the 1s nom. or from the 'emphatic clitic' (as in Germanic mik etc.) The laryngeal canjot have been h2 or h3, as even if they're not directly reflected, the i would certainly have been lowered to e.

For the Tocharian, we have h1mne- > mne- > ne- > ñe- (or something like that; I'm not familiar with the specifics of Tocharian sound change). In Hittite, h1mn- > hmm-, written amm-.

[kanejam]

So I've just had a look through A Concise Historical Grammar of the Albanian Language - Reconstruction of Proto-Albanian by Vladimir E. Orel (which has made me fall in love with Albanian and want to make it a sister language). Albanian is usually referred to as a satem language as the palatal velars become dental affricates (according to Vlad) and the pure velars are retained as velars.

However Vlad argues that Albanian, like Anatolian, actually maintains the threeway contrast of PIE, in that labiovelars before front vowels or /j/ become palatoalveolar affricates, but plain velars don't. The evidence he gives is:

- Albanian darkë 'supper' from Proto-Albanian *darkā from PIE *dorkʷom
- Albanian kollë 'cough' from PA *kāslā from PIE *kʷās-
- Albanian djeg 'to burn' from PA *dega from PIE *dhegʷh-

- Albanian pesë 'five' from PA *pentše from PIE *penkʷe
- Albanian sjell 'to bring' from PA *tšela from PIE *kʷel-
- Albanian zi 'black' from PA *džārnā from PIE *gʷed-

Firstly, I have no idea how sound his reconstructions of either PA or PIE are, although they are certainly convincing. Secondly, he doesn't give any examples that I can find of velars before front vowels, so this could be the result of a merger between labiovelars and velars. Thirdly, he gives the odd pathway: /kʷ/ -> /kw/ (and the /w/ was lost before non-front vowels or /j/) -> /tsw/ -> /tš/.

Vlad also implies a close genetic affinity between Albanian and Baltic (not necessarily saying that Albanian is Balto-Slavic, though - maybe a macro-grouping such as Greco-Armenian?). Either way, Albanian has some crazy changes and features: Latin ratiō came out as arsye (that <y> is /y/), causa became kafshë, spleneticum became shpretkë and episcopus became ipeshkv. The book also deals with Greek and Slavic loans. Interesting read.

[R.Rusanov]

Thracian and Balto-Slavic were definitely related; if Albanian comes from a Thracic or Dacic predecessor, instead of Illyrian (which is closer to Italic), it's certainly possible for a closer relationship to exist.

Cursorially looking at the roots you copied, kasla (cough) looks a lot like kashlya, kashlica (ibid) in Bulgarian.

[Pabappa]

Im not sure I even understand what the Vlad guy is arguing. Those who dont believe in three series always unite the plain velars with the palatals. So showing that labiovelars were distinct from velars proves nothing since basically everyone agrees with that already. What he should try to prove is that PIE velars have distinct outcomes from PIE palatals. Unless it's just some subpoint of a larger argument which proves that there were three series after all and that Albanian, despite being satem, is different from other satem languages because it doesnt merge labiovelars with velars before front vowels. But if so he should first show that the velars were distinct from palatals.

[Sleinad Flar]

Velars distinct from palatovelars is kind of the definition of Satem. Centum: velars merge with palatovelars, labiovelars distinct and satem: velars merge with labiovelars, palatovelars distinct. Albanian is satem (k' > ts > th, k^(w) > k), and Orel does show that, but apparently the merger of k and k^w is incomplete or, as Orel says, PIE k^w went through palatisation processes which k didn't. In IE languages this is rather unique (the only other language I can think of which does something similar is the extinct Anatolian language Luwian).

[hwhatting]

Thirdly, he gives the odd pathway: /kʷ/ -> /kw/ (and the /w/ was lost before non-front vowels or /j/) -> /tsw/ -> /tš/.

It's not that odd; it's similar to what happened to /tw/ and partially to /k^w/ in Greek, only in Greek it went /t^sw/ > /t^s/ > /s:/ (or > /t:/ in Attic).

Thracian and Balto-Slavic were definitely related; if Albanian comes from a Thracic or Dacic predecessor, instead of Illyrian (which is closer to Italic), it's certainly possible for a closer relationship to exist.

Nowadays, only Albanian nationalists believe that Albanian is a descendant of Illyrian (beacause that would mean that Albanians are living in Albania since antiquity). We don't know very much about Thracian and Dacian, but as they seem to have been Satem and Romanian seems to have a Proto- or Para-Albanian substrate, it's more likely that Albanian is related to Thracian and Dacian.

Im not sure I even understand what the Vlad guy is arguing. Those who dont believe in three series always unite the plain velars with the palatals. So showing that labiovelars were distinct from velars proves nothing since basically everyone agrees with that already. What he should try to prove is that PIE velars have distinct outcomes from PIE palatals. Unless it's just some subpoint of a larger argument which proves that there were three series after all and that Albanian, despite being satem, is different from other satem languages because it doesnt merge labiovelars with velars before front vowels. But if so he should first show that the velars were distinct from palatals.

I assume that kanejam didn't quote Orel on the outcomes of the palatals, becaue it's already a known fact that they are distinct from the velars and labiovelars - see the Wikipedia article for details. Actually, the idea that Albanian distinguishes the three series is not new - I remember that being discussed in the literature when I went to an introductory course about Albanian historical linguistics during my studies, in the late eighties.
Edit: ninja'd by seinad

[hwhatting]

On the discussion about /e/ and /o/ - one thing that needs to be looked into is the model how ablaut originated in stress patterns. The models I know all assume some variant of this sequence:
1) Older Accent pattern: Accented syllable e-grade, unaccented zero-grade
2) Newer Accent pattern, including some shifts by which zero-grade syllables can receive the accent, and some e-graded syllables loose it; the latter became o-grade (and then, depending on the model, analogy and other re-arrangements further divorce ablaut grade and seat of accent).
AFAIK, a part of all these models is that o-grade arose in unaccented syllables; in that case those who posit an initial value for ablaut /o/ as a long vowel (/a:/ or similar) need to explain why this unaccented vowel is long and accented /e/ is short. (Another possibility is, of course, to reject the usual theories about the origin of o-grade). I don't say it's impossible - it could be some kind of compensatory lengthening -, but it would be nice to see an explanation and possibly examples of similar developments in attested language history..

[kanejam]

Thracian and Balto-Slavic were definitely related;

Definitely, they're both Indo-European.

But if so he should first show that the velars were distinct from palatals.

Sorry, I was sort of unclear in my post; as the others have said, the velars were definitely distinct from the palatals:
PA kapmi 'have' < PIE *kap- 'grasp' (cf. Lat capiō)
PA atsara 'sharp' < PIE *aḱ- 'sharp' (cf. Lat acidus)

The interesting point is if the Albanian branch keeps all three dorsals distinct, as in Anatolian, and thus it would be apart from the centum/satem isogloss. However Vlad shows that labiovelars and pure velars merge everywhere except before front vowels and /j/, so what's necessary is showing that they are distinct in that environment. I have been able to find a single piece of evidence in Vlad's work:
Albanian gloq /gloc/ < PA *kalāukja < PIE *kom 'with' + *leuk- 'shine'
A PIE kʷ in the same position would yield PA **kalāutša, although I don't know where the *-ja suffix comes from and so it's possible the outcome doesn't necessarily reflect Pre-PA k+j.

If not, then it would be reasonable to assume a set of changes kʷ -> k unconditionally (ie satemisation) and then k -> tš before front vowels and /j/. Without evidence for plain velars before front vowels it's impossible to know, and probably more reasonable to assume satemisation.

Thirdly, he gives the odd pathway: /kʷ/ -> /kw/ (and the /w/ was lost before non-front vowels and /j/) -> /tsw/ -> /tš/.

It's not that odd; it's similar to what happened to /tw/ and partially to /k^w/ in Greek, only in Greek it went /t^sw/ > /t^s/ > /s:/ (or > /t:/ in Attic).

Well it is odd when there's the nice alternative of /kʷ/ -> /tšʷ/ -> /tš/, especially because he's suggesting that /kwi/ -> /tswi/ when /ki/ -> /ki/. I suppose what he's trying to do is reconcile the change with the fact that the palatals became /tš/ in front of /u ū w/ instead of /ts/.

Actually, the idea that Albanian distinguishes the three series is not new - I remember that being discussed in the literature when I went to an introductory course about Albanian historical linguistics during my studies, in the late eighties.

I only brought it up because I'd never seen it before and because Albanian is very often (in less linguisticky places e.g. Wikipedia) referred to as a satem language.

[hwhatting]

Actually, the idea that Albanian distinguishes the three series is not new - I remember that being discussed in the literature when I went to an introductory course about Albanian historical linguistics during my studies, in the late eighties.

I only brought it up because I'd never seen it before and because Albanian is very often (in less linguisticky places e.g. Wikipedia) referred to as a satem language.

That's true; actually, my remark was aimed more at publipis, who seemed to be under the impression that this was Orel's own idea.

[CatDoom]

As far as I know, it's hard to say anything "definite" about Thracian and Dacian. That they were IE languages that had undergone something resembling satemization (though if they're related to Albanian I guess they would be only sort-of satem languages) seems clear from the attested vocabulary, but positing a closer relationship to Balto-Slavic than that is a bit shaky considering the relatively few comparative sets available to work with. Heck, I think there's even some debate as to how closely the two languages were related to one another.

[hwhatting]

As far as I know, it's hard to say anything "definite" about Thracian and Dacian. That they were IE languages that had undergone something resembling satemization (though if they're related to Albanian I guess they would be only sort-of satem languages) seems clear from the attested vocabulary, but positing a closer relationship to Balto-Slavic than that is a bit shaky considering the relatively few comparative sets available to work with. Heck, I think there's even some debate as to how closely the two languages were related to one another.

IIRC, Thracian (I don't know about Dacian) seems to share with Balto-Slavic (and Germanic) the insertion of /t/ in the sequence PIE *-sr-, e.g. the river name Στρυμών (nowadays Struma) < PIE *srew- "flow". That alone doesn't establish a closer relationship, of course.