21 Jan 2008

I tripped over Pre-IE the other day

I tripped over this link as I was surfing the net lately and I was floored by the eery similarity with my own independent contemplations. It's Oswald Szemerenyi's book entitled Introduction to Indo-European Linguistics that I never got my hands on to read in a more physical format until I found it on the net. On page 112 (see here), he cites words from C. H. Borgstrøm who published his work on Pre-Indo-European (Pre-IE) in 1949. According to his theory, using the Pre-IE verb "to be" which he reconstructs as *häsä-, Borgstrøm proposes that the third person singular (3ps) was once *häsä-tä (> *est) "he/she/it is" and that the third person plural (3pp) was *häsä-nätä. It's not a mirror image of my theory since he proposed only one vowel in Pre-IE, namely as you probably guessed, but what it shares with mine which I arrived at independently and what was apparently already understood sixty years ago is that Proto-Indo-European (PIE) at some point in its past had dropped unstressed syllables.

My own search for a clear theory of Pre-IE began years ago with the observation that if PIE's accent alternates between one syllable and the next without an intuitive and common motivation, it must have been more regular in the past. I found out that most accent alternation patterns reconstructed for PIE (such as, *wódr̥ "water" vs. *udn-ós "of the water"[1]) could be easily regularized to the penultimate accent by presupposing that there was once a Pre-IE vowel positioned after case endings such as genitive *-ós. Along with the more obvious examples of an earlier drop of unstressed vowels (a.k.a syncope) in forms like *bʰr̥tós "carried", I became convinced that PIE must have once had a strong stress accent as in English or Italian and that this early syncope affected all unstressed syllables equally. (PIE itself is reconstructed with tonal accent, as in French. Tonal accents don't erode vowels the way stress accents do.) I normally capitalize the word as Syncope when I'm talking about this specific sound rule in Pre-IE, by the way. I was quickly struck by the idea that PIE's so-called "mobile" accent system was mobile (i.e. untuitive as to where accent is placed in a word) precisely because of Syncope. Where word-final vowels were lost, penultimate accent automatically became ultimate while those forms that did not lose a final-vowel would retain the old penultimate accent. In this way, Syncope in one fell swoop obscured the original, clockwork regularity of the accent. There's a lot more I need to say on Mid IE (MIE) stress accent, a lot more, but let's just say that somewhere down the line I realized that the accent in MIE could also occasionally occur on the antepenultimate syllable if an ending was formed in Old IE by the addition of an agglutinized element, as in the MIE 3ps and 3pp verbal endings with the attachment of deictic *ta (> PIE *to- "that") to the pre-existing Old IE 3ps ending *-a and 3pp ending *-éna. I now refer to this Mid IE rule of accentuation as QAR (Quasi-Penultimate Accent Rule).

I also reasoned that qualitative ablaut (i.e. the alternation between *e and *o in many verb forms), which is afterall the rather wild and unexplained alternation between phonetically polar opposites (between an unrounded, front vowel *e and a rounded, back vowel *o) , was probably once a simpler ablaut involving differences in height contrasts only. So at some point in Pre-IE we can reconstruct mid-central alternating with low-central *a. This is what Allan Bomhard has proposed, although he also believes that Indo-European's ablaut originates at the Nostratic stage preceding PIE by more than 10,000 years. To me, this added assumption is too fantastical to believe. At any rate, this signals that PIE underwent a recent vowel shift of *a to *o, particularly considering the otherwise scarcity of *a in PIE proper. From such a "height-contrasting ablaut" stage, the ablaut system could be further pushed back to a stage where there was no ablaut at all once the initial conditioning factor of ablaut (possibly vowel harmony) can be conclusively found, several millenia before PIE. So in a nutshell, this is why I reconstruct two vowels in Pre-IE, rather than one.

So back to Borgstrøm's 3ps *häsä-tä and 3pp *häsä-nätä, I had come to similar but more detailed conclusions, as you can see. I've reconstructed late Mid IE 3ps *ʔésatai and 3pp *ʔasénatai (n.b. the regular accent by way of the above-mentioned QAR). Unaccented *a was dropped in most circumstances in early Late IE via the Syncope rule, producing *ʔésti and *ʔsénti (traditionally written as *h₁esti and *h₁sénti) . The problem of the 1pp and 2pp forms that Szemerenyi uses to disprove Borgstrøm's proposal have no affect on my somewhat different explanation: MIE 1pp *ʔasména "we are" and 2pp *ʔasténa "you are". However, Szemerenyi admits that Borgstrøm's fatal flaw here comes from "the assumption of a stage with open syllables only".

NOTES
[1] The genitive form *udnós is reflected in Sanskrit udnás. Hittite witenas however reflects an alternative genitive case form *wedéns. Personally, I feel that forms with *wed- are in all likelihood the more conservative because of the same *o/*e alternation in other ancient paradigms like that of *pod- "foot" (nominative *pōds "foot" versus genitive *pedós "of the foot"). I would surmise that *udnós is the result of speakers generalizing the more prevailing pattern in PIE of stressed vowels reducing to zerograde when unaccented.

UPDATES
(Jan 24/08) I've decided to alter what I originally stated in footnote #1 above. After a good debate with Phoenix in the comments box on curiosities of Hittite spelling, I'll cave in and admit that Hittite genitive witenas "of water" is likelier to reflect a pronunciation such as /wɪténs/ and thus would be a reflex of *wedéns rather than *wednós as I originally stated (which I based in part on Sanskrit's reflex and partly on my theories concerning Syncope and QAR in Pre-IE). I'll assume for the moment at least that the Anatolian genitive form hasn't been affected by other case forms such as the locative form, *wedéni. I still have a funny suspicion that the underlying paradigm of "water" in PIE contains nomino-accusative *wódr̥, genitive *wedn-ós and locative *wedén-i.

12 comments:

  1. witenas doesn't actually point back to *wednas.

    The full paradigm of Hittite watar is as follows:

    Nom-Acc. wātar
    Gen. witenas
    Dat-loc. witēni
    all. wetena
    erg. wetinanza
    instr. witanta/witenit

    One can not assume that wi goes back to *we if there's a glyph for we (which there is).

    Kloekhorst assumes here that the i in 'wi' is a somehow reduced vowel /ɨ/. While the e seen in the Allative and Ergative are real e from Indo-European. Though of course the /-ants/ suffix is an Anatolian innovation.

    witenas thus points towards a Proto-Anatolian wténas. Maybe the w instead of a u is due to analogy of the rest of the paradigm. Then the fact that there is an -én- suffix with accent and ablaut there, seems to point towards a proterodynamic paradigm. Thus:
    Nom-acc. */wátr/
    gen. */uténs/ (or */wténs/)

    Actually the spelling wi-te-na-as could still represent /wténs/

    The fact that Hittite displays a proterodynamic paradigm while the rest of Indo-European displays hysterodynamic is also quite interesting.

    Okay I sort of went of in a different direction that my point. My point was that the wi- in the genitive absolutely can not come from *we. Can't really come from *u- either, but the /wɨ/ can be explained through analogy.

    I have quite a few questions about your ideas on Syncope, but I'll leave them until they're a bit more structured in my mind :D

    ReplyDelete
  2. Phoenix: "One can not assume that wi goes back to *we if there's a glyph for we (which there is)."

    Excellent, a fight is brewing! I beg to differ. See Fortson, Indo-European Language and Culture: An Introduction (2002), p.111 and Trask, Dictionary of Historical and Comparative Linguistics (2000), p.5. Both refer to an acrostatic paradigm *wódr̥/*wédn-. I prefer to think of it as showing an accent alternating between stem and suffix myself. As I said, such a pattern follows the *o/*e ablaut of other attested paradigms. Finally, in Hittite, unfortunately for you, wetenas is also attested.

    Phoenix: "Kloekhorst assumes here that the i in 'wi' is a somehow reduced vowel /ɨ/."

    "Assumes" is a touchy word with me. Assumes based on what exactly? Certainly not PIE phonotactics as is normally understood and this theory seems to be in blunt ignorance of overall variations in Hittite spelling particularly concerning the representation of e and i (gen. witenas/wetenas).I hardly think that kissarit "by hand" and kisseran "hand (acc.)" are from **ǵʰsor- while nominative kessar is from *ǵʰésr̥. Entirely ad hoc. Can you explain how Kloekhorst is able to ignore these important details?

    ReplyDelete
  3. I really wish I could show you the grammatical/phonological introduction of his up and coming Hittite etymological dictionary.

    The good part is, I'll be digitalising it shortly. As soon as it's up and running I'll post about it on my blog ;)

    Back to the theory.

    ki/ke are not distinguished in writing unless plene spelling is used. The plene spelling to determine the e in the nominative is often seen. Plene spelling in oblique cases to determine the i is not attested.

    wi/we are always distinguished, there's absolutely no ambiguity, and it's not common for them to be mixed up, except in places where we expect epenthetic vowels. The fact that there is a change of wi/we in the Genitive is seen as a confirmation that it is in fact an epenthetic vowel.

    As a side note: There's scholars that theorise that there's no i/e distinction at all, I think this is completely implausible. It's just not likely that eszi was in free variation with iszi when iszi is almost never attested in any written form for example.

    Now for our hands:
    The reconstruction Kloekhorst gives in his dictionary is the following.
    PIE *ǵʰés-r / *ǵʰs-ér-m / *ǵʰs-r-ós

    These are based on the following attested forms:
    ke-es-sar /kéSr/ ki-is-se-ra-an /kɨSéran/ and ki-is-ra-as /kɨSrás/

    There's quite a few variants to these forms, but these seem to be the oldest, and at least all originate from Old Hittite.

    Some more information on the epenthetic vowels:
    Kloekhorst proposes three types of epenthetic vowels.

    1. In clusters that have the shape *CRC always have an epenthetic vowel spelled -a-. Thus: /CaRC/. Kloekhorst continues to argue that this can never be a true /a/ while it is often reconstructed as such, but rahter an epenthetic vowel that was phonetically [ə] or [ɐ].

    He bases this on the verb ārs-(zi)/ars-. This ablaut can't be ā/a because we'd expect an ablaut between *h1ers- and *h1rs-. Thus phonemically the logical stems in Hittite are /ʔars/ and /ʔrs/. He then concludes that phonetically these stems come out as [ʔars] and [ʔərs] respectively. The plene spelling fo the first stem is then only a way of indicating a glottal stop + vowel a. While the other one has a single spelling jsut pointing out a "a like sound" or just the glottal stop.

    2. In clusters that have an -s- and stops or laryngeals we find epenthetic vowels spelled with either -e- or -i-.
    *-TsK- > Hitt. -ze/ik(k)- *-PsK- > Hitt. -p(p)e/isk(k)-, *KsC- > Hitt -k(k)e/isC-, -*Vh2s > Hitt. Vhhe/is.

    And there's quite a few more like this. Also *wT- clusters seem to fall in this category. As there's no consistency in which vowel is spelled either -i- or -e- Kloekhorst assumes that its phonetic value must have been either [ɨ] or [ɘ].

    3. The last one he mentions will hopefully not bring up any disagreeances, since this one is by far the most obvious. In intial cluster *sT (where T is any stop or H) in writing an i- is conistently written. *stu- > Hitt. istu- *sh2oi- > Hitt. ishai- etc.

    This vowel can not be a true [i] because it doesn't partake in the New Hittite vowel lowering of Old Hittite /i/ to /e/ before /s/, /n/ and /m/ abd cluster containing /H/. It can't be of the same phonetic value as the i/e epenthetic vowel either because it's never written with e-. Therefore one might imagine a phonetic [ɪ].

    Because these three vowels occur in their own unique environments and never overlap, we could probably assume an epenthetic vowel phoneme /ə/ with different reflexes in different environments.

    Thanks for trying to do everything you can to make sure that I am not just copying what I've been told, but that I actually have to search for arguments to believe what I do ;)

    ReplyDelete
  4. Wow, excellent justification! With such a thorough account, I hate to shoot you down so bluntly. But I will anyway because Logic has no mercy :)

    Primo: In PIE, the initial sequence of sonorant + consonant is treated the same consistently. Sonorants are always reduced to their vocalic variants, period. Hence *n̥s "us" which would otherwise be humanly impossible to pronounce if either *n or *s weren't phonetically vocalic. Ergo, we find *ud-, never **wd-... ever!

    Secundo: The theorist is shamelessly defying not only PIE's rules but the rules of human language itself. Look up SSP (Sonority Sequencing Principle). Notice that PIE and the data justifying its reconstruction are in line with SSP. In this case, being that *w (sonorant) trumps *d (stop) on the sonorancy scale, the two in that order cannot consitute a legal syllable onset. Only *dw- or *ud- is legal, never **wd-.

    Since **wd- is theoretically impossible, it's *NOT* an epenthetic vowel. Thus, the claim, as presented, is conclusively falsified.

    ... Although perhaps if you stop talking about "epenthetic vowels" in illegal onsets, we could discuss more credible reasons behind Hittite's spelling variations :)

    ReplyDelete
  5. Well the epenthetic vowel might not even be the most correct name for it, since it clearly is a phoneme. Maybe even two (I left that out for simplification).

    I am aware of the Indo-European impossibility of *wd. And even in Hittite the reflex of *ud- should just be *ut.

    Therefore I believe the original form must have been **uténs/uténas. But due to the Nom-acc stem wátr, it has tried to copy the same consonant shape while retaining the ablaut.

    I'm saying that's what it looks like to me. Obviously this is difficult to justify if you don't see it back in other ablauting paradigms, but even in Hittite these have already become incredibly rare.

    but the oblique cases having the shape witen- just can't be explained as *wedn-. The second vowel is clearly an accented /e/, as it would otherwise never be written with plene spelling.

    so we'd at least have a shape *wedén-. The fact that we is interchanged with wi and is never written in plene spelling is a indication that the value of this vowel is suspicious. We can't say it's proof that the phonemic form is /wɨtén-/ but from a Indo-European point of view this would still be more plausible than either /witén-/ or /wetén-/.

    We would also expects in text that do use plene spelling for the second /e/, to also use plene spelling for the first **/e/ if this were a true /e/. Though of course we can not fully rely on the consistency of spelling of Hittite scribes because they often had funky moods. I'd say it depends on the amount of attestation (of which I do not have the figures).

    wi/we is not an indistinguishable pair of glyphs. but wi simply can't be from Indo-European *ue, in this context (before /t/). /e/ becomes /i/ before /nK/, and that's, as far as I know, the only context where this happens.

    /i/ as pointed out before can become /e/ in several contexts. But once again non of these are before /t/.

    There's then three conclusions we could draw:
    1. There's some kind of vowel in play that is somewhere in between e and i.
    2. The genitive has two stems which are freely mixed at the whim of the speaker.
    3. i and e aren't dinstinguished, which in other words is clearly wrong.

    I personally find the first conclusion the most plausible.

    But my main point of all this is: even if wi/we was /e/, one still can't ignore the second e which is clearly a real e. Otherwise we'd expect the spelling:
    wi-it-na-as or we-et-na-as. We don't, therefore the e is not an e to be ignored. Especially not when you look at the Dat-Locative where it appears, and is accented (and therefore is long).

    ReplyDelete
  6. Phoenix: "Well the epenthetic vowel might not even be the most correct name for it,[...]."

    But now you're changing your story since you stated originally: "Actually the spelling wi-te-na-as could still represent /wténs/
    ". Clearly now this is false. Since we both agree that **wd- is a crosslinguistically illegal onset that would never have occured in PIE nor Proto-Anatolian and that *ud- is likewise impossible given the Hittite data (and Luwian), then it's not only logically inevitable that the genitive did indeed retain *wVd- in the first syllable
    (note also Gothic watō/watins) but that other forms that you claim show this same "epenthetic vowel" fall into doubt. It really sounds off-the-wall.

    The paradigm that you give for *ǵʰésr- is talked about in Fortson's above-mentioned book on page 112 and there is no mention of a bizarre alternant **ǵʰsér- as you claim, even despite his pointing directly to the Hittite form kissaras in the same paragraph.

    Phoenix: "Otherwise we'd expect the spelling: wi-it-na-as or we-et-na-as. We don't, therefore the e is not an e to be ignored."

    I never said it should be, but in the context of this blog entry, I was focussed on PIE and Pre-IE, not Hittite or any specific details about paradigmatic leveling in the Anatolian branch.

    It seems to me that any second *e in the genitive must have been imported from the locative. Sure, maybe it's true that Hittite witenas reflects *wedéns but then this form would still have to be the product of paradigmatic leveling in order to explain the curious placement of accent and subsequent reduction of genitive *-ós. Whatever the precise details, it hardly reflects **wdens.

    Maybe you'll understand better my point of view when I talk about the origin of heteroclitic stems in Mid IE, word-final denasalization and QAR :)

    ReplyDelete
  7. The paradigm that you give for *ǵʰésr- is talked about in Fortson's above-mentioned book on page 112 and there is no mention of a bizarre alternant **ǵʰsér- as you claim, even despite his pointing directly to the Hittite form kissaras in the same paragraph.

    I don't see what's so bizarre about the form *ǵʰs-ér-m. It's the exact same form we see in the paradigm of *ph2t-ēr
    (*ph2t-ér-m)
    (*ph2t-r-ós)

    the Nominative kissaras is clearly a New Hittite form. Just the Nominative -s should be enough of an indication for that. Older text don't have this -s; Just -r. This i found is this Nominative then, can easily be taken by analogy from the genitive.
    It's not at all unthinkable that the i in the genetive and accusative was generalised to the nominative.

    And as for the epenthetic vowel that isn't one in wVd-.

    It is not an epenthetic vowel that developed naturally, but looking at the way it behaves, I'm quite sure that the phonetic value of the vowel in between w and d should have the same phonetic value as the type 2 epenthetic vowel.

    ReplyDelete
  8. Phoenix: "It's the exact same form we see in the paradigm of *ph2t-ēr"

    No, it's not because the first syllable is full-grade and accented in "hand" but unaccented and zero-grade in "father". They are opposite. Once you follow the link I already supplied above, you'll see a published explanation of the word that doesn't acknowledge your claim.

    Phoenix: "It's not at all unthinkable that the i in the genetive and accusative was generalised to the nominative."

    All you're doing now is offering ad hoc solutions here and there to legitimize an epenthetic vowel that I've already disproved. You need a new theory to account for the spelling differences that doesn't require so many assumptions.

    ReplyDelete
  9. No, it's not because the first syllable is full-grade and accented in "hand" but unaccented and zero-grade in "father". They are opposite. Once you follow the link I already supplied above, you'll see a published explanation of the word that doesn't acknowledge your claim.


    I feel really stupid for asking this, but I really do not see the text giving any other alternative, it purely gives the nominative *ǵʰésor. which could very well be the origin of kessar. Just not kissar. But it definitely doesn't say anything about the accusative and whether it would be *ǵʰsér-m.

    You're right that there's a difference between ph2tēr being different, in the nominative. The rest of the paradigm is similar though. A Noun that has the accent shift between Root-Stem Suffix-Case ending in Nom-Acc-Gen is a holodynamic paradigm and not unattested. Something similar happens with *dʰéǵʰ-ōm As can be seen here.

    I personally do not feel that the epenthetic vowels are so highly assumptive as you claim them to be. A certain type of vowel consistently appears in predictable environments, the only thing that's in predictable is how it's written either e/i. Though I agree that the 'wt' environment is odd, or even impossible, just because the epenthetic vowel is to be dismissed there, I see no reason why it should be dismissed all together also in kessar. Kessar has an epenthetic vowel in a position that is phonetically viable, also from a phonotactic point of view in Indo-European.

    If you're truly set on finding another explanation though, feel free to have a go at it. I am just not sure what other explanation there must be. We simply cannot assume that i and e weren't contrastive in Hittite, there's too much proof against this.

    As a side note: It might be interesting to know that in the earliest Hittite Kessar still functions as Neuter, where Kessar is simply the nom-acc. Later in Hittite both the Neuter form and a Commune form starts to get used. The commune form having a -s suffix to clearly mark the commune nominative. We clearly perceive a time where both forms are perfectly legal to use though, the semantic difference isn't clear.

    All you're doing now is offering ad hoc solutions here and there to legitimize an epenthetic vowel that I've already disproved. You need a new theory to account for the spelling differences that doesn't require so many assumptions.

    kessar is one of the few clearly ablauting nouns in Hittite. If suddenly an /i/ spreads to the nominative, that in earlier forms was only seen in oblique cases, I am inclined to believe that this analogy. Especially considering this is a later Hittite form. I know of no Old Hittite text that write ki/e-es-sa-ar as ki/e-is-sa-ar. There's should be no reason to doubt the original /e/, and the disappearance of it, when it started to be consistently written with ki/e-is.

    Maybe, I should write a blog entry soon on Hittite phonology. It might be easier to organise my thoughts and points like that. Because this is turning in an: is not!-is too! discussion. :D (still, feel free to reply to this comment though, if neither of our convincing powers will bring the other one to change his mind, I'll write an entry in blog on it ;), references to texts, articles and everything.)

    ReplyDelete
  10. Okay, as you can see, I enjoy a good hardcore debate :)

    Phoenix: "Though I agree that the 'wt' environment is odd, or even impossible, just because the epenthetic vowel is to be dismissed there, I see no reason why it should be dismissed all together also in kessar."

    Think about this. If it's dismissable in the case of watar, then there's a vowel running around in Hittite that is conclusively *not* caused by this epenthentic vowel. In other words, epenthesis is likely to be unnecessary. Unlike a totally unproven epenthesis, rules concerning sonorancy prove its absence in at least one case. The score is: Epenthesis - 0, Absence - 1. The onus is on you to prove its presence conclusively, to prove that this is not just an ad hoc rule.

    Let me try a competing hypothesis just to drive you mental, hehe. What if the e/i spelling variation is truly a reflection of a special vowel as you say? I would like to claim for fun that it is a reflection of unstressed *e. Thus witenas/wetenas = *wedén(a)s. Also eshar/ishar = *eshár, kissaras = *kesárs and kissaran = *kesáran. The latter paradigm of "hand" might show leveling of the vowel and accent, hence the variation of the first vowel in the nominative, reflecting perhaps an alternation between accent on first syllable and accent on second (matching the accusative's accent position). This would also explain why eszi doesn't show alternation since it's *észi with accented *e and of course asanzi reflects zerograde PIE *h1s-, not *h1es- afterall.

    Any loopholes I missed?

    ReplyDelete
  11. P.S. Absolutely write a blog entry on Hittite! This stuff is great as a mindteaser to ponder and inspire everyone to research. :)

    ReplyDelete
  12. Any loopholes I missed?

    That seems to cover quite well, at least, for later Hittite, I'll most definitely look into it. I'll also give Kloekhorst an e-mail ask him how he's going to talk himself out of that one ;)

    And, of course, I'll be updating my blog soon. It's going to be a big one. :D

    ReplyDelete