11 Oct 2007

Reinterpreting the Proto-Indo-European velar series

I've pondered for years now about Indo-European (IE) phonology and the problems associated with it. It annoys me that IEists don't finally address them by updating their transcription system. One of these major issues with the sound system involves the so-called velar stops, that is, the reconstructed sounds *ḱ, *k, , *g, *ǵʰ and *gʰ.

You may wonder what's wrong with them? Afterall, satem dialects show us clearly a distinction between *ḱ and *k. Of course, we should have no issue towards this contrast at all. However the problem lies with how certain we are that *ḱ is phonetically realized as a palatalized consonant in IE. The nitty gritty of it is that we only assume that *ḱ is palatal since this is how it ends up in satem dialects where we find the palatal affricate in its place. So it's long been believed that since IE *eḱwos becomes Early Indo-Iranian *ećwos then it follows that IE itself had palatal consonants. Oh dear, what a careless leap of logic! The evidence from satem dialects merely shows conclusively that early Post-IE dialects had palatal consonants. Centum dialects lacked palatalization altogether. But how can we explain IE without palatalized sounds?

To explain IE without palatalized stops we need to first explain what we plan to do to fix IE. This is how we should update IE's sound inventory for the 21st century:


  • Palatalized stops are to be reinterpreted as plain stops:
    *ḱ, , *ǵʰ -> *k, *g, *gʰ
  • Plain stops are to be reinterpreted as uvular stops:
    *k, *g, *gʰ -> *q, , *ɢʰ
Why that's blasphemy! Delicious, isn't it? Now get ready. Here comes the reasoning behind it.

We know that pronouns and numerals contain the so-called palatalized stops exclusively and yet this is completely counter to the principle of phonological markedness. We expect simpler sounds to be used for such common words and yet clearly IE is a theoretical maverick: *eǵoh2 'I', *ḱo- 'this', *sweḱs 'six', *déḱm 'ten'. This in itself is clear proof that these sounds must be interpreted as plain, not marked with added palatalization.

There is also the consideration that the sequences *ke or *ek are rarely reconstructed for IE. By acknowledging that traditionally transcribed *k is in fact marked and pronounced further back in the mouth, and further by pairing it with *h₂, we realize that the reason for the lack of these sequences is because *q, and *ɢʰ colour vowels just like *h₂ does. So whenever we see *ka or *ak reconstructed, we should remember that they are in fact *qe and *eq (e.g. *kap- 'to seize' is to be understood as *qep-). I will go out on a limb and bet that the few words that are reconstructed with these sequences of *ke or *ek are falsely reconstructed, either because they are based on false evidence, because the proof points rather to its "palatal" counterpart, or because the vowel in question should be long (n.b. that long vowels resist colouring normally caused by neighbouring *h₂).

When we start pondering the effects of this reinterpretation, we begin to see a different story concerning the development of satem dialects unfold. We then realize that the Satem dialect area was the innovator, pushing the two stops *k and *q frontward in the mouth. Hence briefly, dialectal *ḱ and *k spread across a portion of the IE-speaking area where the rest of the dialects kept original *k and *q. This regional isogloss was now the seed for satem dialects like Indo-Iranian and Balto-Slavic. However, palatal velar stops are unstable and quickly turn to affricates, so it wouldn't have been long before and *k were heard throughout Satem IE as became the norm in later Indo-Iranian.

10 comments:

  1. I had to let this theory sink in for a while, before I felt confident to say anything about it. Now I have, and I'm moderately sceptical.

    One of your arguments has to do with the phonological markedness. I absolutely agree that you expect more 'natural' consonants in basic words. Nevertheless Indo-Iranian apparently did not have a problem with this, and used more marked phonemes for very basic words. If this is possible in Indo-Iranian, why would it be a problem to the Indo-Europeans?

    Nevertheless, your theory is quite attractive, k-q series seems to match better with Proto-Finno-ugric, and, don't shoot me, is in my views a good candidate to be combined with Proto-Indo-European in a larger macro-family. But more about that later.

    What surprises me though is your assumption of "voiceless-voiced-voiced aspirated", this is also quite a strong violation of what typologists would find a natural system. Nevertheless, Sanskrit is a clear proof that is possible, although it does have the voiceless aspirates too, they're incredibly rare.

    Over at my university though, we support the preglottalisation theory. And that in fact the stops were not *k - *g - *gʰ but rather *k *ˀk *kʰ where the last two were 'Lenis' and tended to become voiced intervocally, but not as a phonetic distinctive feature.

    So the second in the series wasn't voiced but preglottalised, this is based on vowel lengthening seen throughout Latin and Balto-Slavic before voiced consonants, as though there was a non-colouring laryngeal, thus, preglottalisation of the 'voiced' consonant.

    This isn't too damaging for the Uvular theory, except that a preglottalised q might be a challenge to pronounce, but I believe several Native-American languages don't seem to have any problem with it, so I won't have a problem with it either.

    Your uvular a-coloring theory does have enormous implications. If this theory theory is correct, a fair amount of *h2's could be thrown out of the window, right?

    Now for the uvular fricatives giving these different colourings, I'm not at all opposed. Since it'd be silly, my native Dutch /r/ is [ʀ] or [ʁ] and clearly has vowel lowering qualities in coda position.

    I've somewhere lost track of my own story, but I hope you can pick up enough of my terribly disorganised argumentation to understand some of my scepticism. ;)

    ReplyDelete
  2. Phoenix: "Nevertheless Indo-Iranian apparently did not have a problem with this, and used more marked phonemes for very basic words."

    Actually, not quite. PIE *ḱ already became in Proto-Indo-Iranian (PIIr). Finno-Ugric loans show us exactly what sounds PIIr had at the time. Palatalized *ḱ is unstable, both phonetically and within this sound system. It's brief occurence isn't really an issue since I'm merely talking about strong language tendencies, not absolutes. But in fact, this idea provides a perfect motive for the merger of labiovelar *kʷ with the newly fronted plain velar *k of PIIr (< IE *q). That is, any imbalances in phoneme frequency and markedness caused by fronting would be quickly resolved by a new abundance of fresh plain velar stops derived from PIE *kʷ. In other words, this merger made "plain k" more common than its marked counterpart again, thereby reestablishing harmony in the cosmos.

    In the traditional view, palatalized and plain stops merge to plain in centum dialects but there's no clear reason why there were these unstable palatalized stops in the beginning! Nor is it clear what on earth would have palatalized them.

    The current view is simply wrong because they can't account, for example, as to why clusters of the type *ḱC- are more common than *kC-. However in this reinterpretation, everything is clearer because we see that *kC- is more common than *qC-, just as we should expect.

    Phoenix: "And that in fact the stops were not *k - *g - *gʰ but rather *k *ˀk *kʰ"

    I would rather say that it's just a three-way contrast involving voicing onset. I think of *g as an "English-like" semi-voiced /g/ while *gʰ is the "French-like" fully voiced /g/. In such a system, aspiration is not a key feature, just voice. Luckily I'm bilingual in both English and French and can distinguish the two g's easily, so I know it's possible.

    Phoenix: "This isn't too damaging for the Uvular theory, except that a preglottalised q might be a challenge to pronounce"

    On the other hand, if the contrasts are to do with position of voicing onset, as I state above, then traditional *g is to be rewritten as and pronounced as a semi-voiced uvular stop [q͡ɢ] (<- I hope you can see these crazy IPA symbols).

    Phoenix: "Your uvular a-coloring theory does have enormous implications. If this theory theory is correct, a fair amount of *h2's could be thrown out of the window, right?"

    I'm not sure what you mean. I'm simply suggesting that these sounds can colour *e to *a like *h2 because the underlying phonetic quality that colours vowels in IE is [+uvular]. So rather, we would see a few *a's thrown out the window (and replaced with *e) ;)

    Phoenix: "Now for the uvular fricatives giving these different colourings, I'm not at all opposed."

    Uvular *fricatives*? Did I miss something? I was talking about uvular stops.

    ReplyDelete
  3. The uvular fricative part, was about your comments on h2 and proto-semitic ayin in your previous post. For some kind of reason I somehow assumed it was in this post. sorry for bringing up unnecessary confusion.

    And yes, I see the funky IPA, albeit, a bit distorted. Which font do you use for IPA?

    The voiced - semi-voiced - fully voiced idea is definitely a more 'natural' system, reminds me a bit of lenis fortis extra fortis contrast in Estonian (written as g k kk).

    But, does this mean you completely abolish the preglottalisation theory?

    Obviously, voicing of consonants can lengthen the preceding vowel, which could explain why it happens. But in your system, it doesn't explain why it DOES happen in *g but not in *gʰ.

    ReplyDelete
  4. Phoenix: "The uvular fricative part, was about your comments on h2 and proto-semitic ayin in your previous post."

    Oh right, yes. However the Semitic fricative was not uvular. It was a pharyngeal, further back in the mouth, and the claim I made in Mid Indo-European, Semitic and Neolithic was that the pre-IE stage of MIE only had velar fricatives. In fact, I don't even think that these uvular sounds (aka *h2 and the "plain" series) were distinctive until Late IE.

    Phoenix: "But, does this mean you completely abolish the preglottalisation theory?"

    That's great. You're definitely observant ;) But actually I would describe myself as a "Glottalic Theory sympathist". What that means is that I think that the arguments in favour of ancient ejectives in IE's past (but not pre-glottalization) are sound. However, I strongly do not believe that IE itself had ejectives. Rather, I claim that MIE had them.

    To be honest, I was shocked (and maybe even disturbed) that they are teaching you that *bʰ was an unvoiced aspirate. This sounds very fringe.

    Phoenix: "And yes, I see the funky IPA, albeit, a bit distorted. Which font do you use for IPA?"

    As far as I know, I don't have any control over the font selection in this commentbox. Maybe I need to dig deeper into the convoluted Blogger system. I was assembling the IPA from this cool interface and then copy-and-pasting.

    ReplyDelete
  5. Hey Glen! It's your old buddy Rob from Cybalist. I've been following your blog pretty religiously, and now I actually have something to comment on.

    Lately I've been reading about vertical vowel systems. These are well-represented in the Northwest Caucasian languages, among others. Anyways, it's led me to start re-evaluating the development of Indo-European.

    It seems that IE's vowel system has many indications of having descended from an earlier vertical vowel system. The big piece of evidence is the labiovelar series. I had earlier proposed that this series came from an earlier uvular series, but now I think that is mistaken. There would have to be an intermediate stage with a labiouvular series. Phonologically, this is very unlikely at best. It makes much more phonological sense for the labiovelars to have arisen from earlier sequences of [velar + back vowel].

    On another note, I'd like to challenge your thesis regarding the IE velar series. At least one language, namely Ubykh, has palatalized and labialized velars at much higher frequencies than plain velars. I also seem to recall similar situations among some Altaic languages, but I can't be sure at the moment. So while markedness theory says that plain velars should be the most common, that isn't *always* the case.

    Hope you enjoy this food for thought!

    - Rob

    ReplyDelete
  6. Hi Rob! Welcome! Of course I remember you.

    Rob: "It seems that IE's vowel system has many indications of having descended from an earlier vertical vowel system."

    Yes, precisely! However, I've been saying this for quite a while now (read my message from Dec 16, 2003: Re: [Nostratic-L] Pronouns and Nostratic). This view has also been in print since at least a decade (Allan Bomhard, Indo-European and the Nostratic Hypothesis, 1996). I commend Bomhard not so much for his tentative reconstructions but for his brilliant insights into pre-IE. He suggests intimate pre-IE contact with Abkhaz-Adyghe (aka. "North-West Caucasian") and it has since convinced me. It's then natural to suspect that both the vowel systems of AbAd and of IE may have evolved in tandem by areal influence. If so, I estimate that this contact would have occured between about 9,000 and 8000 BCE as both languages were gradually moving westward from Central Asia towards the Black Sea. I'm sure I've stated this before on the forums years ago.

    Rob: "On another note, I'd like to challenge your thesis regarding the IE velar series. At least one language, namely Ubykh, has palatalized and labialized velars at much higher frequencies than plain velars."

    Excellent! I love challenge. Your example of Ubykh is important to know but it's not relevant. I've already discussed how rare possibilities don't make things probable (see How NOT to reconstruct a protolanguage). The question you should be asking is not whether something COULD exist but whether it is NECESSARY, by way of Occam's Razor. If it's not necessary, you have no logical reason to reconstruct it. It's the ol' K.I.S.S. principle: "Keep It Simple, Stupid."

    So in the case of IE, it's in fact unnecessary in light of the reinterpretation that I and others suggest (i.e. "palatal" stops are plain and "plain" stops are vowel-colouring uvulars). Furthermore, I think in light of Semitic loans, there is indication that these stops were indeed plain and that palatalization was not a phonemic feature of Mid IE (c. 6000-5000 BCE) or Late IE (c.5000-4000 BCE). The pre-IE adoption of *sweḱs from Semitic *šidθu can only make full sense when we understand that Semitic *d was alveolar (like in English) while IE *d was dental (like in French). As a result, pre-IE speakers understandably misheard an alveolar for a velar which is produced near the same region, hence *šidθu (pronounced ['ʃʷɪd͡tθ̙ʊ]) became at first MIE *sʷeksa (pronounced ['sʷeks̙ə]) evolving to IE *sweks (traditionally written *sweḱs) with a plain velar stop throughout.

    Initial consonant clusters in IE also overwhelmingly contain the "palatal" variant instead of the plain, and so the traditional view when taking into account all of the evidence defies credulity. Occam's Razor will prevail regardless of Ubykh phonology and other interesting but rare exotica.

    ReplyDelete
  7. Maybe one last note: Upon consideration however, I admit that when the vowel system first moved towards centralization in pre-IE (in a stage I now call "Indo-Aegean"), it's possible for palatalization to have become a phonemic feature of consonants at first. However, if this were true, we'd still have to theorize that palatal and plain stops merged to plain, leaving labialized phonemes intact by the neolithic (because of Semitic loans in Mid IE). As I said before however, I try to follow Occam's Razor and so I have yet to see a necessity beyond reasonable doubt of palatalized phonemes contrasting plain ones in pre-IE phonology, despite it's possibility.

    ReplyDelete
  8. I'm sorry, just a self-correction: I previously stated "defies credulity". Hahaha, I'm such a twit. I meant to say "defies credibility". I don't know why my swiss-cheese brain mixed these two words up but I blame my coffee which was surely decaffeinated against my will.

    ReplyDelete
  9. There's one thing I don't get. In northern German the coronals are apical-alveolar like in English, in southern German they are, to be fully pedantic, laminal denti-alveolar like in French, Russian and so on. (The line runs perhaps between Low and Middle German, though I have no real idea.) I'm Austrian, so my native coronals are "dental". I had 10 years of English at school. I was able to tell that something sounded a bit wrong about my coronals, but I didn't manage to figure out what. Only later a linguist told me that the English ones are apical-alveolar. A lightbulb lit up above my head, and since then, my English coronals sound convincing. My point is that... OK, fine, it's just an argument from personal incredulity, but I can't for the life of me imagine how anyone (outside of Hawai'i, I suppose) could ever hear apical alveolars as velars.

    The rest sounds interesting; I certainly need to read more of your blog. I'd just like to mention that there are languages that have palatalized, plain, and labialized velars but only plain labials and coronals. (I'll look for references tomorrow.) Furthermore, the fact that the plain ones are much rarer than the palatalized or the labialized ones in the conventional reconstruction of PIE is not surprising if we assume that the palatalization and the labialization as well as the funky vowel system come, AbAd-style (...what a coincidence...), from the transfer of frontness and roundedness from vowels to consonants so that plain vowels are only found in front of former central vowels (which most likely means only /a/). Indeed, in AbAd, the plain velars are rare enough that they merged into the palatalized ones (not the other way around!) in Ubykh, so that, in the attested stage of Ubykh, plain velars only occurred in Adyghe and Turkish loans. (Hey, what are three more consonant phonemes when you already have 80.) Clearly, however, your explanation is very parsimonious for the PIE clusters, where other hypotheses have to make the ad hoc assumption that some vowel was lost!

    OK, so "I'd just like to mention that" is misleading, and I got carried away... :-]

    Oh yeah, and Bomhard writes in his 2007 book that he thinks the Proto-Semitic interdentals were actually something else. I forgot what, though. I'll look it up tomorrow.

    ReplyDelete
  10. David Marjanović said: "I was able to tell that something sounded a bit wrong about my coronals, but I didn't manage to figure out what."

    Great story. My parents who are 'clinically monolingual' anglophones decided to enroll me in the French Immersion program and so I know the eureka one gets from realizing how to mimick another language's strange sounds. I recently was elated when I finally found out how to mimick the strange "v"-like allophone of initial "w" produced by some Mandarin speakers. It's a labiodental approximant /ʋ/, something that doesn't exist in my dialect of English at all. Teehee!

    David Marjanović said: "[...] but I can't for the life of me imagine how anyone (outside of Hawai'i, I suppose) could ever hear apical alveolars as velars."

    I'm certainly not suggesting this for every case of Semitic loans in Mid IE. It's only to explain the unexpected medial cluster in this one case and to use its oddities to extrapolate specifics on the phonetics of the two languages.

    If you think this alteration is bizarre, you should ponder on the city name "Carthage". In Greek, it ended up as Καρχηδών (Karchedón) and yet it shows up as Carthago in Latin and *Carthaza in Etruscan. We know that it derives from Phoenician Qart Ḥadašt 'New City'. Obviously there is only one explanation for the velar aspirate 'ch' in Greek's reflex and you're not going to like it. :) This is a perfect parallel to my Semitic-IE comparison. It's not as unreasonable as you assume. But again, I stress, that this change was likely sporadic due to occasional faulty transmission of a word across language boundaries.

    David Marjanović said: "I'd just like to mention that there are languages that have palatalized, plain, and labialized velars but only plain labials and coronals."

    Erh, I'm assuming that you possibly mean to say "... but not just plain labials and coronals"? But then, this would be false because of languages like Klallam (spoken on the West Coast of Canada) that only treat labialization, not palatalization, as a distinctive phonemic feature.

    David Marjanović said: "Furthermore, the fact that the plain ones are much rarer than the palatalized or the labialized ones in the conventional reconstruction of PIE is not surprising if we assume that the palatalization and the labialization as well as the funky vowel system come, AbAd-style (...what a coincidence...), from the transfer of frontness and roundedness from vowels to consonants so that plain vowels are only found in front of former central vowels (which most likely means only /a/)."

    Nothing changes the fact that the most common words and morphemes in PIE use palatal stops, not plain, and this is indeed a bizarre thing that can be cured by "making the switch".

    David Marjanović said: "Oh yeah, and Bomhard writes in his 2007 book that he thinks the Proto-Semitic interdentals were actually something else."

    Ah yes... :( That would probably be lateral affricate? Unfortunately, Bomhard is the same person that mistakes a loan for a Nostratic root. Consult his improbable Nostratic reconstruction, *sʷak[ʰ]sʷ- 'six'. Bomhard is what I would call a "big-picture man". He has some great perspectives on Proto-Nostratic as a whole, but tends to make mistakes concerning the details, such as here.

    ReplyDelete