7 Jan 2008

Markedness and the uvular proposal in PIE

And now I will confess some important technical issues concerning the aforementioned uvular proposal for Proto-Indo-European (PIE). I still think that reinterpreting the non-palatal series of stops in traditional PIE reconstruction as uvulars is heading in the right direction. In this proposal, all we're really doing is pushing the articulation of all of the dorsal stops forward. By doing this, the exceedingly common palatal consonants thereby become plain and unmarked while the relatively much rarer "plain" consonants become uvular and marked. This solves markedness issues nicely by reversing which series is marked and which is unmarked with the least amount of effort. Yet...

There are a few problems and some extra ones that have been noticed by one observant commenter, Phoenix. Phoenix has commented in my last article that languages with uvular fricatives without velar ones are typologically unusual. And he's correct[1]. I've also noted that a three-way contrast between uvulars that this proposal would require (namely *q, *q:[2] and for traditionally notated *k, *g and *gʰ, respectively) would also be quite rare. So it seems that the old, pre-Nietschean saying might be true: Damned if we do and damned if we don't.

But I'm not giving up on this brilliant uvular idea without a fight! It just makes too much sense and I think that getting rid of a large-scale oddness in phonemic distribution in favour of an odd typology with normal distribution between marked and unmarked phonemes is a positive trade-off. There are languages like French that lack a velar fricative but contain a uvular one instead. In French, the uvular trill as a favoured pronunciation of "r" sprang up recently in 17th-century Paris[3] and this trend towards uvular-r, whether trill or fricative, gained popularity ever since, even in neighbouring languages like German, Dutch and Norwegian! If uvulars are really so odd in the grand scheme of things, it's rather amazing that this wave occurred. In Abkhaz (a language spoken in the Caucasus), there is a complex uvular stop series of qʲʼ, and qʷʼ without any non-glottalized counterparts! So we must realize that typological rarity is sometimes a matter of perspective. One must be reminded that areal environment often has a way of "concentrating oddities" in a single location and time as well[4].

Speaking of areal diffusion of curious phonological features, languages throughout the Caucasus mountains are known in fact to have these so-called "odd" uvular stops. For example, Lezghian of the Nakh-Daghestanian family contrasts aspirated uvular (), non-aspirated uvular (q) and ejective uvular () stops[5]. This is a 3-way contrast reminiscent of what I've suggested for an updated PIE. So that being said, is it too much of a stretch to wonder if perhaps pre-IE was inspired by these languages from the neighbouring Caucasus to adopt uvular phonemes?

I also find it interesting that, after I've suggested that uvularity was a rather recent distinct feature only developping from earlier allophonic variation once Syncope began in early Late IE, Lass' comments on phonemic typology appear to be quite apt: "Remembering that these are phonemic - not phonetic - 'normalcy statements', we can be justified in suspecting that, for instance, if a language has a pharyngeal fricative or uvular stop phonetically, it is more likely than not to be better characterized as an allophone of something else than a primary allophone."[6] Indeed, I propose that MIE *k had two conditioned allophones, /k/ and /q/, and that it was only in the millenium preceding PIE proper where we find newly distinct phonemes *k and *q (traditional *ḱ and *k). For that matter, it didn't take long for *k and *q to merge in Centum dialects to plain *k while pushing forward to and *k respectively in Satem dialects. This then seems like a more natural solution overall than the traditional account which would have us believe in palatalized velars which extend far into pre-IE despite being unstable and despite lacking any indication of a recent source of their supposed palatalization. The traditional account, keep in mind, is not just in violation of markedness in PIE itself, but in violation in the aeons preceding its theoretical development.


NOTES
[1] Spencer, Phonology: Theory and Description (1996), p.16 (link here): "The uvulars are much rarer sounds than the velars, and it is especially rare for a language to have a uvular sound but no velar."
[2] Here, I use "colon" as a casual marker for 'fortis' stops. Don't take it literally as a geminate consonant, as might be the interpretation of those familiar with IPA notation. I suggest giving it the phonetic value of an unaspirated, possibly semi-voiced, stop.
[3] Gross, Speaking in Other Voices: An Ethnography of Walloon Puppet (2001), p.15 (link here).
[4] Lass, Phonology: An Introduction to Basic Concepts (1984), p.155 (link here): "In fact, one characteristic of areal and genetic groups is the way they often concentrate 'oddities': a particularly striking example is the virtual restriction of phonemic clicks to a portion of southern Africa. So what's rare universally may actually be the NORM for a family or area: we may have 'family universals'."
[5] Haspelmath, A Grammar of Lezgian (1993), p.2 (link here).
[6] Lass, Phonology: An Introduction to Basic Concepts (1984), p.155 (link here).

UPDATES
(Jan 7 2008) I added a footnote after questions about my uncommon colon notation for IE stops. Sorry, people. Sometimes I forget to explain these important details.

15 comments:

  1. What kind of sounds are *q: and *k: ?

    ReplyDelete
  2. The colon sign ":" is to mark a 'fortis' stop, without imposing exact phonetics. Personally, I suspect they are unaspirated (and possibly semi-voiced much like in English). Such sounds may easily originate from earlier ejectives via simple weakening of the glottal component of the ejective. The fortis stops, being marked, are much less frequent and have greater restrictions than the 'lenis' stops (such as *t or *dh in the traditional notation).

    So I think of *q: as uvular unaspirated and *k: as velar unaspirated. Both sounds should sound much like voiced /g/ to English ears, although *q: will sound a little alien unless one has dabbled in Lezghian or other uvular-enriched languages before.

    ReplyDelete
  3. Sorry for all the questions, but I'm new at this. This is a 3-way distinction in voicing onset?
    *k - voiceless
    *k: - semivoiced
    *g - voiced

    And the reasoning for this is to give the phonemes a more normal distribution, but also to account for the fact that no daughter languages have ejectives?

    ReplyDelete
  4. Why apologize for being new? Maybe I'm newer than you. I figure we all have to start somewhere, otherwise we won't start at all and there's nothing more tragic than ending before we've begun. <:)

    As for these stops, yes, a 3-way VO contrast of unvoiced (*k), semi-voiced (*k:) and fully voiced stops (*g) makes things tidy. Thai has just such a contrast, by the way. Savour this tasty pdf showing phonologies from around the world.

    ReplyDelete
  5. thanks!

    Could you recommend a reference that covers PIE phonology? Where I could read about, for instance, how long high vowels are the product of compensatory lengthening due to a lost laryngeal.

    ReplyDelete
  6. No worries.

    In the specific case of the word "20" which you mentioned according to the older reconstruction of *wīḱm̥tī on your blog, it is generally understood that its suffix *-íh₁ marks the dual in general (see here, for example). Fortson on p.393 of Indo-European Language and Culture: An Introduction (2004), a must-have book nowdays (link here) offers another example of such a dual: *h₃ekʷ-íh₁ "both eyes". Note that *-íh₁ is an allomorph of *-h₁ as in *pód-h₁ 'two feet'. In the age of Pokorny, it was once written *-e.

    That being said, I'm probably leaping ahead of myself because for me, I find it a closed matter regarding the existence of long high vowels in PIE. I don't see the justification in them at all and believe that almost all can be traced to a laryngeal. There are cases regarding pronominal and particle forms however that lead some to believe that long vowels are conclusively proven. Clearly I have to write about this. You've inspired me. Thanks!

    ReplyDelete
  7. Hah! I'm here, back to annoy you again challenging your theories, which I myself would love to embrace (How Ironic, but essential to not become like Starostin and others).

    You say the semivoiced stop was indo-european, and that once at some point it was an ejective.

    I'm currently working on the digitalisation of a wonderful Slavic Etymological dictionary, and with all it's entries it also gives Proto-Balto-Slavic reconstructions. I'm not sure what it is based on, but preglottalised voiced consonants are actually reconstructed as such.

    For example the word:
    *adati v. ‘investigate, explore’
    with indo-european root *h3ed- relatad to Greek: Gk. ὄζω ‘smell’.

    This has a Balto-Slavic reconstructed root as follows:
    BSl. *oʔd-

    Now I'm not sure what they base this on, but I'm pretty sure there must be a reason why they're reconstructing pre-glotallisation here rather than semi-voicedness.

    The only reason I can think of is following:
    The registered vowel lengthening before your semi-voiced, Beeke's pre-glottalised consonants, is not exactly universal in all Indo-European languages. As far as I know the process is only registered in Balto-Slavic and Italic. Therefore we reconstruct a still existing pre-glottalisation, which only in those two branches came to act exactly the same as *h1.

    ReplyDelete
  8. So you think that lengthening can only occur before preglottalized consonants? Hmm. That's news to an English speaker then because in order to account for lengthening of preceding vowels as in "nab" vs. "nap", "bid" vs. "bit" and "bag" vs. "back", I guess you'll have to claim that "b", "d" and "g" are preglottalized instead of semi-voiced. Perhaps you should do a perusal through this nifty pdf: Voice Onset Time and the Scottish Vowel Length Rule in Aberdeen English. By way of the Scottish Vowel Length Rule, vowels are lengthened before voiced fricatives. No glottalization required.

    ReplyDelete
  9. Also, you may want to respond to Fortson's skepticism of Winter's Law, the lengthening law that you're alluding to, in his influential book Indo-European Language and Culture: An Introduction (2004) on page 365. The original formulation of Winter's Law, from what I understand, involved voiced non-aspirated stops, not preglottalized ones as the Glottalic Theory of Kortlandt and others demands.

    ReplyDelete
  10. Well obviously I don't think that pre-glottalisation is the only factor that can lengthen vowels.

    But if a semi-voiced consonant lengthens a preceding vowel, I do find it highly unlikely that a full-voiced consonant would not.
    I can't think of any languages that have that.

    And then there's also my argument that in Proto-Balto-Slavic they reconstruct a glottal stop before the voiced consonant. But as I said, I am not sure which proof there is for this reconstruction. I was sort of hoping you'd know ;)

    (Balto-Slavic languages are definitely not my forté haha)

    ReplyDelete
  11. What I'm really objecting to here is the notion that if this preglottalized version of Winter's Law is correct (and I admit it could be to some extent since if I understand correctly it explains the tones, although it could just as well be the product of preaspiration, no?), that it's somehow "proof" that PIE had ejectives. It's not proof; merely a possible indication. It seems like Glottalic Theory proponents are stepping over themselves too much to provide "proof" of their ejective theory, beyond the sensible things they've already established.

    Preglottalized stops don't necessarily come from ejectives as is evident in Sienese Italian (see pdf) where preaspirated consonants are allophones of geminate consonants. (Preaspiration and preglottalization are said to alternate in many Germanic languages. And then there's Saami which is a Uralic language.)

    So if PIE *d in early Proto-Balto-Slavic came to be pronounced as a long tense stop /t:/ instead of an ejective, let's say, it might just as easily develop preaspiration or preglottalization. A long stop can certainly stem from an ejective, although I don't know what *?d (rather than *?t) is supposed to mean phonetically other than an implosive where voice would be possible... which is a little different than a preglottalized stop!

    When I read sensationalist things like Kortlandt's How old is the English stop?, I start to get the impression that Glottalic Theory proponents have now become desperate and are starting to pull rabbits out of a hat instead of recognizing that their proposed phonology is ironically too stable to be considered plausible given the data. All protolanguages with ejectives that we know of (e.g. Proto-Kartvelian, Proto-Salish, Proto-Nakh-Daghestanian, Proto-Abkhaz-Adyghe) show that ejectives are inherently stable. Ergo, PIE at the time of fragmentation simply could not have had ejectives per se but rather some derivative phoneme like a semi-voiced stop or a tense stop of some kind. I have to admit there's an attractiveness to thinking that traditional IE's *d and *g are tense long stops, at least in some early IE dialects.

    ReplyDelete
  12. Thanks for the great reply.

    As for the appeal of *d and *g being tense long stops, it's quite the contrary from what you'd expect considering Hittite where we find <tt> as a reflex of *t. And <t> as a ref;ex pf *d.

    Although there's some discussion on the phonemic value of <tt> I find it likely to be /t:/ and <t> being /t/ simply because if it was a contrast of voice, the Hittites wouldn't have mixed up d/t all the time. And I can't really think of anything that you'd spell as tt in contrast to t that doesn't represent length.

    So Hittite displays the exact opposite. It would seem a bit odd that it switched up.

    ReplyDelete
  13. It seems to me that if we know that *dh and *d merged into Proto-Anatolian *d (leaving a simple contrast of *d and *t), and if *t became Hittite -tt- /t:/, then your issue is with all theories, whether "traditional", "glottalic" or the "hybrid" variety that my view is.

    Keep in mind that what is considered fortis in one stage of a language is not necessarily what is fortis in the subsequent stage.

    So regardless of the exact phonetics of PIE *d (glottalic, semivoiced, tense, long, plain voiced, whatever), it became PAnat's general *d, which was in some form or fashion voiced. At some point, Anatolian *t became "fortis" in Hittite. The result is the same, no matter which theory you choose: There was a "stop switcheroo" in Hittite and there's nothing you can do about it. Don't blame me; blame Aegeo-Anatolian areal influence :P

    ReplyDelete
  14. Phoenix has commented in my last article that languages with uvular fricatives without velar ones are typologically unusual. And he's correct [...].

    "Typologically unusual" is clearly an exaggeration. My impression is that, in the absence of a phonemic contrast between velar and uvular fricatives, linguists tend to transcribe uvular fricatives as velar. Off the top of my head, Swiss German, some kinds of Dutch as Phoenix mentioned, most kinds of Arabic, Modern Israeli Hebrew and Uyghur have uvular fricatives only.

    Now, a language with a uvular but no velar plosive, that is Klingon!

    Here, I use "colon" as a casual marker for 'fortis' stops. [...] I suggest giving it the phonetic value of an unaspirated, possibly semi-voiced, stop.

    That's a lenis, not a fortis, under any of the at least 3 definitions I've seen so far.

    As for these stops, yes, a 3-way VO contrast of unvoiced (*k), semi-voiced (*k:) and fully voiced stops (*g) makes things tidy. Thai has just such a contrast, by the way.

    Judging from the soundfiles I've listened to, Thai has a contrast between voiceless aspirates, voiceless lenes (the way I use that term -- more or less the same as your "semi-voiced" ones, though completely voiceless), and fully voiced ones. Hindi has the same (plus voiced aspirates). Spanish, too, has the same (minus aspirates). Pure voice contrasts, as opposed to voice + fortis/lenis contrasts (like in French or Russian or Japanese), are not that uncommon.

    On the other hand, contrasts between voiceless unaspirated fortes and voiceless (inevitably unaspirated) lenes are extremely uncommon. Austrian Standard German and the majority of the Bavarian-Austrian dialects, like my native one, are examples. Some Englishes have it in a few positions (co-chair vs code-share; discuss vs disgust maybe); Haida has it between two affricates; northern variants of German probably have it in word-final clusters (-nt vs -nd), but I need to find some speakers and listen; and that, AFAIK, is it.

    BTW, when I say "fortis", I don't mean the Korean stiff-voiced phenomena. Those sound majorly weird to me.

    although I don't know what *?d (rather than *?t) is supposed to mean phonetically

    Well, a consonant cluster -- as opposed to a coarticulation, which would require a tie bar in IPA. :^) [ʔb ʔd] are in free variation with [ɓ ɗ] in Vietnamese, says Wikipedia.

    And I can't really think of anything that you'd spell as tt in contrast to t that doesn't represent length.

    Swiss German has a contrast between long and short voiceless lenes.

    And now I'll go to bed at last.

    (Well, perhaps I should mention that the Swiss long lenes are what the term "fortis" was originally invented for. That's quite a pity, because it makes the term useless...)

    ReplyDelete
  15. David Marjanović: "'Typologically unusual' is clearly an exaggeration."

    You misread. I specifically said that languages with uvular fricatives **and yet no velar fricatives** are typologically unusual and then even backed it up with a footnote citing a published reference stating exactly that!

    I'm naturally dismayed that you didn't read carefully or follow my links. So sufficed to say, all the languages you mention, even Klingon, contain velar fricatives alongside uvular ones. Standard French however is a language that has a uvular fricative ("r") without a velar fricative but this has only been the case since the 17th century and so perhaps it may not last.

    David Marjanović:"My impression is that, [...]"

    Why should I accept idle impressions as a valid form of rebuttal? Back it up with facts and references.

    David Marjanović: "Well, a consonant cluster -- as opposed to a coarticulation, which would require a tie bar in IPA. :^) [ʔb ʔd] are in free variation with [ɓ ɗ] in Vietnamese, says Wikipedia."

    Wikipedia, the encyclopedia of fools. Sorry to be blunt but come on. Back to reality now...

    Greenberg addresses the lack of clear articulatory difference between a "preglottalized voiced stop" and a "voiced implosive stop" and mentions a similar purported free variation in Chrau (see here) that he questions. Even if these free variations are accurate, there's no basis to reconstruct such a thing nor present it on a phonemic level, as is also mentioned in the link.

    Plus, the term "consonant cluster" signifies "two or more distinct consonantal phonemes side by side", so it's not appropriate here since the preglottalized component theorized for this sound in order to explain Winter's Law is not in any way whatsoever a phoneme in its own right.

    ReplyDelete