1 Apr 2009

PIE "look-alike stems" - Evidence of something or a red herring?

On his blog, Phoenix has been dwelling on several suspicious stem look-alikes in Proto-Indo-European. While he's been attempting to work up a case for consonant gradation in PIE or Pre-PIE, I am far too skeptical to accept this Kortlandt-inspired view. In relation to this topic, Frederik Kortlandt has suggested that Proto-Uralic too had consonant gradation when in fact most Uralicists today accept that consonant gradation was a post-Uralic innovation[1], one of many flaws in his work that makes it difficult for me to take seriously. (Other important flaws in his theories being a flamboyant artistic license in his reconstruction of familiar words in both Uralic and IE).

However my busy mind also can't resist exploring a new perspective. We can at least say that while the roots Phoenix identifies do indeed appear to "look alike" (although this is assessment is a little too subjective for my tastes), there is no clear pattern in voicing or devoicing that we can immediately ascertain:
#1 *pieh₂- ~ *bʰeiH-
#2 *pleh₂g- ~ *bʰleh₂g-
#3 *trep- ~ *strebʰ-
#4 *treup- ~ *dʰreubʰ-
#5 *terp- ~ *(s)dʰerbʰ-
Yet what if there is actually something behind these pairs? For the sake of brainstorming, the third and fifth pairs in Phoenix's list appear to suggest a possible solution that might be enticing for future examination. In reconstructed Proto-Indo-European, just like in any living language, there appear to be phonotactic pressures which determined how sounds could validly fit within a word.

One important curiosity whose origins I have yet to solve to my satisfaction is that there appears to be some sort of "voicing harmony" such that voiced stops (ie. voiced aspirated stops, traditionally speaking) can only appear in a verb root with other voiced stops while voiceless stops can only appear with other voiceless stops. Creaky stops (ie. traditional plain voiced stops), no doubt descendants of ejectives, could occur with both voiced and voiceless stops but not with other creaky stops. So *tep- or *ped- are valid root shapes yet **dʰep- or **deg- are not. Putting aside how this pattern arose in the first place, I wonder if this well-known phonotactic constraint in unison with the presence or absence of the so-called mobile *s-prefix could be to blame for these apparent pairs above.

Here's how I imagine this could work. We start with a root like *dʰerbʰ- which has an *s-marked variant, *sterbʰ-. Following *s in an onset cluster, the voicing of a subsequent stop is neutralized as in English but otherwise the phonotactic constraint I mention above is unaffected. Now what if PIE speakers begin to drop the *s in *sterbʰ-? I'd imagine that, if the phonological constraint still holds, a resulting form *terbʰ- might be pressured into becoming *terp- to maintain the stop voicing harmony of yore. This then could especially help to explain the fourth stem pair in this list: *treup- ~ *dʰreubʰ- (via an *s-marked intermediate variant of *dʰreubʰ-, namely **streubʰ-).

[1] Denis Sinor, The Uralic languages: description, history, and foreign influences. Handbuch der Orientalistik, v.8 (1988), p.274: "Relating gradation to Uralic or even Finno-Ugric has been criticised because it is only a feature of the languages mentioned above and is not found in any other Uralic language. Scholars have therefore seen gradation in Balto-Finnic and Lapp as the result of parallel, but separate development to the gradation in Samoyed." (see link).


  1. I too have been reading his blog and find it very interesting. I haven't looked up his roots, but I would suspect that if a voiced aspirated stop devoices after *s then Indo-Iranian and possibly Greek should retain the feature of aspiration. Sieb's Law is a classic explanation for the existence of some voiceless aspirates in Indo-Iranian.

  2. That's indeed quite an interesting solution.

    By now I am convinced that Kortlandt's gradation idea can't possible work the way he says it does.

    And indeed even the who gradation is starting to feel fishy to me too.

    The solution you opt seems plausible, but there are still roots out there that clearly violate this. Like *(s)dʰerbʰ- which yields Dutch sterven. With devoicing of the initial dental due to the s-mobile, but a still existant *bʰ. Though maybe that could be solved through Verner's law, but we'd like to see alternation of *f/*b then in at least one of the germanic languages and I don't think we do.

    I must say when working with Germanic evidence for consonant gradation that Kortlandt proposes things obviously become extremely complicated since there will be 4 levels of consonant gradation at work.

    Just realise sterven in dutch could go back to both *sterban and *sterfan, so in fact there could be Verner at work.

  3. Mordrigar: "Sieb's Law is a classic explanation for the existence of some voiceless aspirates in Indo-Iranian."

    Szemerenyi rejects this rule because of Avestan zdī from *s-dʰi "be!" (imperative plural). See Szemerenyi, Introduction to Indo-European Linguistics (1999), page 144. Could you explain what Szemerenyi might have missed or is Siebs' Law really doomed?

  4. Phoenix: "Like *(s)dʰerbʰ- which yields Dutch sterven. With devoicing of the initial dental due to the s-mobile, but a still existant *bʰ."

    If one thinks of these roots on both the phonemic and phonetic levels, then one understands the resultant phonetic [t(ʰ)] as merely an allophone of *dʰ following *s (or perhaps more specifically tautosyllabic *s?). Thus there is no contradiction since *dʰ, despite devoicing, is properly paired with *bʰ. In contrast, a root that shows a devoiced stop but which confuses the allophone with the homophonous phoneme should instead properly pair with another unvoiced stop. In a nutshell, I'm talking about phonemes, allophones and the confusion of the former two to help explain your stem look-alikes.

  5. Mordigar, perhaps I can just answer my own question about Siebs' Law right now. Can we not just say that it demands that the sounds are adjacent and tautosyllabic before the law can operate?

    Thus *sdʰi (as Szemerenyi writes it in his book, but which is more accurately *ʔsdʰí) does not violate the rule because *s and , while adjacent, are not in the same syllable. This would probably help to explain Szemerenyi's other opposing examples.

    Ok, nevermind then. Carry on. ;-)

  6. The other possibility is that Sieb's Law only words word initially (which as you point out is not the case for *?sdHi). Internally *s has [z] as an allophone before voiced stops (and possibly word finally). Szemerenyi misses a lot because he insists on one laryngeal, /h/, and phonemic voiceless aspirates.

  7. Mordrigar: "The other possibility is that Sieb's Law only words word initially (which as you point out is not the case for *?sdHi)."

    Can you cite a word with an instance of an adjacent, tautosyllabic combination of sibilant plus aspirate stop where we know that Siebs' Law cannot apply? If not, it suffices to state that the rule is about adjacency and tautosyllabicity, not about its position in a word.

    In fact, any word with a reduplicated Sieb's-Law-governed stem indicates that Siebs' Law in fact has nothing to do with word-initial position.

  8. I'm going to toss a bunch of ideas in here to see if they might be of any help at all. This will be long, so I’ll have tobreak it up into more than one post.

    First off, I haven't done historical linguistics in ages now, apart from my work on diachronic changes in sign language handshapes. (I've specialised in sign linguistics for a very long time now.) I visit this blog every now and then because of my curiosity on where PIE reconstruction is at nowadays and the likelihood of PIE and (proto-)Semitic contact and borrowing at some point or points in history.

    I was not aware of the putative voiceless/voiced root alternations in PIE, though I recall having glanced years ago at an article about similar manner alternations in the first two consonants of semantically related roots in Semitic. An example set (close to what the paper gave) from Arabic as a representative of the family, involves roots related to the idea of cutting (plus one related to breaking): the first consonant is a velar stop or j from PS *g and the second a dental stop or fricative. I'll give these examples in a second comment because they take up so much space and this one is going to be long enough as is.

    It just occurred to me that the variety of realisations of the reconstructed voiced (aka aspirated) and voiceless stop series across daughter families might instead reflect varying phonetic realisations across PIE dialects of a rather different phonetic phenomenon. Let me throw out the hypothesis that voice (and/or aspiration) might not have been a phonological feature of stops; instead, breathy voice may have been a prosodic feature with the syllable as its domain. Thus PIE would have had only two stop series: plain voiceless stops and creaky voiced or pre-glottalised stops. Plain stops would vary allophonically depending on whether they appeared in modal voicing syllables of breathy voiced syllables.

    Syllabic breathy voice would have spread onto both stops in a stop-(X)-stop syllable (where K = any stop), and where the feature was absent, both stops would remain voiceless. This would explain the "voicing harmony" with no special constraints needed. Breathy voice being incompatible with creaky voice already present on a consonant, it would fail to affect these, explaining why they can co-occur with voiceless or breathy voiced consonants. (This is an argument for these stops having creaky voice as opposed to a plain voice feature, by the way.) Similarly, under gemination or feature assimilation with a following onset consonant, the breathy voice feature would be lost, assimilating the final C voicing feature to that of the following onset C.

    With breathy voice a syllable-level prosodic feature, this leaves open the possibility that consonants affected by this feature could undergo a variety of phonetic changes depending on the dialect. They could take on breath but remain voiceless; both voice and delayed breath; only voice; or full voice and aspiration at consonant onset, resulting in voiced fricatives. These allophonic realisations varying by dialect would have resulted in the variety of realisations in daughter languages. It would even make sense that /k/, for example, might be more susceptible to full spirantisation than p, t, kʷ or q.

    (...part 2)

  9. (Part 2...)

    Back to the topic at hand, though. It may be that this putative prosodic breathy voice played a (limited) morphological role analogous to ablaut or n-infixation, explaining to some extent the apparent voiceless/voiced ("aspirated") root doublets. I have two other candidates to add that I have come across while looking up etymologies from time to time:

    #9 *kap- ˜ *gʰabʰ/gʰebʰ grasp, hold (have, give)
    #10 *kap-ut ˜ *gʰebʰ-el head (top?)

    The root *kap in #9 gives English 'have', 'heavy', 'heave', Dutch 'hebben', German 'haben' and Latin 'capere' (seize), capsa (case); *gʰabʰ/gʰebʰ (or *ka̤p/ke̤p under the Syllabic Breathy Voice Hypothesis) gives Latin 'habere' and English 'give'.

    In #10, *kap-ut gives Latin 'caput' and Germanic *hauβuð > German 'Haupt", Dutch 'hoofd', OE 'heafod' > Modern English 'head'. Doublet *gʰebʰ-el (or SBVH ke̤p-el) gives Greek 'kephalos', English 'gable'.

    Obviously, the difference in vowel in the corresponding roots in #10 is a problem. If *kap is actually reconstructible as *qap - you guys would know if this is the case - this might be a simple case of allophonic variation entering into the equation.

    With this, after blithely reconfiguring the PIE stop and voicing system (chuckle), I float my hypothesis to you guys to see what you think might be its advantages and disadvantages. Though I am a phonologist, I am far too unfamiliar with the literature and common knowledge of PIE and other proto-language studies to be able to anticipate the problems my ideas might face. Nonetheless, I hope it might eventually turn out to be useful!


    For purposes of comparison, here is a list of apparent semantically related root alternations in Arabic/Semitic. I give proto-Semitic consonant sequences corresponding to the initial two consonants of the Arabic roots above each group of glosses. (All roots were gleaned from the Hans Wehr Dictionary of Modern Written Arabic edited by J. M. Cowan.)

    < *kt
    kt-b 'write'

    < k't
    qt-l 'kill'

    < *k't'
    qtˁ 'cut'
    qtˁ-ʕ 'cut'
    qtˁ-l 'cut'
    qtˁ-m 'cut off, break off'

    < k's
    qs-m 'divide, split'

    < k's'
    qsˁsˁ 'cut off, clip, shear'
    qsˁ-b 'butcher, carve up'
    qsˁqsˁ 'shatter, clip, trim'
    qsˁ-l 'cut off, mow'

    < k'ɬ'
    qdˁ-b 'cut off, prune, trim'
    qdˁ-y 'settle, judge, limit'

    < gd
    jd-ʕ 'cut off, amputate'

    < gz
    jzz 'cut off, clip, shear'
    jz-ʔ 'cut up, break up, divide'
    jz-r 'slaughter, butcher' (also, 'island')
    jz-m 'cut off, cut short, clip, settle, judge'

    < gð
    jðð 'cut off, clip'
    jð-m 'cut off, chop off'

    As you can see, the combined phonological and semantic relationships are pretty suggestive of something going on in the morphophonological history of Semitic that brings about these alternations in voicing and pharyngealisation (from PS ejective consonants) as well as manner changes in the coronal second consonant. (For what it's worth, a look at the (rather short) Bartleby list of proto-Semitic roots occasionally brings up a note referring you from the voiced consonant root you click on to another corresponding root with voiceless consonants.)

    I leave it to you, the specialists, to dig up the relevant literature, but these apparent multiple correspondences seem quite akin to the PIE data PhoeniX is looking at, though the SBVH I propose for PIE would do nothing to explain why this kind of thing appears to be going on in Semitic.

  10. Kiwehtin: "Let me throw out the hypothesis that voice (and/or aspiration) might not have been a phonological feature of stops; instead, breathy voice may have been a prosodic feature with the syllable as its domain.".

    Yet how then would you explain all 3ps subjunctives in *-et whose verb ends in a traditional voiced aspirated stop? Surely then we'd see suffix alternations like *-et with **-edh, depending on the phonation of that syllable.

    Additionally, what of *dhghōm 'earth' (ie. your *tkō̤m) which requires that the phonation extend past the asyllabic root and onto the following suffix *-om-?

    Your idea is very tempting and it almost seems to work and yet there seems to be some frustratingly insolvable conundra that shatter it here. The idea of spreading phonation sounds on the right track though.

  11. Actually, I made a mistake in the way I formulated this, because I decided to stay up and run with the idea between 1 and 2 am for fear of forgetting it. (I see a few other inadvertent but minor mistakes in what I wrote, for what it's worth.)
    I meant to say that this would be a morpheme-level phenomenon but confused it with the syllable since most roots are superficially monosyllables (an abstraction that ignores the fact they are almost always bound to following morphology). I see the spreading as local and phonetic within the morpheme at an early stage of the language, with the resulting harmony functioning as a correlate of morphemicity, especially for roots. This is analogous to other phonetic asymmetries between content and function morphemes across languages.
    Of course, some inflectional morphemes can also bear this feature - such as the *-bʰ- (or -b-/-p+BV-) of some instrumental endings and other case-number combinations (depending on the reconstruction) - but it is interesting to note that they *also* contain a noncoronal occlusive unlike most other inflectional endings. It's interesting to note the phonetic and semantic similarity with *mbʰi- here; I'm sure I can't be the first to see this and wonder about a possible relationship... Are there any *other* noncoronals in inflexional suffixes apart from the ubiquitous *m, I wonder?
    From this perspective, it isn't really a surprise that breathiness wouldn't cause changes in the coda consonant of a following inflexional ending. (This would be nonlocal spreading, unlike that involving a following onset consonant, where we *do* see voicing assimilations in daughter languages.) As for *dhghōm, I see no real problem here. As far as I knew, this was a monomorphemic root (with o-grade in this allomorph); it's interesting to see there is another analysis now! In any case, if the root is simply biconsonantal, there would be nothing to prevent breathy voice from being a floating feature comparable to floating tone features in many languages. Additionally, breathy voice is entirely compatible with a sonorant.
    Now, the reduction from three to two series I proposed is probably an oversimplification, taking the hypothesis as far as it can go. Another direction to take this would be to assume that at some early stage in PIE history, the three stop series were present but there was no constraint against voiced and voiceless consonants in a single morpheme. Once breathy voice on the syllable peak evolved after a voiced consonant, the voicing harmony effect was then in a position to develop, leading to the familiar co-occurrence restrictions.
    I think the most interesting consequence of this hypothesis (apart from the way it can be used to account for PhoeniX's doublets) is the way it accounts for the voiced and voiceless aspirates of Indic and Greek languages respectively without having to ever posit common voiced aspirates at any stage of PIE. The overall evolution in Indic and Greek would then be very similar to what happened in Khmer, Mon and Nyah Kur in the Mon-Khmer family:
    The Mon & Nyah Kur languages (http://people.anu.edu.au/~a108009/languages/Monic.html)
    Given a consonant inventory consistent with the glottalic hypothesis, this gives a phonetically realistic explanation for the development of the Hellenic and Indic aspirates that don't surface in the other branches (except perhaps in Italic as voiceless fricatives).
    I can't help wondering, given this hypothesis, if there is evidence for exceptions to voicing harmony in some branches of PIE that might point to earlier pre-harmony (and therefore pre-BV) forms and also, perhaps, uneven application of BV spreading? I see a few apparent exceptions in Calvert Watkins' PIE roots supplement to the American Heritage Dictionary of the English Language, scattered between Germanic, Greek and Latin, but I wouldn't know whether they are to be attributed to borrowing or to the kind of thing I suggest here.

  12. This is already a very interesting idea despite any details you feel you've missed.

    Your link above didn't come through in your comment so let me first help others who may want to follow it: Proto-Monic. I'll have to ponder more on this.

    Concerning PIE *dʰǵʰōm 'earth', we find Hittite tekan which suggests an alternative vocalism for *dʰǵʰ-ōm, namely *dʰéǵʰ-ōm (see Kimball, Hittite historical phonology (1999), p.265).