23 February 2016

The Strange Case of the Jumbled Vowels

Though scattered traces of athematic reduplicated presents can be found in several branches of Indo-European, it’s only Indo-Iranian and Greek that preserve them well enough to enable reconstruction. Indo-Iranian evidence is especially important, since that branch seems to distinguish two types reduplicated presents, one with *e and the other with *i as the echo vowel. Moreover, the ablaut (vowel alternations) in the conjugation of reduplicated presents can be seen there more clearly than in Greek.

  • Vedic bábhasti, bápsati ‘chew, devour’, as if from *bʰe-bʰes-ti, *bʰe-bʰs-n̥ti [1]; 
  • Vedic jígāti, jígati ‘go’, as if from *gʷi-gʷah₂-ti, *gʷi-gʷh₂-n̥ti.

Some Indo-Europeanists believe that the two types are inherited and their coexistence in Indo-Aryan is an archaism rather than an innovation. In the LIV (p. 16) [2] they are reconstructed with different PIE vowel grades and accent patterns:

  • Type 1: *dʰé-dʰoh₁-/*dʰé-dʰh₁- (root *dʰeh₁- ‘put, place’); 
  • Type 2: *sti-stéh₂- [*stistáh₂-]/*sti-sth₂- (root *steh₂- ‘stand’).

In Greek, on the other hand, the echo vowel is invariably *i, and the root vowel (when accented, as in the singular) is always a reflex of *e. Note the characteristic triad of examples (three very common verb roots, each with a different laryngeal):

  • Greek títʰēmi ‘I put’, as if from *dʰi-dʰeh₁-mi;
  • Greek hístēmi ‘I cause to stand’, as if from *s(t)i-steh₂-mi [*sistah₂mi];
  • Greek dídōmi ‘I give’, as if from *di-deh₃-mi [*didoh₃mi].

Type1 and Type 2
Wikimedia Commons
It seems that Type 1 disappeared completely in the prehistory of Greek and all verbs originally belonging to it were absorbed by Type 2. The o-grade reconstructed in the LIV for Type 1 is not directly confirmed by Indo-Iranian evidence (all non-high vowels merged as /a/ there); it is inferred from rather complex assumptions about Proto-Indo-European vocalism. The only fact cited in its support is the anomalous o-grade present of Germanic *ðō- ‘do’ (found only in West Germanic). The idea that it represents dereduplicated *dʰé-dʰoh₁- inherited from Proto-Indo-European is hard to reconcile with our understanding of other reflexes of genuinely reduplicated *dʰeh₁- in Germanic (as we shall see).

Relics of reduplicated presents derived from *dʰeh₁- and *deh₃- [*doh₃-] can also be found in Balto-Slavic. The former had e-reduplication there, as shown by Lithuanian dẽda (3sg.) ‘lay, put’ and Old Church Slavonic deždǫ (1sg.) ‘put’ (< Proto-Slavic *de-d-je/o-, transferred to the *-je/o- conjugation). The latter, curiously, is reduplicated with Balto-Slavic *ō, as in Lith. dúodu, OCS damь (< *dad-mь*dōd-mi, with athematic inflections). This *ō reflects earlier short *o, lengthened before non-aspirated *d (Winter’s Law). We can therefore reconstruct parallel reduplicated stems at an earlier stage of the Balto-Slavic parent language: *dʰe-dʰ- ‘put’, *do-d- ‘give’.

It’s clear that the “weak” form of the stem (with the root in zero-grade) was generalised in each case, but why are the echo vowels different? The most parsimionious explanation is that *dʰe-dʰ- is a straightforward reflex of PIE *dʰe-dʰeh₁-/*dʰe-dʰh₁- (levelled out in favour of the weak variant), whereas in the Balto-Slavic descendant of PIE *de-deh₃- [*dedoh₃-]/*de-dh₃- the echo vowel was assimilated to the laryngeally coloured root vowel of the “strong” stem (*dedoh₃- > *dodoh₃-). Subsequently, this new pronunciation was generalised across the paradigm (*dedh₃- > *dodh₃- > Proto-Balto-Slavic *dōd-), and only the weak variant survived into historical times. For this hypothesis to work, it is necessary to assume that the original strong vocalism of the reduplicated present of *dʰeh₁- was *e, not *o; otherwise it would also display the echo-vowel assimilation visible only in *dōd- ‘give’.

It seems reasonable to conclude that Type 1 and Type 2 differed much less than the LIV reconstruction suggests. The ablaut pattern of the root syllable seems to be the same in both types; the only significant difference between them concerns the choice of the echo vowel. This is how the two types are reconstructed e.g. by Don Ringe (2006: 28)[2]:

  • Type 1: *dʰé-dʰeh₁-/*dʰé-dʰh₁-;
  • Type 2: *stí-steh₂-/*stí-sth₂-.

Note the fixed accent on the echo syllable, consistent with most of the comparative evidence. On the other hand, this reconstruction doesn’t tell us why the root syllable alternates between e-grade and zero-grade. Nor does it help to account for the different echo vowels. Is the occurrence of e-reduplication beside i-reduplication just a messy fact of life, or are we missing something?

The two reconstructions can’t both be right, although they can both be wrong. I actually believe that neither of them is correct, and I’ll try to justify my opinion in the next post.


———

[1] The forms cited here are 3sg. and 3pl. The sequence *bʰs must have developed into something like Proto-Indo-Iranian *bzʰ as a result of progressive breathy-voice assimilation (Bartholomae’s Law). Although it ended up as voiceless [ps] in Vedic, the aspiration survived long enough to trigger the deaspiration of the initial consonant of bápsati by Grassmann’s Law.

[2] Helmut Rix, Martin Kümmel et al. 2001. Lexikon der indogermanischen Verben (2nd edition). Wiesbaden: Dr. Ludwig Reichert Verlag.

[3] Don Ringe. 2006. A linguistic history of English. Vol. 1: From Proto-Indo-European to Proto-Germanic. Oxford: Oxford University Press.

30 comments:

  1. The sequence *bʰs must have developed into something like Proto-Indo-Iranian *bzʰ as a result of progressive breathy-voice assimilation (Bartholomae’s Law). Although it ended up as voiceless [ps] in Vedic, the aspiration survived long enough to trigger the deaspiration of the initial consonant of bápsati by Grassmann’s Law.
    If I recall correctly, it has been argued that the PIE development was (e.g.) *-dh-t- to *-t(s)t- and that the "Bartholomew" Clusters are due to re-composition in Indo-Iranian. In that case, the -ps- in bápsati would be old (and how would you get -ps- from *-bzh- anyway?); the media in the reduplication syllable in the plural would then be due to analogy / paradigmatic levelling after the singular.

    ReplyDelete
    Replies
    1. There is independent evidence that such clusters underwent devoicing in Vedic; suffice it to mention kṣam-, kṣiṇāti, etc. This devoicing was not even general in Indo-Aryan, and did not happen at all in Iranian. Compare the desiderative stems, Ved. dipsa-, Av. dißža-, both from *dʰi-(dʰ)bʰ-(h₁)s-e/o- 'intend to deceive'.

      Even if the conditioning of Bartholomae's Law is morphophonological (a position I tend to I agree with[*]), we still need the *bzʰ cluster in early Indic to account for the application of Grassmann's Law. The latter did not operate unless breathy voice had been restored in the second stop/cluster in the stem, cf. Ved. bʰut < *bʰudʰ-s [*bʰuts] 'awakening'.

      ---
      [*] Though I'd like to know how to explain e.g. a voiced cluster in the Iranian words for 'wasp, stinging insect', derived from *wobʰsah₂. Was the relationship with 'weave' still transparent at that stage? I somehow doubt it.

      Delete
  2. Good point on the "wasp" word. But for Grassman in baps-, as I said, the deaspiration of the reduplicated syllable could be an analogical transfer from the singular bábhas-.
    I dimly remember from my IE lectures that there are a few cases in Iranian where Barthlomae didn't happen and we have voiceless clusters. But I heard those lectures a quarter-century ago and I don't think I still have my notes..

    ReplyDelete
  3. Hi Piotr, may I kindly ask you to review the material I've consolidated to illustrate the proposed Pre-Laryngeal Plosive Loss Law. It's now hosted as a guest post on Eli Nelson's site (http://indoeuropeanetymology.blogspot.com/2016/03/discussion-of-vladimir-diakoffs.html#comment-form). It's dealing with very standard IE vocabulary, and I do hope to have a productive conversation with a highly trained Indo-Europeanist such as yourself about it. In the comments section there's an add-on dealing with the *gwiH3wo- root you wrote a paper on. Thank you!

    ReplyDelete
    Replies
    1. I'm (very) sceptical but it's certainly ingenious!

      The semantic correspondences are surprisingly good, with a few exceptions... 'tooth' and 'foot'!

      Delete
  4. The connection between tooth and foot is not direct. *ped- is the 'foot' root enlarged with -on- to mean 'bottom, below' and only from there 'lower tooth' with a body-part suffix -t-. But what it tells us is that PIE differentiated between lower teeth and upper teeth (compo. mandibula and maxilla as two different jaws). This in turn explains why two tooth forms *H3edont- and *gombho- are reconstructible for PIE > one set of languages generalized the lower tooth form, the other - the upper tooth form.

    ReplyDelete
  5. I know, but some similar 'up-down' logic is conceivable for an enormous number of potential semantic correspondences.

    ReplyDelete
  6. Some of them are very real and well-known: toe-finger, toe-nail-finger-nail, arm-thigh, palm of hand, sole of foot, cheek of face and cheek of buttocks. If the phonology and morphology are right I wouldn't be too skeptical about the semantics.

    ReplyDelete
  7. I mean that a large number of roots that you might have found with the right phonological shape could have had a meaning vaguely connectable to the z axis in physical space.

    ReplyDelete
  8. But yes, as I say the semantic correspondences in general look pretty strong to me.

    Though I don't know why you imply that PIE having more than one word for concepts like 'drink' and 'tooth' is something needing explanation. Does any known language only have one word for 'drink'?

    ReplyDelete
  9. In general, it's a valid point. There could be two words for the same concept, but there must be some slight semantic or pragmatic (e.g., high-register vs. low register) differences to justify the existence of two words. Comparative method may not capture those protolinguistic nuances. But if there's a way to either clearly differentiate between the two (lower tooth vs. upper tooth) or lump them as one form (drink), then it's better than to just be agnostic about the solution.

    ReplyDelete
  10. There's still a strange unexplained phonemic split in this hypothesis... among many other things.

    ReplyDelete
  11. Here are a few points that may answer your question about the "strange" split: 1). PIE had a mobile accent that shifted within a paradigm; 2) cluster T+H is the ancestral condition, while TH as a single phoneme is a derived one; 3) judging by the 'earth' example (which everybody agrees had a plosive loss but only my Pre-Laryngeal Plosive Loss Law explains why the plosive got lost), different root shapes correlate with different accent placements (*dH1e'ghom vs. *(dH1)ge'm; 4) one of the accent placements (either on the syllable immediately following the cluster, or on any other syllable down the string) favored the retention of the the ancestral cluster condition, which resulted in plosive loss and vowel coloring; 4) leveling took place across paradigms that may have obfuscated the original accent-based conditioning.

    What are the many other things?

    ReplyDelete
  12. The divergences in stem and suffix are an issue, though if the change were very ancient I suppose that would be expected to some extent.

    But the thing I find most unconvincing is the idea that in these cases it's the initial consonant that got lost in favour of the laryngeal. Especially as we know that in PIE proper and all the daughter languages laryngeals were frequently lost, and from typology that h-like sounds in general are prone to loss.

    ReplyDelete
  13. "The divergences in stem and suffix are an issue."

    Could you be more specific?

    " the thing I find most unconvincing is the idea that in these cases it's the initial consonant that got lost in favour of the laryngeal."

    It's a good observation (even clusters of kl- and st- seem to be subject to elision in front of a "laryngeal"), but I guess I would rephrase this argument as a question regarding the exact phonetic nature of laryngeals. If it's typologically unlikely for a stop to be lost in front of a fricative, then laryngeals were not originally fricatives. (They may have turned into fricatives in Anatolian only.) But the etymological evidence I presented needs to be debated on its intrinsic grounds and typological evidence can't invalidate it.

    ReplyDelete
  14. Sorry for the delay...

    judging by the 'earth' example (which everybody agrees had a plosive loss but only my Pre-Laryngeal Plosive Loss Law explains why the plosive got lost), different root shapes correlate with different accent placements (*dH1e'ghom vs. *(dH1)ge'm;

    Actually, the constraints on syllable shape that PIE clearly had explain this quite easily. In short, PIE speakers found dhgh- unpronounceable. Read Andrew Byrd's work on academia.edu.

    4) leveling took place across paradigms that may have obfuscated the original accent-based conditioning.

    Sure, but to propose such a hypothesis you need to present it in much more detail; in particular, you need to show that the grammatical forms with the accent placement that led to the reconstructed forms were more common than the others.

    What are the many other things?

    I'll reply on that blog.

    then laryngeals were not originally fricatives. (They may have turned into fricatives in Anatolian only.)

    They behaved a lot like *s in PIE, though: they were able to occur in the same places in a syllable.

    typological evidence can't invalidate it

    It can, however, make a strong probability argument.

    ReplyDelete
  15. @David

    " In short, PIE speakers found dhgh- unpronounceable. Read Andrew Byrd's work on academia.edu."

    "Unpronounceable" looks declarative to me. And a "constraint" argument is not as strong as a "phonetic regularity" argument. Plus it doesn't include forms such as Gk ekhis that don't have the same syllable shape but still show the loss of the plosive. You can of course simple exclude it as a completely different root but then the material looks potentially a little doctored.

    "They behaved a lot like *s in PIE, though: they were able to occur in the same places in a syllable."

    So, all three laryngeals were like -s-? Even if one of them behaved like /s/, we have cases when a plosive is lost before s, as in *kswek's- 'six'.

    ReplyDelete
  16. Although we need to be careful about ascribing phonetic values to reconstructed PIE phonemes, I find it hard to see how at least h2 and h3 could be anything but fricatives.

    ReplyDelete
    Replies
    1. I'd be equally careful about adscribing the status of a real language to the theoretical construct called "PIE". But yes, h₂ and h₃ are surely fricatives like their Semitic counterparts.

      Delete
  17. Wouldn't it also matter that these "fricatives" had a very far back articulation and H1 may have been glottalized?

    ReplyDelete
  18. *h1 was almost certainly glottal – [ʔ] or perhaps [h].

    "In short, PIE speakers found dhgh- unpronounceable. Read Andrew Byrd's work on academia.edu."

    "Unpronounceable" looks declarative to me.


    It's a largely testable conclusion. As I said, read his work, it's on academia.edu. :-)

    Plus it doesn't include forms such as Gk ekhis that don't have the same syllable shape but still show the loss of the plosive. You can of course simple exclude it as a completely different root but then the material looks potentially a little doctored.


    I'd rather say "the material looks potentially a little doctored" if you include this word...

    Even if one of them behaved like /s/, we have cases when a plosive is lost before s, as in *kswek's- 'six'.

    No; the only evidence for this initial *k is the Avestan form, overlooking the fact that initial *šw regularly became *xšv in that language.

    ReplyDelete
  19. @David

    "It's a largely testable conclusion. As I said, read his work, it's on academia.edu. :-)"

    How can "unpronounceable" be testable? I read Byrd but his interpretation came prior to my formulation of Pre-Laryngeal Plosive Loss Law. No need to postulate the ad hoc emergence of unpronounceable clusters if you have a law that postulates what exactly happened to "unpronounceable" clusters across a wide swaths of examples.

    "I'd rather say "the material looks potentially a little doctored" if you include this word..."

    "Doctored" does not mean what you want it to mean. Try another word.

    "overlooking the fact that initial *šw regularly became *xšv in that language."

    OK.

    ReplyDelete
    Replies
    1. How can "unpronounceable" be testable?

      By reconstructing a phonotactic system with its constraints on syllable structure.

      No need to postulate the ad hoc emergence of unpronounceable clusters

      Who does that? Not Byrd.

      Delete
  20. Replies
    1. Or here? I recommend reading the last paragraph first.

      Delete
  21. There are, there are. Sorry, I know it's bloody embarrassing to take such a long leave in the middle of the series. I can only assure everybody that I'm alive and well, and shall be back to regular business soon.

    ReplyDelete
  22. It's the hope that kills me...

    ReplyDelete
  23. Welcome back, Piotr! To inaugurate your return, I posted a couple of new etymologies based on Pre-Laryngeal Plosive Law and explaining a couple of IE numerals at https://indoeuropeanetymology.blogspot.com/2016/03/discussion-of-vladimir-diakoffs.html?showComment=1468127893876#c7971162477202243877.

    ReplyDelete
  24. To find out a different way of phonetic hypothesis of language evolution see this blogb http://langevonew.blogspot.co.il/

    ReplyDelete
  25. Doesn't Greek εὔχομαι represent Type 1, *h₁e-h₁ugwh- (thus Beekes)?

    ReplyDelete