Traditional reconstruction gives the following velars in PIE:

*/ḱ/, */ǵ/, */ǵʰ/

*/k/, */g/, */gʰ/

*/kʷ/, */gʷ/, */gʷʰ/

I wonder what evidence is there to consider velars */kʷ/, */gʷ/, */gʷʰ/ separate phonemes rather than combinations of a velar and a labial sounds: */kw/, */gw/, */gʰw/?

I know there is a word for "horse" *h1eḱwos but here ḱ reconstructed to be palatal so it is not a minimal pair. Is there any occurrence of /kw/ in the language, so to take it as a minimal pair?


Andrew Sihler (Sihler 1995) argues that Greek is "the only centum language [...] attesting a contrast between reflexes of ḱw and *kʷ". In Greek "two distinct sounds *ḱw give a double consonant medially, while the unitary *kʷ gives a single consonant" (p. 159). For example,

PIE *ḱw > Greek ππ, e.g. Greek ἵππος, Latin equus, equos, Myc. iqo

PIE *kʷ > Greek π, e.g. Greek ἕπομαι, Lat. sequitur

He also acknowledges the fact that this argumentation is based on one word only, Greek ἵππος, "whose peculiarities do not inspire confidence" (p. 160) [emphasis mine - Alex B.].

As for the contrast PIE *kw - PIE *kʷ, I think the evidence also comes from Greek (and Mycenaean): e.g. the so called boukolos rule states that "a labiovelar lost its labial element when adjacent to the vowel *u" (Fortson 2010: 70). For example,

Greek βουκόλος 'cowherd' (PIE *gʷoukʷolos > PIE *gʷoukolos), but αἰπόλος 'goatherd' (PIE*aikʷolos).

To conclude: PIE was not a monolithic, 'timeless' language that never changed. There is good reason to believe that there were different stages in its development, too.

  • No, in h1eḱwos there is no *kw, only *ḱw. Look at the question, I already mentioned this word. – Anixx Nov 13 '12 at 3:44
  • Okay, but this is not an example of contrast between kʷ and kw. What is the proof that these two different varuants existed in the proto-language? I saw numerous reconstructions with *kʷ but all reconstructions with *kw were doubtful and different by different authors. Actually my hypothesis is that there was only kw and no kʷ (more exactly, it was uvular *qw) – Anixx Nov 13 '12 at 16:57
  • @Anixx, your hypothesis cannot explain the boukolos rule, for starters. – Alex B. Nov 13 '12 at 21:18
  • why this rule cannot be formulated without labiovelars? Something this way: wkw->wk – Anixx Nov 13 '12 at 21:22
  • Why uvular? I hope you don't read the letter "q" as an IPA uvular sound? – Alex B. Nov 13 '12 at 21:22

-w- can act as a vowel: *drew- > *drw- (zero grade) > *dru- etc. If kʷ was -kw- we would see kʷ "disintegrate" in certain situations into -kew-, -kow-, -ku- etc. which doesn't happen.

  • Could you please elaborate so that it becomes clearer how this answers the question? – robert Feb 20 '14 at 18:41

All these arguments are legitimate. But you could also ask whether there is any real human language (not reconstructed) that has a phonological contrast between /kʷ/ and /kw/.

  • Probably not /kʷ/ and /kw/ but /q/ and /kw/. – Eleshar Aug 31 '16 at 13:42
  • @Eleshar: I don't know of anyone who reconstructs traditional PIE *kʷ (labialized velar stop) as /q/ (uvular stop). Some people think it was /qʷ/, a labialized uvular stop. This would contrast with non-labialized /q/ (traditional PIE *k). – brass tacks Aug 31 '16 at 19:35
  • I believe this is actually quite pointless since most historical linguist reconstruct just more or less empty phonemes without much ambition to know how these were realised (and some will actually admit it), so it is largely irrelevant if you mark it as *kʷ, *q, *qʷ, or k2 - the "k" there is there mostly to symbolise that the reconstructed phoneme evolved into k-like sounds. – Eleshar Aug 31 '16 at 19:47
  • My idea of the system is 4 degree positional contrast (*p, *t, *kʲ, *q) which would be far from unheard of in existing languages. I find this more plausible than 3 degree contrast where just the last one differentiates roundedness. – Eleshar Aug 31 '16 at 19:54
  • None of this actually addresses my question, which was whether there is any evidence for a phonological contrast between /Xʷ/ and /Xw/, whatever /X/ might be. – fdb Aug 31 '16 at 19:58

I feel like this is a common question people have when they first learn about the labiovelar series, but none of these answers are very satisfying.

As said, if we can't find a direct difference in reflexes, then we might look for differences in behaviour: we would always expect to see *kʷ as *kʷ, but *kw would, in certain circumstances, become *ku instead. The bare fact that *kʷ doesn't so disintegrate isn't enough—we need demonstrated (not just theoretical) evidence that *kw does.

And if we can't find any instances of *kw because plain velars are so rare to begin with, we could look to *ḱw instead, since all of the centum languages collapsed the palatovelar series into the plain velars. Centumisation happened post-PIE, so we might have to add a few extra caveats to our conclusions, but whatever.

But what we can't do is look for alternation between *ḱw and *ḱu and take that as evidence of a contrast between *kw and *kʷ unless we can also show that this alternation happened post-centumisation, otherwise we've only demonstrated a contrast between *ḱw and *kʷ, which I don't think anyone doubts. The alternation present in the paradigm of *ḱwṓ certainly predates centumisation, so it's not relevant, and *h₁éḱwos doesn't show any alternation at all. Really, only evidence of different outcomes of *kʷ and *ḱw themselves would do, and there is none (Sihler's ἵππος notwithstanding).

There is, however, a controversial root *kwep- 'to smoke, steam', which is apparently attested in the full-grade in Latvian kvêpt 'to smell', kvêpêt 'to smoke', and Lithuanian kvė̃pti 'to smell' (and a bunch of other languages in other grades: Latvian and Lithuanian aren't the only basis for this root's existence). If you accept this root, that's your smoking gun: Latvian and Lithuanian are satem languages, so they turned their labiovelars into plain velars, but here they preserve PIE *kw as /kv/ or /kʋ/ instead of /k/. Direct evidence of different outcomes.

(That root also shows the predicted alternation between *kw in the full-grade and *ku in the zero-grade, but we don't even need it now.)

  • "The bare fact that *kʷ doesn't so disintegrate isn't enough—we need demonstrated (not just theoretical) evidence that *kw does" -- I don't think we do, specifically. The putative *kw contains *w, and we know that *w was an alternant of *u. So the only way we could explain a lack of *kw~*ku alternations would be to say that *w~*u alternation was blocked after the one segment *k, which goes against everything we know about PIE syllabification. – TKR Apr 15 at 19:00
  • @TKR Taking the theory as evidence of the theory is circular. Yes, it'd introduce inelegancies, but that doesn't mean it's necessarily false. – Cairnarvon Apr 15 at 21:33
  • I'm not sure what you mean by "taking the theory as evidence of the theory" -- the *w~*u alternation is established independently of whatever views one might hold about the dorsal series, so of course it can be used to inform those views. – TKR Apr 15 at 23:07
  • @TKR The alternation is established independently, but that doesn't mean there can't be a conditioned exception after *k. That's the whole point. – Cairnarvon Apr 15 at 23:15
  • Well, nothing's impossible, but given how radically incompatible such an exception would have to be with what we know of PIE ablaut and syllabification, I think it can safely be ruled out. – TKR Apr 15 at 23:33

There are several indications this was not just combination of *k+*w.

1) Morphology: Semi-vowels were vocalised in zero grade verbs but this does not happen for *kʷ. To the contrary, there are cases where there is *ḱ followed by /w/ that can vocalise to /u/ ("dog" - a.gr.: kyón, gen.: kynos, skrt.: śván, gen. śunah) which would be fairly peculiar if it did not happen for regular /k/.

2) Alphabetic representation: Some pre-classical Greek dialects use letter "koppa" where *kʷ is reconstructed (which is actually the source of Roman letter Q). There is little reason to assume they would record one particular combination of phonemes as a single phoneme (not that it would be unprecedented in Greek, just the coincidence seems unlikely).

3) Contrast type: It seems certain that PIE *ḱ corresponded to some sort of mildly palatal form of [k] as in the so-called satem languages, it developed quite indiscriminately to sibilants. Therefore if you have a palatal version of [k], and another version of [k], it is likely the contrast will be somewhat reinforced by distancing the places of articulation from each other, the latter being moved more to the post-velar regions of the mouth (as in [q] in IPA), which tends to produced effect not entirely dissimilar to labialisation (same as back vowels tend to be rounded too). I believe it is not difficult to find supporting evidence in linguistic typology for this.

4) Satemisation: When IE languages changed the spawn of *ḱ into something else (either by satemisation or frequent palatalisation), the spawn of *kʷ tended to lose the labial element, as if it were no longer needed to distinguish a contrast because the contrast ceased to exist due to one part of it changing almost entirely to something wildly different. Also it makes sense with regards to (3) - the contrast might have been lost due to it being reinforced too much one way or the other. Again, this is not ubiquitous and does not constitute a proof but it points out in this direction.

  • 1
    The other arguments are good ones but 2 is incorrect: koppa was used for any [k] that did not precede a front vowel. Most such instances did not come from labiovelars, which had been lost centuries earlier. – TKR Aug 31 '16 at 21:01
  • if wonder if the Greek kyon can be just leveling by analogy with oblique cases such as kynos – Constantine Geist Apr 19 '17 at 21:23
  • Alphabetic representation works for Gothic where the descendants of kʷ and gʷ (hʷ and kʷ) had their own letters (𐌵 and 𐍈). But that argument would make them the only Germanic language with surviving labiovelars pronounced as labiovelars. – Ned Apr 7 at 17:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.