What is the evidence for *ḱw > *kʷː in Greek?

Question

It seems to be commonly accepted that Proto-Indo-European *ḱw became something very close to *kʷ in Greek, hence ἵππος (Mycenaean i-qo = *hiqqos?) showing the same develarization as ἕπομαι. The geminate /pp/ seems to imply that the earlier labiovelar was geminated, which would also make sense if it came from a cluster—in other words, *ḱw > *kʷː, separate from *kʷ.

As TKR says in these comments:

I've always seen [this theory] presented as communis opinio. Is there other evidence for the Greek reflex of *ḱw? A priori the idea that *kʷ > p and *ḱw > pp looks obviously plausible phonologically, whatever the problems with the initial vowel of ἵππος.

Weiss presents the Greek contrast as a fact in his Latin grammar (p. 34), and Palmer implies the same. Certainly it would be good to have more evidence, but the idea is widely accepted AFAIK.

However, Cairnavon disagrees:

I'm not aware of any other direct reflexes of *ḱw (*ḱwṓ obviously became κύων and is word-initial anyway, none of the other instances I can think of seem to have left any reflexes in Greek), but *ḱ merged with *k everywhere else, including before *u, as it did in the other centum languages, none of which show a contrast between *ḱw and *kʷ. As far as I know only Sihler has argued for such a contrast in Greek (could be he was repeating someone else, he's not good about indicating sources), and even he admitted the evidence, which is just ἵππος, isn't strong.

And indeed, ἵππος is a somewhat odd word for other reasons, so it's a hard word to pin a whole theory on.

Are there any other clear reflexes of *ḱw in Greek, or good reasons apart from ἵππος to assume that the reflex of *ḱw was geminate? Or are Weiss, Palmer, and Sihler all basing this idea solely on ἵππος?

Answer 1

I know that Germanica non leguntur in some quarters but anyway,

Fritz and Meier-Brügger 2021 (10th ed.), section G506 (pp. 69-71), btw this is a popular introductory textbook:

"Für equus bedeutet dies, dass vom Lat. aus nicht zu entscheiden ist, ob in diesem Fall -qu- auf *-k^u̯ oder *-k̑u̯ zurückgeht. Das Iir. macht in unserem Fall die Entscheidung für *-k̑u̯ aber leicht." [p. 70, emphasis mine - Alex B.]. See p. 70 for their analysis of the relevant Indo-Iranian data.

"Zum -pp-: Die ursprüngliche Doppelkonsonanz -*-ku̯ ist vermutlich bereits vormyk. durch -k^u̯ k^u̯- ersetzt worden (daraus dann nach Beseitigung der Labiovelare neu ein -pp-, s.u. P 343 Abs. 4).

Hinweise auf ein altes -k^u̯ k^u̯- geben die bereits myk. Schreibung als i-qo (ein mögliches *i-ko-wo kommt nicht vor) und die im alphabetischen Gr. notierte Doppelkonsonanz."

(cf. p. 55, English translation of the 8th German ed. by Charles Gertmenian, 2003):

also see NIL s.v. *h₁ék̑u̯o- and Meillet 1908 Introduction a l'étude comparative des langues indo-européennes (p. 64).

I have not seen any other serious proposal yet but I am very open-minded and I would be happy to consider another analysis backed up by relevant language data (examples!) and linguistic theory. And I also hope we can examine other data re: *ḱw > *kʷ rather than focusing solely on possible ad-hoc explanations of the root vowel or gemination in ἵππος with a heterosyllabic cluster, Wetzels 2002: 312 explains the gemination due to w losing its segmental status, Meier-Brügger 2017 writes it "underwent gemination to maintain the original prosodic structure of the etymon" (p. 701), i.e. *kw > *kʷkʷ > *kwkw > pp etc - proposals galore, for an excellent review see

Bernabé, Alberto. "The Noun for ‘Horse’ in Mycenaean and Some Related Terms". Synchrony and Diachrony of Ancient Greek: Language, Linguistics and Philology, edited by Georgios K. Giannakis, Luz Conti, Jesús de la Villa and Raquel Fornieles, Berlin, Boston: De Gruyter, 2021, pp. 115-124. https://doi.org/10.1515/9783110719192-008

Some important ideas from Bernabé 2021:

• “In Mycenean the spellings for ancient k+w (as in ‘horse’) are clearly distinguished from the ones for k+w in morpheme boundary” (118), e.g.

Cf. i-qo, e-qe-a-o, e-qe-o vs. te-tu-ko-wo-a, o-da-ku-we-ta

• Aspiration in Greek is secondary (p. 120),

Tarentine ἴκκος, Macedonian Ἰκκότας, compounds like λεύκιππος; Mycenian e-pi-qo-i (if aspirated, it would have been †e-pi-i-qo-i), pace Perpillou 2004:175 and Garcia Ramon 2016: 218, 220, 228.

• Melena 2014: 15 (ft. 29) “the labiovelar series can be also used for writing clusters of inherited dorsals plus /w/”, e.g. he reads i-qo as /ikwos/ and not the usual /ikk^wos/

Bernabé also argues that

"The use of a sign of the q- series instead of a sequence like †i-ku-wo seems to denote that the phoneme that existed between the two vowels was pronounced in a very similar way to that of the labiovelars, whereas the fact that the “morpheme boundary” spelling was not used can probably be attributed to the fact that the sequence was not perceived in the same way in one case (-kk^w-) as it was in the other (-k+w-). But neither would be an absolute coincidence with the results of the original labiovelar, because this cluster resulted in geminated -ππ- in the first millennium, while the labiovelar before o gives π, not a geminated. It therefore seems most likely that what we had in Mycenaean was a geminated labiovelar, /kk^w/, so when the sequence with *k^wo resulted in πo, the sequence with geminated *-kk^wο- resulted in -ππo-. On the "other hand, /-kk^w-/ and /-k^w-/ are not distinguished in the Mycenaean script, because there is never a graphic distinction between simple and geminated consonants. We found other possible cases of geminated labiovelar in Mycenaean: mo-ro-qa, *μοιρόκκ^wᾱς, ‘owner of a plot or lot of land’, composed of μοῖρα ‘part’ and the verb πέπᾱμαι (see Boeotian τὰ ππαματα, where ππ- comes from -kw-), and also the anthroponym of the same root pi-ro-qa-wo[, *Φιλοκκ^wᾱϝων. Likewise, through comparison with the geminated Eol. ὄττι we can postulate a geminated voiceless labiovelar, product of an assimilation *d + k^w > *kk^w, for the term jo-qi, yόκκ^wι(δ) < *yod-k^wid ‘what’. We still have a voiced correlate in the compound pe-qa-to, which must be reconstructed as *πέγγ^wατον “floor” (of the chariot) < πεδ- followed by the initial *g^w- of the root of βαίνω" (pp. 121-122).

If you want to get a broader view, with an extensive historical review, see Otkupshchikov 1989/2001 (Откупщиков 1989/2001), in Ряды индоевропейских гуттуральных

Answer 2

Here's what Buck's Comparative grammar of Greek and Latin, retained in Sihler's 1995 reworking, has to say about it:

Greek is the only centum language (so called, II) attesting a contrast between reflexes of *ḱw and *kʷ. Although the notion that *ḱw > G ππ distinct from *kʷ > π is reasonable on its face, the only obvious evidence for it is G ἵππος, whose peculiarities however—the unexplained rough breathing and the unexplained ι- from PIE *e-—do not inspire confidence. Myc. i-qo at least underwrites the ι, but raises another question: where -kw- occurs elsewhere it is written differently, as te-tu-ko-wo-a₂ (tetukʰwoha) 'wrought' perf. pple. of τεύχω. Perhaps a monomorphemic /(h)ikwos/ was different (phonetically? or just graphically?) from formations with an analyzable suffix beginning with w. But the writing i-qo may mean that the -ππ- of ἵππος is just one more odd thing about the word, and not a straightforward reflex of *ḱw.

Even slenderer evidence is afforded by in Boeot. πᾱμα 'possessions', on account of such epigraphic attestations as τα-ππαματα 'possessions', θιο-ππαστος 'belonging to the god'. But the supposed cognate, Ved. śvātrá-, is a word of uncertain meaning used only of soma. An etymon *ḱweH₂- would account for all forms, Boeot. -ππ- and all; but the absence of semantic controls would enjoin extreme caution even without the discouraging evidence of the Homeric hapax πολυ-πά̄μονος gen.sg. 'exceedingly wealthy', which must scan πολῠ-. (This accords with the usual treatment of the initial, as in the aor. ἐπᾱσάμην 'get, acquire', and many others.)

(This is the passage I mistakenly remembered as coming from Sihler himself in one of those comments. Sihler actually completely uncritically goes in for *-ḱw- > -ππ- with no further comment or evidence.)

FWIW, Skt. śvātrá- is usually taken to mean 'invigorating', and while the etymology of πᾱμα and its relatives remains doubtful (Beekes discusses them under πέπαμαι; he does not connect them with śvātrá- and suggests an ad-hoc reconstruction with initial *ḱw- by presupposing the distinct reflex is fact, so his reconstruction can't be taken as evidence that it is), Lucien van Beek recently argued that it's from the extended root *peh₂s- 'to pasture, guard, oversee' instead.
(If the root does start with *ḱw- and Greek does have a unique contrast with *kʷ there, an explanation is still needed for why the geminate doesn't show up in all the places it doesn't.)

The awkwardness of ἵππος is certainly the only reason it was ever posited that Greek, unique among the centum languages, had distinct reflexes of *ḱw- and *kʷ in the first place. It's not academically tenable.

Answer 3

Not an answer, but FWIW here is Lejeune's discussion in Phonétique historique du grec.

Answer 4

There's evidence for a broader Greek sound change *-Tw- > -TT- (where T = voiceless stop), of which ἵππος would be an example. A couple of relevant data points:

1. The change *-tw- > -ττ-~-σσ- is widely accepted, as in τέτταρες~τἐσσαρες : quattuor, dialectal ἥμισσον "half" < *ἡμιτϝ-ον.
2. The geminate of πέλεκκον "axe-handle" (from πέλεκυ-ς "axe") has been explained as from *-kw-. If so, this would show a development *-kw- > -kk-, which would parallel the gemination in ἵππος. The different outcome -κκ- would of course require explanation; an obvious reason would be the influence of πέλεκυς (and maybe also dissimilation from the initial π-).

I don't think there are any Greek data for the reflex of *-pw-.

Further afield but still possibly informative is the treatment of *-Ty-, which also produces geminates or clusters: e.g. μέλιττα~μέλισσα (-ty-), φυλάττω~φυλάσσω (-ky-), κλέπτω (-py-). Of course there's no a priori reason why the two glides should pattern together, but it wouldn't be surprising if they did.

Sihler's objection about te-tu-ko-wo-a₂ (quoted in Cairnarvon's answer) doesn't seem very strong: the idea is that i-qo reflects a geminate [kʷ:] which had arisen diachronically from *ḱw, but that doesn't mean that a sequence [k⁽ʰ⁾ +w] synchronically produced at a morpheme boundary should be expected to give the same result.

Also worth noting is the dialectal variant ἴκκος, preserved in a gloss, which seems to point to a different dialectal treatment of the geminate.

Finally, though it's true that the initial vowel and breathing of ἵππος are a longstanding problem, they've been recently argued to result from regular sound change. If this is correct it obviously strengthens the *ḱw > pp theory.