Groucho Marxism

In a series of recent blog posts I have argued that Proto-Indo-European (PIE) had just two types of velar consonant: plain velars and labiovelars. It has been claimed that there is evidence for three distinct types of velar being preserved in Albanian and in the Luvian branch of Anatolian. The theory that three types of velar remained distinct in Albanian was first put forward by Holger Pedersen in 1900, according to whom labiovelars were palatalized before (PIE) front vowels, whereas plain velars were not. Specifically, Pedersen argued that before front vowels, there is a threefold reflex in Albanian of palatovelars, plain velars, and labiovelars, as follows: *k’ > Albanian th, *k > Albanian k/q, *kʷ > s; and *g’,*g’h > Albanian dh, *g,*gh > Albanian g/gj, *gʷ,gʷh > Albanian z.

Pedersen’s most convincing examples of the labiovelar developments are: *penkʷe ‘five’ > pesë, *kʷiH₁ ‘how’ > si, *H₃okʷiH₁ ‘eyes’ > sy, *gʷherm- ‘heat’ > zjarm, and *kʷel- ‘turn’ > sjell. The main examples of words supposedly containing plain velars that were not palatalized by a following front vowel are: *ghedh- ‘find’ > gjej, *suerghe- ‘be ill’ > dergj, *Horghi ‘nit’ > ergjëz, *suolke- ‘pull’ > heq, *keH₁s- ‘time’ > kohë, *kel- ‘drive’ > qell, and *keid- ‘cut’ > qeth. The problem with Pedersen’s theory is that, as Lars Steensland demonstrated in 1973, plain velars do not occur before front vowels in PIE. Or, more accurately, plain velars were subsequently palatalized before front vowels in the satem languages. This suggests the unpalatalized velar must have been introduced analogically in at least some of these words.

The unpalatalized velar in gjej ‘find’ may have been taken from the nasal variant *ghnedh- ~ *ghndh-, as palatalization was blocked before resonants in Albanian. Similarly, the unpalatalized velar in qell ‘drive’ may have been taken from the zero grade form *kl-. The unpalatalized velars in dergj ‘be ill’, heq ‘pull’ and kohë ‘time’ can be explained by noting that these words are derived from a root which originally contained an *s; as I noted in a previous blog post, there is evidence that palatalization was blocked in words containing  *s in Albanian. The supposed reflexes of *Horghi ‘nit’ cannot be regularly derived from it, so this root probably did not exist in PIE. Finally, the word qeth ‘cut’ is not probative as the underlying root is actually *kH₂eid-, so the velar never stood before a front vowel.

The theory that the three PIE velar series remained distinct in Luvian was first put forward by Craig Melchert in 1987, who originally claimed that *k’ > Luvian z, *k > Luvian k, and *kʷ > Luvian ku. Melchert (2013) subsequently revised his claim with the assertion that *k’ > Luvian z only before front vowels, non-syllabic *u, and syllabic resonants, with *k’ > Luvian k elsewhere. To demonstrate that the three PIE velar series remained distinct in Luvian, it must therefore be shown that *k > Luvian k before front vowels, non-syllabic *u, and syllabic resonants. The main examples adduced by Melchert to support this claim are: *kes- ‘scrape’ > kišanda, *krs- ‘cut’ > karš-, and *tuke- ‘statue’ > tukedre. None of these examples holds water.

The unpalatalized velar in kišanda ‘scrape’ may have been taken from the zero grade form of the root *ks- (contra Melchert). The unpalatalized velar in karš- ‘cut’ can be explained by noting that the underlying root was *(s)ker- with a mobile *s and positing that palatalization was blocked by the preceding *s, as it was in the satem languages. Alternatively we can note that Melchert only provides one example of *k’ > z syllabic resonants, namely *k’rH₂- ‘horn’ > zurnid, where  the palatalization could easily have spread from the well-attested *e-grade form *k’erH₂-. Thus, the change *k’ > z before syllabic resonants is not well-established. The velar in the third word, tukedre ‘statue’, could represent an original labiovelar which was delabialized by the preceding *u.

In conclusion, the claims that three distinct types of velar were preserved in Albanian and in the Luvian branch of Anatolian do not stand up to scrutiny. There is no evidence from these languages which disproves the hypothesis that PIE only had two types of velar constant: plain velars and labiovelars.