Groucho Marxism

In a previous blog post I argued that Proto-Indo-European (PIE) had two types of velar in word-initial position: plain velars and labiovelars. In the satem languages, labiovelars were delabialized and plain velars were palatalized apart from after *s, and before *r, *s, and *H₂; palatalization was also blocked before *l in Albanian and *lo in Balto-Slavic. In the centum languages, labiovelars were delabialized before *u, and labiovelars were also delabialized before *o and before consonants in the western languages. This raises the question whether the two-velar hypothesis holds in word-medial position too. This is a more difficult question to answer, as there are more environments word-medially that need to be considered.

In his 1973 PhD thesis, Lars Steensland showed that in word-medial position, palatovelars occur everywhere apart from after non-syllabic *n; plain velars occur everywhere apart from after *e and syllabic *n; and labiovelars only occur after *e, *i, *o, and *n. Steensland also showed that only plain velars occur in words beginning with *(s)t, whereas only palatovelars and plain velars occur word occur in words beginning with *(s)m. This includes all of his examples of words with a plain velar after *i, and most of his examples of words with a plain velar after *o. The obvious explanation is that PIE had just plain velars and labiovelars in word-medial position, and the latter were delabialized in the Centum languages apart from after *e, *i, *o, and *n, and in words beginning with *(s)m.

The problem with this explanation is that it implies that the labiovelar developments described must have happened separately in all of the centum languages. However, a closer look at the data suggests we do not need to assume this. Let’s start with the position after *H₂. Steensland provides two examples of words where a plain velar occurred in this position, both of which are derived from the root *bheH₂g- ‘share’: *bheH₂geti ‘share (verb)’, and *bheH₂gos ‘share (noun)’. The plain velar in these words is reconstructed on the basis of Greek reflexes meaning ‘to eat’, with a semantic shift: ‘I received a share’ > ‘I consumed’ > ‘I ate’. This is clearly not sufficient evidence to conclude that plain velars occur after *H₂, and we therefore need not assume that labiovelars were delabialized in this position.

Steensland demonstrated that labiovelars do not occur after *u in PIE, but both palatovelars and plain velars do occur in this position. This is consistent with the two-velar hypothesis if we assume that plain velars were palatalized in this position in the satem languages and labiovelars delabialized in this position in the centum languages, which is plausible. Steensland provides two examples of words with a plain velar after *r: *H₂uergeti ‘turn’, and *(s)morkos ‘decayed’. The plain velar in the former may have been blocked from palatalizing in the nasal variant *(H₂)urengeti, whereas latter word is highly doubtful and may therefore be disregarded. Thus, we need not assume that labiovelars were delabialized after *r in the centum languages either.

There is one example of a word with a plain velar occurring after *l: *uolkos ‘wet’. As this word only has reflexes in the western languages, it may have originally contained a labiovelar which was subsequently delabialized before the *o. There are five examples of words with a plain velar which begin with *(s)m: *moghtis ‘might’, *muktis ‘freedom’, *(s)morkos ‘decayed’, and *H₃mighleH₂ ‘mist’. As the first word only has reflexes in the western languages, it may have originally contained a labiovelar which was subsequently delabialized before the consonant. Similarly, the second word may have originally contained a labiovelar which was subsequently delabialized after the *u. The third word may be disregarded, as noted above.

The plain velar in the fourth word, *H₃mighleH₂ ‘mist’, is more difficult to explain. The word is derived from the root *H₃meigh- ‘drizzle’, which must surely be related to the root *H₃meigh’- ‘urinate’. We may hypothesise that they are in fact the same root, and that palatalization was blocked in descendents of the former but not the latter. As *H₃mighleH₂ is the only reconstructable word derived from the first root, we can posit that palatalization was blocked before the following *l. This works fine for the Albanian reflex, as palatalization was blocked before *l in Albanian. To explain the Indo-Iranian and Balto-Slavic reflexes, we must assume that palatalization was blocked before *leH₂ (> *laH₂) in these languages; this is rather ad-hoc, but aligns with palatalization being blocked before *lo in Balto-Slavic.

Thus, except for the position after *u, there is no evidence that labiovelars were delabialized word-medially in the centum languages, other than in the positions in which they were delabialized word-initially. We are not quite finished yet though, as there are still a few words with plain velars occurring in word-medial position that require an explanation. The first is *tuekos ‘body’, which is reconstructed as *tuakos by Steensland, but the *a is based on a Greek reflex which is probably unrelated. We may posit that this word originally contained a labiovelar which was subsequently delabialized in the zero grade; this zero grade is unattested, but that is not surprising given that underlying the root *tuek- is barely attested anyway.

The plain velar in *H₂uogseieti ‘enlarge’ probably represents an original labiovelar which was delabialized by the following *s, as the word only has reflexes in Germanic among the centum languages; alternatively, it may have been delabialized in derivatives of the root variant *H₂eugʷ-. On the other hand, the plain velar in *uogsos ‘wax’ probably represents an original plain velar which was blocked from palatalizing after *s in the metathesized form *uosgos from which the Balto-Slavic reflexes are derived. The word *knH₂kos ‘safflower’ is clearly related in some way (via an adjective ‘honey-coloured’) to the word reconstructed as *knH₂onks ‘honey’, where the plain velar would have been blocked from palatalizing after the non-syllabic *n.

That just leaves one word with an unexplained plain velar: *loghos ‘lair’. The Greek reflex of this word precludes an original labiovelar, which makes accounting for the lack of palatalization in the Slavic and Albanian reflexes difficult. The absence of a palatalized reflex in the Slavic can be explained by a rule set out by Antoine Meillet in 1924 which states that in Slavic, palatalization was blocked in words containing *s. Two of the words used by Meillet to reach this conclusion – *gh’eH₂ns ‘goose’ and *k’oseH₂ ‘hair’ – also have unpalatalized velar reflexes in Albanian, suggesting that the same rule operated in that language as well. If so then we have accounted for all the data, which demonstrates that the two-velar hypothesis holds in word-medial position too.