Groucho Marxism

In a previous blog post I explained how Proto-Indo-European (PIE) is usually reconstructed with three types of velar constant, which are referred to as palatovelars, plain velars, and labiovelars. I also mentioned that in 1973, Lars Steensland demonstrated a complimentary distribution between the three types of velar. Steensland first showed that in word-initial position, only plain velars occur after *s, except before *i where only palatovelars occur. He then showed that in word-initial position not following *s, palatovelars occur everywhere apart from before *r and *s; plain velars occur everywhere apart from before *e, *i, and what we would now call *h₁; and labiovelars only occur before *e, *i, *r, and what we would now call *H₁.

Next, 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. Thus, in word-initial position, there isn’t a single environment where all three types of velar occur; and in word-medial position, all three types of velar only occur after *i and *o. Steensland also showed that only plain velars occur word-medially in words beginning with *(s)t, whereas only palatovelars and plain velars occur word occur word-medially in words beginning with *m. This includes all of his examples of words with a plain velar occurring after *i, and most of his examples of words with a plain velar occurring after *o.

The obvious explanation for this distribution runs as follows. PIE originally had two types of velar: a palatovelar and a labiovelar. In the satem languages, plain velars became palatalized in word-initial position in all environments apart from (1) after *s, unless followed by *i; and (2) before *r and *s; and in word medial-position in all environments apart from (1) after non-syllabic *n, and (2) in words beginning with *(s)t. In the centum languages, labiovelars were delabialized in word-initial position in all environments apart from before *e, *i, *r, and *H₁; and in word-medial position in all environments apart from (1) after *e, *i, *o, and *n, except in words beginning with *m. There is a problem with the latter part of this explanation however.

As I noted in my previous blog post, whereas the satem languages were geographically contiguous so could have undergone such changes together, the centum languages were not. The explanation above implies that the labiovelar developments described must have happened separately on seven separate occasions: in Anatolian, Tocharian, Greek, Germanic, Italic, and Celtic. This is implausible and removes one of the key arguments in favour of the two-velar reconstruction. However, a closer look at the data suggests we may not need to assume that all these developments occurred in all the centum languages. To see this, let us consider one environment where delabialization seems particularly common: word-initially before before *o.

Steensland provides 20 examples of words with a plain velar occuring before *o in word-initial position, of which 11 only have reflexes in the western languages (Celtic, Germanic, and Italic). Of the remaining 9, three are derived from roots with a mobile *s, which accounts for the lack of palatalization in the satem reflexes. One of the remaining six, *kolH₂os ‘strike’, may be derived from a root *(s)kelH₂- ‘cut’, and another, *korios ‘army’, may be derived from a root *(s)ker- ‘cut’, both with a mobile *s. One of the remaining four, *ghordhis ‘garden’, is derived from a root *g’her- ‘garden’, which displays gutturalwechsel in the satem languages, suggesting that the word originally had a plain velar that was blocked from palatalizing in the zero grade *g’hr-.

Another, *koH₂peH₂ ‘handle’, is derived from a root *keH₂p- ~ *kH₂ep- ‘grab’, where palatalization may have been blocked before *H₂. The plain velar in one of the remaining two words, *kom ‘with’, is based entirely on Albanian kë, which is now usually derived from the interrogative root *kʷ-. The final word, *kom, is a particle of unclear meaning and etymology which apparently only has reflexes in Anatolian among the centum languages, and even that isn’t certain, so we cannot be sure that this word contained a plain velar. Thus, to explain the data, we only need to posit that delabialization occurred before *o in the western languages – Germanic, Italic, and Celtic -rather than all of the centum languages. This seems a lot more plausible.

One potential counterexample is *gʷous ‘cow’, which yields labial reflexes in Italic (Umbrian bum, Latin bōs)  and Celtic (Irish bó, Welsh  bu). One way around this is to posit a proto-form *gʷeH₃us and assume either that the delabialization before *o  occurred before the colouring of *e to *o in these languages. The problem with this is that the colouring of *e to *o is usually dated all the way back to PIE, or at least to a period very soon afterwards. Instead, we can simply assume that the change *gʷ > *b occurred in Osco-Umbrian and Celtic prior to the delabialization before *o in these languages (Latin bōs is known to be a borrowing from  Osco-Umbrian). This seems a better explanation as it does not create any issues from a chronological perspective.