Cowgills law (in Greek 1 ) states that the PIE vowel *o, of whatever origin, develops into Greek u when flanked by a labial (*m, *b, *bh, *p, *g w , *k w , *gh w , *w, *h3) and a resonant (*m, *n, *l, *r), or vice versa.
Some classical examples: PIE meaning Greek *nk w t-s night nks *bhli -om leaf phllon *ml-eh2 mill mle *h3ngh w -s nail nuks
The law is clearly invalid as stated. There are countless counterexamples, including such emblematic Greek phoneme sequences as /pol/ (2075 dictionary entries in a Perseus Greek Dictionary Headword Search) or /phor/ (1378). Even allowing for compound words, repetitions and entries involving morpheme boundaries, it is clearly the case that not every *o in Greek turns into u in the given environment.
On the other hand, there may be some cases of Greek *o > u where the vowel does not stand between a labial and a resonant, or a resonant and a labial. I think a pretty good case can be made for the preverb/preposition sun(-) with, which shows a number of exact lexical correspondences with *som- or *kom- in other Indo-European languages. Initial s- is explained by the original shape ksun(-), which may simply be an amalgamation of *kom with *som > *ksom. Unexpected u then reflects the working of Cowgills law 2 .
So, while it is undeniable that some PIE *os give u in Greek, the conditions under which this happens are far from clear. It has been suggested that apophonic *o is not input for Cowgills Law (see for instance this blogpost by Christopher Culver), but in that case, what about G. mle, from a seemingly ablauting verbal root *mol- ~ *mel-?
My suggestion for a solution involves getting rid of two of the conditions stated in the first paragraph: of whatever origin, and resonant. As I will try to demonstrate below, only PIE *os of a specific origin turned into u in Greek, and only when in the neighbourhood of a labial consonant (*m, *b, *bh, *p, *g w , *k w , *gh w , *w, *h3).
The Paradebeispiel for Cowgills law in Greek is arguably the word nks night, and this same word may also offer an insight into the exact circumstances under which Cowgills law operated. As I have argued elsewhere, I believe that PIE had four series of tectals: velar *K, uvular *Q, labialized velar *K w and (marginally) labialized uvular *Q w , corresponding to traditional *K , *K, and *K w (twice). Originally there had been only two series, the velar and the uvular, and the labialized series had arisen through labialization caused by neighbouring pre-PIE **u or **.
In the word for night, the labialization of the velar is most economically explained if it was caused by a preceding **. The word can then be reconstructed as **ngt- 3 > *nok w t-. In Greek, long ** developed into u here instead of o, because of the labial consonant following it (which had been
1 There is another Cowgills law in Germanic, which concerns the hardening of *h3 to *k w (as in quick). 2 Cf. also Hom. ksns common < *ksom-is (with Cowgills law), vs. koins id. < *kom-is (without it). 3 I prefer **ngt over **nkt, as the latter would in my opinion have given *noh3t- caused by ** itself in the first place). The presence of u, finally, caused the following labiovelar to lose its labiality, and the result is Greek nukt-.
The same argument applies more or less to G. nuks nail. We can reconstruct pre-PIE **fngh- > PIE *h3nogh w -, but pre-Greek *h3nugh w -. Paradigmatic levelling between NA *nukh- (with Saussures effect) and oblique *onkh w - ~ *onukh- (with prothetic vowel) must have produced attested nux, nukh- 4 .
Another case where we can deduce the presence of pre-PIE ** from the following consonant is the word for name, PIE *h1nh3mn , G. *numa, numa, noma. If *h3 is the labialized counterpart of *h2 (phonetically / w / vs. //), then we can reconstruct something like pre-PIE **dnx-man-, giving pre-Greek *n
ma, obl. *enmat-, with further diverse developments according to dialect.
In the other cases where we have evidence for Cowgills law, the origin of *o (G. u) out of earlier ** cannot be demonstrated directly, but can be assumed on the strength of the evidence provided by night, nail and name. That would allow us to reconstruct pre-PIE **bhl-, **ml-, **(k)sm-, etc.
In the majority of cases, where *o in a labial environment gives G. o, the origin of the *o can be assumed to have been pre-PIE ** or a secondarily lengthened pre-PIE **.
Finally, if Greek really provides evidence for a different treatment of PIE *o from ** vs. PIE *o from **, then we should be using different symbols in our PIE reconstructions. My suggestion would be *:
PIE meaning Greek *nk w t-s night nks *bhli -om leaf phllon *ml-eh2 mill mle *h3ngh w -s nail nuks
4 Thereby eliminating an awkward homophony between the nominatives of nail and night.