I've just read
this article by Melchert, which reviews the Luwian and Lycian evidence for three dorsal stops. His conclusion is pretty remarkable:
Melchert wrote:The unconditioned three-way contrast I claimed in 1989 for the voiceless dorsals in Luvian and Lycian is to be rejected. Luvo-Lycian is rather the “mirror-image” of Albanian: before the merger of the front and non-front velars, the voiceless front velar, but not the non-front velar, underwent conditioned palatalization. A similar conditioned palatalization of just voiced front velars is also possible, but cannot at this point be affirmed. Thus Anatolian is in terms of the traditional classification “centum”: it does show an eventual phonological merger of the front and non-front velars.
The conditioned palatalization he's speaking of is the sound change k' > ts before {i, e, e:, y, w}, but k' > k elsewhere. Back velars (and in Luwian labiovelars) didn't show this conditioned palatalization, however there are just a few (albeit strong) pieces of evidence for this: *ke(:)s- > kisa:- and *kwi- > Luwian kui- Lycian ti-.
Remarkably, the palatalization also occurs before /w/, e.g. *ek'wo- > azuwa- and *k'won- > zuwani-. Melchert also remarks that 'there's a strong tendency of labiovelars to be palatalized" and references (Hock 2009) for that conclusion. We know this conditioned palatalization of labiovelars occurs in Albanian and Greek, but I'm wondering about other examples, so I'm going to try to find that Hock article.
I'm mentioning this as there is no reason to assume that /k'w/ and /k
w/ are less likely to show fronting than simple /k'/ or /k/, in fact they are more likely to show palatalization, at least before front vowels. However, this doesn't go against KathAveara's proposal for the three-dorsal series /k q k
w/, in fact, it supports it: the fact that the labiovelars in Albanian show conditioned palatalization shows that the merger of back velars and labiovelars in Satem languages was rather late. Similarly does the Luwian evidence show that the conditioned palatalization of front velars happened before the merger of front and back velars in Centum languages. The merger processes in the dorsal series were rather late, and a late fronting of /q/ would be a neat explanation for this.
To summarize, and starting from a three dorsal series /k q k
w/, the order of post-PIE sound changes affecting the dorsal series:
1a. Conditioned palatalization of front velars in Luwo-Lycian, k > ts before front vowels and /w/.
1b. Conditioned palatalization of labiovelars in Albanian (something like k
w > ts
w before front vowels)
2. Unconditioned palatalization of front velars in Satem languages (maybe inhibited in some environments), k > ts, s , s' or the like.
3. Loss of labial element in labiovelars in Satem languages, k
w > k
4. Unconditioned fronting of back velars everywhere, q > k, which leads to a merger of front and back velars in Centum languages and of velars and labiovelars in Satem languages,
In order for this to work, the dialects must still form a continuum, and Anatolian must already have splitted into at least two branches, as sound change 1a. does not affect Hittite and Palaic. Otherwise it's neat: I especially like it how the gap of /k/ left in 2 is filled by 3, probably immediately.
To summarize the summary: both the Luwian and Albanian developments speak in favour of KathAveara's proposal, which starts to look better by the minute, despite my earlier objections.
(I also just realized I've always spelled "palatalization" wrongly as "palatization". Live and learn.)