Joseph Greenberg believed that the native languages of the Americas fell into three macro-families: Eskimos-Aleut, Na-Dene (mainly the Athabaskan languages, spoken in Western Canada and the American Southwest), and Amerind (all of the rest). Often this theory was paired with the notion of three separate migrations: first the Amerinds, then the Na-Dene, then the Eskimo.
It does indeed look as if there were three separate migrations – another gem from Reich and Patterson. The study was marred by complete non-cooperation from Amerindians in the US, but they managed. The Canadian Na-Dene tribe they looked at was mostly Amerind but had about 10% of the second wave. The Eskimos were about half Amerind and half the third wave. I would guess that Na-Dene groups such as the Navaho and Apache have a higher fraction of that second wave, because they simply look more like modern East Asians, as do the Eskimo.
Both of the later waves seem to have started out settling areas in the far North, regions that became habitable late and where settlment required specialized techniques. Interestingly, the genetic sample from the old Saqqaq culture in Greenland looks to be almost entirely second wave ( 85%) – so some people linked to the Na-Dene likely occupied all of the Arctic coast of North America before the Eskimo. Curiously, the Eskimos don’t show a noticeable fraction of second-wave ancestry. They must have replaced them without much admixture. Perhaps more detailed studies will show a bit. Lesson: when one human population expands into another, there is almost always some admixture (enough to transmit adaptive alleles, certainly) but there isn’t always much. Laurent Excoffier thinks that such events are always similar to a ‘range expansion’ – diffusion and random mating – but that happens not to be the case. Not when AMH met Neanderthals and Denisovans, not in Neolithic Europe, not in the Philippines, not in the Americas.
Fairly recently, Na-Dene has been linked to a particular family of Siberian languages, Yeniseian, using standard linguistic techniques rather than Greenberg’s multiple comparisons approach. Of course, Merrit Ruhlen, a student of Greenberg, had made the same connection earlier using multiple comparisons, while Alfredo Trombetti suggested the same thing 90 years ago.
Most linguists studying Amerindian languages disagreed with Greenberg. They thought (and still think) that it is impossible to identify a language macro-family that originated as much as 15,000 years ago. They also to tend to dismiss evidence from genetics, not least because they don’t know any.
It is conceivable that Greenberg’s method is invalid, and that it repeatedly gave the right answer (on the three-wave settlement of the Americas, on Dené–Yeniseian, and in his generally accepted classification of the African languages) by chance. Maybe Joe Greenberg was just lucky.
Certainly his critics aren’t.