Undermining the Hybrid Hypothesis

It was only about six weeks ago that I linked to the paper on the new hybrid theory of the origins of Indo-European Languages. (Link to the paper there.) I mentioned there that I had previously favored the steppe hypothesis, but this paper provided strong evidence for the Anatolian being at least part of the story. 

(If you don't want to read all the internal links in this essay, just go to the next one only. Unless it isn't available to you and is subscriber only, in which case, well, I don't know. Cut-and-paste of a section of a PDF file can result in a run-together text where there is no spacing between words, let along sentences and paragraphs.  We'll see.)

Not so fast.  David Anthony, author of The Horse, The Wheel, and Language (which is now the standard text on the Indo-Europeans and Yamnaya), which I have also referenced a few times, is not convinced, and he lists others who are not convinced as well. (Re: The transcript of the interview with Razib Khan, at about the 22 minute mark.) The main problem seems to be that the new paper is using a method which has worked for biological trees of descent, but gives odd numbers when applied to languages.  This is because it intentionally discards any linguistic paleontology. That would be the interpretation of reconstructed and protolanguages to discern where they were and what their culture was.  If you have taken any historical linguistics you have seen this done for Indo-European. There are words that occur in widely scattered daughter languages that are very similar and have similar meanings, such as types of trees and animals that have ranges (such as honeybees, birches), words for farming but not metallurgy, lakes but not oceans, patrilineal (vs "all other") descent of property. Sanskrit and Icelandic have some words with obviously similar roots. 

This becomes key in the Indo-European splits, because the whole idea of wheeled vehicles pulled by oxen and spreading across the Steppe north of the Caucasus is foundational to the idea of Yamnaya and Corded Ware spread. Kyklos - words, such as circle, cycle in English are in all the daughter languages after a particular split.  But there are archaeologists who reject all such evidence, stating you can only get a vague idea of "thing that turns" from the PIE word, not the specific idea of an axle. The linguists counter that the derivative words do in fact mean "axle" in the daughter languages as soon as we can pick them up in writing. It is ridiculous to suppose that they all independently settled on that vague "turn" root to describe an axle. That it would happen a few times seems possible.  Every time, no. Language is never so orderly even on much shorter time scales. British and American English, starting from a common origin, did not make identical decisions over the centuries.  Heck, Canadian and American English didn't make the same decisions. 

I find David Anthony's arguments about how the splits cannot have occurred before 4500 BC as they claim, are simply devastating. Even an amateur like me can pull out a fair bit of "But wait, there was no evidence of Balto-Slavic split in the archaeological record for another thousand years, right?  Have I got that wrong?"

The skeptics insist we can't know, not for certain, even if many guesses turn out to be correct. And they like their own tools, such as Bayesian reconstructions, much better than the subjective interpretations of whether equus and yakwe (thousands of miles apart) both not only accidentally mean "horse," but mean "horse" because of descent from a common PIE word.  The argument is nicely summarised by someone who thinks both of the groups are being a bit extreme and difficult.  Even with that split-the-difference offering, however, I am back to siding with Steppe Hypothesis.  When I passed along the hybrid theory a few weeks ago I did not realise that they had simply discarded all linguistic paleontology from their Bayesian method.  As with chaos theory, there is a sensitive dependence not only on initial conditions, but all conditions under all outside influences when you elect to live or die by Bayesian methods. Electing to discard a common practice in a related (competing?) field just because you don't think they have proven their method quite enough is just being difficult.

Applying Bayesian methods to Bayesian decision-making itself would grant that the method you don't like so much nonetheless has some likelihood of being correct, and thus deserves some weight.