Re: [tied] Proto-Steppe Numerals

From: Piotr Gasiorowski
Message: 3268
Date: 2000-08-19

I don't know what's eaten my response to Glen's long posting on Steppe numerals. I'm sending it again just in case -- sorry if you get it twice. The original posting is here (reasons of space):
 
http://www.egroups.com/message/cybalist/3239
 
 
 
Dear Glen,
 
First, thanks for your long and detailed reply. As I find it necessary to answer point for point, this posting will be lengthy as well. My apologies to Cybalist if it develops into an essay.
 
Your explanation of the relation between '3' and '9' in Japanese in ingenious and I have no difficulty accepting it. However, some problems remain: what you gain by solving this problem is a rough match between Japanese, Mongolic and Finno-Ugric, with Samoyedic, Turkic, Tungusic, Korean, Nivkh, Yukaghir, Chukchi-Kamchatkan and Eskimo-Aleut showing apparently unrelated forms and the alleged Indo-Tyrrhenian cognate giving you headaches, as you yourself admit. As the Proto-Japanese numeral is somewhat conjectural, everything really hangs on the agreement between Mongolic gur- and Finno-Ugric kol(me) -- this is in fact the REAL basis for your kWul(mu). You seem to believe that this kind of double agreement (involving '3' and '4') between two languages (Proto-Finno-Ugric and Proto-Mongolic in this case) is unlikely to be mere coincidence. But if you compare, say, the numerals of PIE and Proto-Quechua, you'll find even more spectacularly similar words for '5' and '6'; and if you take Proto-Austronesian and PIE, you'll find the correspondence between the words for '2' and '3' in both families quite literally too good to be true. In other words, you fail to show that the degree of similarity on which you build your reconstruction exceeds that which could occur by chance.
 
By "reaching down", i.e. using forms not from proto-languages but from their arbitrarily chosen daughters you depart from the normal procedure of the comparative method and greatly increase the probability of a chance match, as in Greenberg's mass comparison (especially when you work with very short roots such as *ra or *na). Frankly, I still can't see why Samoyedic should be less likely than Finno-Ugric to have retained Proto-Uralic forms. After all, Tungusic is also spoken way up north, and yet you don't hesitate to use exclusively Tungusic evidence (ignoring Mongolic and Turkic, not to mention the rest of Steppe) for some items. To give a concrete example, Tungusic is your only witness for '6' (with some feeble support from Japanese).
 
Your admission that there is NO ATTESTATION AT ALL for the reconstructed numeral '8' is staggering. What's the point of reconstructing it at all, then? As a matter of fact, you abandon the comparative method entirely in this case and simply project your Nostratic form onto Proto-Steppe by "inverted reconstruction". Since that word does not exist in the Steppe languages, a suspicious question arises: what is the Nostratic reconstruction based on? A naive reader might hope that there is at least some non-Steppe comparative justification for *munri (or the like), but as far as I can see, the corresponding Sumerian, Dravidian and Afroasiatic numerals (if reconstructable at all) can hardly be cognates. From what evidence and by what esoteric method did you distil your Nostratic reconstruction then? The answer is the same as before: you project a still more ancient (Dene-Caucasian) form onto Nostratic -- ALL THE TIME WITHOUT A SHADOW OF EVIDENCE. So here we have a Dene-Caucasian word passed on to Nostratic and then to Steppe without leaving any tangible traces in any of their daughter branches.
 
You promise there is a long story behind it all. I'd be interested to hear it, but I'm not actually holding my breath. Ironically, you seem to have overlooked the single Steppe '8' word similar to *munri - I mean Nivkh minr (Nivkh is the currently preferred term for what you call Gilyak). But I wouldn't get too excited about this word. Your "Steppe" contains several hundred languages. It would be a real miracle if never a one of them had something similar to a prespecified "protoword".
 
If your aim is to demonstrate distant relationship you should claim no data that violates the proposed sound correspondences. Of course actual language change is often irregular, but the requirement of admitting only regular sound change in TESTING a relationship scheme is a sound methodological principle which reduces the danger of chance resemblance. If you relax it ad hoc (as you do for the numerals '3' and '4' in your response), you sabotage your own project. In the case at hand you make things far worse by relaxing semantic constraints at the same time, as when you allow the '5' word for to mean '4' or '6' (even with so much latitude you can't get an unambiguos reconstruction for '5' and have to be content with two different forms in free variation). Of course if you were more rigorous you'd get no matches -- but that's the whole point. If you relax the requirement of regularity you get SPURIOUS matches that can't be distinguished from genuine ones. Again, it's a suicidal departure from the straight and narrow in comparative analysis.
 
Your protoword for '5' (*kit:u/*kWut:u) develops the meaning '4' in one branch (Indo-Tyrrhenian) and '6' in another (Uralic). Oritur quaestio, dominus Glen: is there a Steppe branch in which it means just what it's supposed to mean? Your answer is "Altaic" -- but in actual fact you only mention Old Japanese. Let's look at the words for '5' elsewhere in Altaic: Mongolic *tabun, Tunguzic *tuNga, Korean tasôt, Old Turkic beS. No good. How about extending the search? If we look farther afield (and I've checked Nivkh, Samoyed, Yukaghir, Eskimo-Aleutian and Chukchi-Kamchatkan, lest you should think I'm lazy), *kit:u/*kWut:u just can't be seen anywhere in the wide Steppe. Is it just another Nostratic (or Dene-Caucasian) ghost?
 
It becomes clear that despite paying lip-service to the comparative method you don't actually apply it in reconstructing the Steppe numeral system. You POSTULATE a list of numerals justified mostly by external considerations (namely, by what you believe the Nostratic or Dene-Caucasian forms to have been like). There are isolated matches or near-matches here and there (quite naturally, with so many language groups to choose from) and you claim them as supportive evidence, but even if there are no matches at all you aren't greatly worried. After all, you know in advance what the protoforms should be -- as if Nostratic or Dene-Caucasian reconstructions were more solid than Steppe ones.

I understand your private methodology well enough to see that it's self-defeating. By saying that it's better to reconstruct ANYTHING rather than nothing at all you ignore the distinction between a serious proposal and unfounded speculation. A methodology that ALWAYS allows you to reconstruct SOMETHING would yield a phantom reconstruction of a non-existent protolanguage even for entirely unrelated languages (such a thing would apparently be "better than nothing"). You present a sketchy and highly speculative reconstruction and then challenge other people either to accept it or to offer a superior hypothesis -- in other words, you claim null hypothesis status for your reconstruction (as a reward for pioneering work?) thus evading the burden of proof: "Got a better one? No? Then I win."
 
This is a misunderstanding of scientific methodology. The null hypothesis must always be as conservative as possible in order to exclude unwarranted inferences. Thus, you can't start by ASSUMING that the Steppe families are related and what remains to discovered is just HOW they are related. Such a proposition could never be falsified, for it cannot be proved that two languages are unrelated: the evidence of their genetic relationship may have been obliterated by historical change. The initial null hypothesis must assume their NON-relatedness -- only then can we hope to make some progress by trying to falsify it. You can replace it with a new null hypothesis only if you manage to demonstrate that all or some of the languages being compared show systematic resemblances which exceed non-genetic "background similarity" (due to borrowing, chance, etc.). The QUALITY of proposed cognate words and morphological paradigms is much more important that their number. Their DISTRIBUTION is important (in the initial testing of a hypothesis one should exclude cognates that can't be found in most of the daughter groups). If you postulate a model of relationships which is not solid enough (and even professionals sometimes indulge in such games out of impatience), you won't make it better by "extrapolation and evolution"; you'll just move in circles instead of making progress.
 
We are all lumpers to some extent; otherwise who would believe in Proto-Germanic or PIE? Lumping is a healthy activity: you challenge the current null hypothesis (which assumes by default that languages are NOT related unless demonstrated otherwise) and sometimes you win. Do you think I've never spent evenings studying Etruscan or Uralic word-lists and paradigms? I'm generally sympathetic to your efforts and I hope you'll see my criticism as friendly and constructive even if you reject much of it. At any rate stop seeing the ugly snout of politics everywhere.
 
Cheers,
 
Piotr