The Great Proto-Indo-European Thread's Sequel

[bradrn]

As someone who is professionally used to work with noisy data (specifically from radar)

Oh? I’m doing lidar; always interested to hear how related fields approach things.

it saddens me that comparativists can't work with noisy (a.k.a. dodgy) data.

The problem is, in engineering terms, the signal-to-noise ratio. Given a certain rate of lexical replacement and irregular change, there is a time period beyond which there is simply not enough signal to compare languages reliably.

[Richard W]

The problem is, in engineering terms, the signal-to-noise ratio. Given a certain rate of lexical replacement and irregular change, there is a time period beyond which there is simply not enough signal to compare languages reliably.

One can work with quite low signal-to-noise ratios. I've seen angle tracking demonstrated down to −30dB; the practical problem there is knowing that one is still tracking something,. One ought to be able to work with a set of alleged cognates where 75% are not.

[WeepingElf]

it saddens me that comparativists can't work with noisy (a.k.a. dodgy) data.

The problem is, in engineering terms, the signal-to-noise ratio. Given a certain rate of lexical replacement and irregular change, there is a time period beyond which there is simply not enough signal to compare languages reliably.

Exactly. There is a limit to the comparative method: it eventually runs out of cognates, or rather, the cognate sets multiply and shrink the deeper you look, until each set consists of only one word, at which point it becomes indistinguishable from chance resemblance. The limit of detectability of a relationship is perhaps about 10,000 or 15,000 years; Afroasiatic gets close to this limit (there are currently two major reconstructions of Proto-AA, which disagree in many points and therefore cannot both be correct, which likely means that both are at least in parts wrong), and Mitian perhaps lies narrowly within this limit, too. It is a bit like C14 dating, which cannot date samples more than about 50,000 years old because beyond that age, there is no measurable quantity of C14 left in the sample.

[bradrn]

The problem is, in engineering terms, the signal-to-noise ratio. Given a certain rate of lexical replacement and irregular change, there is a time period beyond which there is simply not enough signal to compare languages reliably.

One can work with quite low signal-to-noise ratios. I've seen angle tracking demonstrated down to −30dB; the practical problem there is knowing that one is still tracking something,. One ought to be able to work with a set of alleged cognates where 75% are not.

The problem is that it’s not 75%: it can easily be more like 95% non-cognate or more. Also, unlike the case you mention, you don’t necessarily know if they are cognate in the first place — indeed that’s precisely the thing you’re trying to prove.

[Richard W]

The problem is that it’s not 75%: it can easily be more like 95% non-cognate or more. Also, unlike the case you mention, you don’t necessarily know if they are cognate in the first place — indeed that’s precisely the thing you’re trying to prove.

Well, it shouldn't be. You should be asking whether there are cognates (sensu stricto) there. The answer will generally be probabilistic.

[bradrn]

The problem is that it’s not 75%: it can easily be more like 95% non-cognate or more. Also, unlike the case you mention, you don’t necessarily know if they are cognate in the first place — indeed that’s precisely the thing you’re trying to prove.

Well, it shouldn't be. You should be asking whether there are cognates (sensu stricto) there. The answer will generally be probabilistic.

I’m not sure I understand your meaning there.

[Brilliant-Resource14]

Well, it shouldn't be. You should be asking whether there are cognates (sensu stricto) there. The answer will generally be probabilistic.

... what?

[Travis B.]

The problem is that it’s not 75%: it can easily be more like 95% non-cognate or more. Also, unlike the case you mention, you don’t necessarily know if they are cognate in the first place — indeed that’s precisely the thing you’re trying to prove.

Well, it shouldn't be. You should be asking whether there are cognates (sensu stricto) there. The answer will generally be probabilistic.

What do you even mean here?

[Talskubilos]

You can always look at relating PIE and Afroasiatic (AA) as a prelude to relating Mitian to AA. As someone who is professionally used to work with noisy data (specifically from radar), it saddens me that comparativists can't work with noisy (a.k.a. dodgy) data.

Does the AA 1st person pronoun *ˀan-Vk- (Semitic *ˀanāku) have a Mitian counterpart?

[Travis B.]

And a big problem here is that a sister relationship between AA and Mitian or PIE makes little sense considering the sheer time depth of AA. Mitian or PIE would have to be a daughter of AA.

[Brilliant-Resource14]

here is an ongoing PIU reconstruction i'm part of https://docs.google.com/spreadsheets/d/ ... 1620298213

anyone want to rate this?

[Richard W]

The problem is that it’s not 75%: it can easily be more like 95% non-cognate or more. Also, unlike the case you mention, you don’t necessarily know if they are cognate in the first place — indeed that’s precisely the thing you’re trying to prove.

Well, it shouldn't be. You should be asking whether there are cognates (sensu stricto) there. The answer will generally be probabilistic.

I’m not sure I understand your meaning there.

Proving two words are cognate can be difficult. One can be uncertain because of divergences in meaning, and sporadic or highly conditional sound changes can be difficult to confirm when the material is slight. Conversely, there are cases where the semantic and
phonetic correspondences are good, e.g. English and Farsi bad or, without any plausible common ancestor, English and Mbabaram dog. Making a yes or no decision is not a useful approach when any actual matches have been buffeted by the millennia. A pair of distantly related languages do not provide a large population to sample from - discarding actual evidence is a problem. Ideally, one would quantify these matches, so that one can combine weak matches as in statistical analyses. Unfortunately, quantification is difficult. There is also the lurking problem that the sounds and meanings of words are correlated, albeit weakly, and I do not get the feeling that this correlation is a result of common linguistic ancestry.

[zompist]

Proving two words are cognate can be difficult. One can be uncertain because of divergences in meaning, and sporadic or highly conditional sound changes can be difficult to confirm when the material is slight. Conversely, there are cases where the semantic and
phonetic correspondences are good, e.g. English and Farsi bad or, without any plausible common ancestor, English and Mbabaram dog. Making a yes or no decision is not a useful approach when any actual matches have been buffeted by the millennia.

This may just be a case of over-summarization, but you seem to forget that the comparative method is iterative. It doesn't matter that proposed cognates are uncertain; that's only the first step. You look for regular sound changes, and if you find them, you revise the cognate list to reflect what you find.

Eg. maybe you start out thinking that diēs and day (or bad and bad) are related, then you find out that this is not in fact how Latin/English correspondences work, and you remove it from the cognate list. You might start out thinking two and yerku are not related, and then find evidence that they are.

People are very very easy to fool when they're comparing lexicons. But that's only a problem if you stop there, like Greenberg. Precisely because word histories are so messy, you can't establish anything by mass comparison, even if you start talking probabilities instead of certainties.

(Now, maybe it would be useful to attach a probability to each asterisk. That is, a reconstruction like PIE is, say, 50% probable in its details. Then a reconstruction based on other reconstructions, like Nostratic, is 25% probable or less. People forget how fast multiplying probabilities reduces the overall certainty level. Chain together 3 statements of 80% probability and the result is just 51% probable.)

[Travis B.]

It also should be noted that sound correspondences across different groups that seem too close are very, very often loans. Take IE and Uralic, for instance ─ most apparent 'cognates' are, when you look closer, obvious loans, especially since very many of these are clearly not from PIE into PU, but rather are from PII into PU. Real cognates usually reflect systematic sound changes from a common ancestor, not superficial similarity.

[WeepingElf]

Does the AA 1st person pronoun *ˀan-Vk- (Semitic *ˀanāku) have a Mitian counterpart?

It doesn't.

It also should be noted that sound correspondences across different groups that seem too close are very, very often loans. Take IE and Uralic, for instance ─ most apparent 'cognates' are, when you look closer, obvious loans, especially since very many of these are clearly not from PIE into PU, but rather are from PII into PU. Real cognates usually reflect systematic sound changes from a common ancestor, not superficial similarity.

Yep. The sound correspondences that are often proposed are mostly trivial - each PIE phoneme corresponds to the closest PU phoneme. That's not what to expect from two languages that had parted ways thousands of years ago, but from loanwords. The pronouns, however, are another matter. Pronouns can be borrowed, but the borrowing of a whole paradigm of them seems less likely.

[Brilliant-Resource14]

Yep. The sound correspondences that are often proposed are mostly trivial - each PIE phoneme corresponds to the closest PU phoneme. That's not what to expect from two languages that had parted ways thousands of years ago, but from loanwords. The pronouns, however, are another matter. Pronouns can be borrowed, but the borrowing of a whole paradigm of them seems less likely.

I also believe that the noun paradigms (or at least Nominative, Accusative, and Ablative, along with Dual and Plural) are shared. You don't just sprachbund all of your damn grammar

[bradrn]

You don't just sprachbund all of your damn grammar

Actually, yes you can. The most famous case is Kupwar but there are many others.

[Richard W]

(Now, maybe it would be useful to attach a probability to each asterisk. That is, a reconstruction like PIE is, say, 50% probable in its details. Then a reconstruction based on other reconstructions, like Nostratic, is 25% probable or less. People forget how fast multiplying probabilities reduces the overall certainty level. Chain together 3 statements of 80% probability and the result is just 51% probable.)

But if looking for 2 or 3 of the 3 to be so, you get 90%.

I do wonder how dependent on Vedic Sanskrit we are for the discovery of Verner's law. We still have Balto-Slavonic and Greek, so the dependence is not total. It has pushed a troubling inconsistency in phonation into the domain of accent placement, and we are less troubled by vagaries of accent placement.

[WeepingElf]

You don't just sprachbund all of your damn grammar

Actually, yes you can. The most famous case is Kupwar but there are many others.

That is called metatypy. It has been quite some time since I read about the Kupwar case, but what I remember is that the languages of Kupwar have syntactic structures that almost perfectly match morpheme by morpheme, but the sound shapes of the morphemes did not converge - Kupwar Marathi still has Indo-Aryan morphemes, Kupwar Kannada still has Dravidian morphemes. What we have in IE and Uralic is the opposite: the structures are quite different (IE is fusional with ablaut, while Uralic is agglutinating with vowel harmony), but the morphemes have similar sound shapes. This suggests that the languages are related, but one (or both) of them was heavily restructured, perhaps by a substratum with a different structure. And as Uralic looks more "typically Mitian" than IE (e.g., Turkic or Eskimo-Aleut appear to have typologically more in common with Uralic than with IE) and generally has a more archaic look and feel, it seems likely that it was PIE which was restructured.

[bradrn]

You don't just sprachbund all of your damn grammar

Actually, yes you can. The most famous case is Kupwar but there are many others.

That is called metatypy.

Indeed, that is the term.

It has been quite some time since I read about the Kupwar case, but what I remember is that the languages of Kupwar have syntactic structures that almost perfectly match morpheme by morpheme, but the sound shapes of the morphemes did not converge - Kupwar Marathi still has Indo-Aryan morphemes, Kupwar Kannada still has Dravidian morphemes.

Yes, and that is the usual situation with metatypy — for another example see e.g. https://web.archive.org/web/20070703083 ... tatypy.pdf.