News Flash: nostratic-esque theory in news

[2+3 clusivity]

NEWS FLASH! Ha, right. Anyway: http://www.washingtonpost.com/national/ ... story.html.

[Kereb]

terrible, just terrible

but didn't this exact story (ie, the same study, I think) come out in a slightly different version some time last year?

[Vuvuzela]

"To flow"? "To spit"? "Mother"? All of these are easily explained as onompotopoeia/ [spIt_h] and [floU] sound like someone spitting and something flowing, and [ma] sounds like the first noises a baby makes.
Using words like these makes me skeptical of any long-distance comparison.
Also, "Sound alike'? I hope this is being misrepresented here, because I hope these linguists aren't that ignorant of historical linguistics.

[zompist]

They're not linguists at all, and it looks like drek.

Some linguists' reactions over at LanguageHat.

[Salmoneus]

terrible, just terrible

but didn't this exact story (ie, the same study, I think) come out in a slightly different version some time last year?

I think newspaper editors (and their modern equivalents) have a pile of "linguistics" stories that they leave to rot until there's a gap in their 'and finally' section, at which point they just shove something in from the file. We tend to see the same "news" from different outlets over months or years.

[BettyCross]

I considered posting this article here, but something intervened. I felt sure somebody would pick it up.

There is no other primate species that speaks. (The calls that many animals use don't count.) Human language is so completely unique, biologically speaking, that it may indeed have only had one source. But since we have to use written languages to reconstruct the history of language groups, there are strict limits to how far back we can go. This is why I'm skeptical not only of Proto-World but also such hypotheses as Eurasiatic and Norstratic. There may be a link between Indo-European and Uralic, but beyond there's just not enough unequivocal data.

Betty Cross

[KathTheDragon]

I likewise doubt the possibility of a reconstructible Eurasiatic/Nostratic, but if the eivdence is sufficiently good, I may read further.

This, on the other hand, is nonsense.

[Radius Solis]

But since we have to use written languages to reconstruct the history of language groups, there are strict limits to how far back we can go.

That is not true - the comparative method of language reconstruction does not rely on languages having a writing system (although it certainly helps a great deal when they did). Rather, the two main reasons there are limits to how far back we can go are 1. that random lexical replacement really adds up over the course of millennia, leaving too little material still shared between very distantly related languages for us to have much data to work with, and 2. sound changes often obscure the patterns of prior sound changes, and this effect also adds up over millennia, greatly reducing the quality of the data for very distantly related languages. In short, both the quantity and the quality of data decrease with time.

[Drydic]

I considered posting this article here, but something intervened. I felt sure somebody would pick it up.

There is no other primate species that speaks. (The calls that many animals use don't count.) Human language is so completely unique, biologically speaking, that it may indeed have only had one source. But since we have to use written languages to reconstruct the history of language groups, there are strict limits to how far back we can go. This is why I'm skeptical not only of Proto-World but also such hypotheses as Eurasiatic and Norstratic. There may be a link between Indo-European and Uralic, but beyond there's just not enough unequivocal data.

Betty Cross

Blanket declarations are a bad thing. Skepticism is a good thing, but not summary dismissal. And Radius' points are better arguments against the macrofamilies than written languages imposing a brick wall at c.6-8k BC (or whereever your 'strict limit' is set).

also, what's with the arbitrary text size?

[WeepingElf]

This is just yet another variation on a theme that people experienced with this matter know as "Nostratic", "Eurasiatic", "Mitian", "Eusian" and some other names. The exact list of presumed member languages varies a bit between different authors, but it usually includes Indo-European, Uralic and Altaic (the latter, BTW, not really an established family, either). There are some tempting similarities in the morphologies and in the vocabularies of these languages, and it is quite possible that some of them are related to each other (and a time depth of 15,000 years would also be plausible). Alas, nobody has so far presented decisive evidence that would be sufficient to nail the case down, and all claims that a 15,000-year-old megafamily has been found in Eurasia are so far premature.

[Drydic]

No, THIS is utter tripe, linguistically speaking. Putting it in the same boat as Nostratic, which actually has hard data backing it up (even if the list isn't set in stone yet) is just insulting.

[Salmoneus]

But since we have to use written languages to reconstruct the history of language groups, there are strict limits to how far back we can go.

That is not true - the comparative method of language reconstruction does not rely on languages having a writing system (although it certainly helps a great deal when they did). Rather, the two main reasons there are limits to how far back we can go are 1. that random lexical replacement really adds up over the course of millennia, leaving too little material still shared between very distantly related languages for us to have much data to work with, and 2. sound changes often obscure the patterns of prior sound changes, and this effect also adds up over millennia, greatly reducing the quality of the data for very distantly related languages. In short, both the quantity and the quality of data decrease with time.

Also, language reconstruction only really works if you start out knowing which languages are related (or, in practice, if you can have a really good guess at it). In cases like IE, this was pretty obvious almost immediately. In other cases, however, lexical and morphological and syntactic borrowings have obscured the family resemblances. The reconstructions are thus lacking important information from an unrecognised cousin, or, even worse, are led astray by attempts to explain developments in a language which turns out not even to be related to the family.

[NB. other primates may not have language, but ravens certainly do]

[KathTheDragon]

The reconstructions are thus lacking important information from an unrecognised cousin, or, even worse, are led astray by attempts to explain developments in a language which turns out not even to be related to the family.

Or other languages vital to the reconstruction died out before they could leave any evidence of themselves behind.

[WeepingElf]

Here's what has appeared on this matter on Language Log. It says exactly where the problems lie.

[Imralu]

My favourite part ...

Pagel’s team used as its starting material 200 words that linguists know to be the core vocabulary of all languages.

"Core vocabulary" is a link to the Swadesh list.

[Terra]

Relevant: http://www.smbc-comics.com/index.php?db ... 2556#comic

They probably got the idea from genetics, where a similar thing is done with genes.

Also, "Sound alike'? I hope this is being misrepresented here, because I hope these linguists aren't that ignorant of historical linguistics.

According to the paper, they're biologists and psychologists.

[Yiuel Raumbesrairc]

According to the paper, they're biologists and psychologists.

Oh great. Someone should remind them to stay in their fields. Languages are obviously too complicated for them.

[Vuvuzela]

So you're trying to predict the relationships between languages at time depths greater than 6,000-9,000 years? Just model it as phylogenetics*, and then add some secondary terms to account for borrowing. Easy, right? So, why does linguistics need a whole journal, anyway?
*Not exactly a simple object, but whatever.

[Soap]

Incidentally I've just discovered that wolves evolved into horses and not just dogs as previously thought. The proof of this is that the PIE word for wolf (*wḷkʷo) looks a lot like the word for horse (*ekʷo) and in fact it even looks more like the word for horse than it does the word for dog! Clearly horses were created by the PIE people as they are the source of our modern words for horses. Therefore we are getting a glimpse of the very first horses ever to walk the earth, thanks to the all-seeing eye of historical linguistics. I'm not a biologist but I expect to have this earth-shattering discovery published in all the major biology magazines this next month as it is clearly the greatest discovery of our lifetimes.

[WeepingElf]

Incidentally I've just discovered that wolves evolved into horses and not just dogs as previously thought. The proof of this is that the PIE word for wolf (*wḷkʷo) looks a lot like the word for horse (*ekʷo) and in fact it even looks more like the word for horse than it does the word for dog! Clearly horses were created by the PIE people as they are the source of our modern words for horses. Therefore we are getting a glimpse of the very first horses ever to walk the earth, thanks to the all-seeing eye of historical linguistics. I'm not a biologist but I expect to have this earth-shattering discovery published in all the major biology magazines this next month as it is clearly the greatest discovery of our lifetimes.

That's a good one.

[Vuvuzela]

This new discovery really challenges the status kʷo, huh?

[2+3 clusivity]

challenges the status kʷo, huh?

I saw what you did there.

[Morrígan]

Something that didn't seem to be addressed in the Language Log post is the apparent lack of any resembling an experimental control - this suggests that their technique could produce results for an arbitrary set of languages regardless of whether they are related or not, or even real or not.

And they way the assembled their cognate lists seems dangerously like cherry-picking or question-begging.

Or did I miss something.

[Richard W]

Or did I miss something.

As the publicity bears little relationship to the content of the paper, yes.

The one statistical result in the paper is that the meanings for which the LWED does best at coming up with cognates are the meanings where cognates are most likely to be preserved. Of course, this could just demonstrate that the likeliest places for cognates have been researched the hardest!