The Great Macrofamily thread: Indo-Uralic, Altaic, Eurasiatic, Nostratic etc.

[Nortaneous]

This begs the question why only back sibilants have a labialization distinction!

Why shouldn't this be the case? Gaps in labialized series are well-attested. AFAICT, /sʷ zʷ/ do not exist in Northwest Caucasian at all.

[KathTheDragon]

yet, the Proto-Uralic and Proto-Eskimo-Aleut consonant inventories still look, at first glance, quite similar to each other (though the vowel inventories could hardly be more different - which probably shows the well-known fact that vowels tend to be more volatile than consonants) - but does that mean that both have preserved the Proto-Mitian inventory especially well?

I'm skeptical of what can be deduced through comparing phonologies.

[mae]

-

[cedh]

There actually seems to be some sort of correlation between post-alveolars (or "back sibilants") and rounding in general; all postalveolar sounds in English are slightly rounded for instance. This makes phonetic sense, because a lowered F3 seems to occur with some post-alveolar sounds (mostly apicals?), and a low F3 is also one of the main auditory cues for labialization. So only having labialized "back sibilants" and no labialized "front sibilants" doesn't really warrant any skepticism.

As for rounded postalveolar sibilants, some western varieties of German spoken in the region around Cologne distinguish /s/ :: /ʃ/ :: /ʃʷ/ (mostly corresponding to Standard German /s/ :: /ç/ :: /ʃ/).

[WeepingElf]

yet, the Proto-Uralic and Proto-Eskimo-Aleut consonant inventories still look, at first glance, quite similar to each other (though the vowel inventories could hardly be more different - which probably shows the well-known fact that vowels tend to be more volatile than consonants) - but does that mean that both have preserved the Proto-Mitian inventory especially well?

I'm skeptical of what can be deduced through comparing phonologies.

I am, too. There was a time when I was into what I called "system comparison" - comparing phonological and morphological system without the drudgery of working out regular lexical correspondences. How badly this can go wrong may be illustrated with Greek vs. Armenian, where one would arrive at such "correspondences" as Gk. ph = Arm. ph, Gk. p = Arm. p, Gk. b = Arm. b etc. - which are of course wrong. And the Uralic and Eskimo-Aleut consonant inventories aren't really that similar, as I pointed out yesterday.

The whole eastern wing of Mitian, now that I take a closer look at it, is very brittle. I am now much more skeptical about "Seefloth's Paradigm" than I used to be. What does really match here, and what does that mean? The order of possessum number, possessor person, possessor number is trivial - one would expect just this from a suffixing agglutinating language. And what regards the morphemes themselves: sure, the plural marker is *-t in both families, but the dual markers and the personal morphemes do not match well; where is the *mi and *ti in Eskimo-Aleut? Is this a Mitian language family at all?

Next, Yukaghir and Chukotko-Kamchatkan. Their morphologies look nothing alike AND nothing like Uralic, IE or anything, not even Eskimo-Aleut, apart from those peaky /m/s and /t/s in the pronouns. The Uralic-Yukaghir hypothesis has been pretty much disproved by Ante Aikio, and the Uralic-looking words are probably just loanwords from a (possibly unknown) Samoyedic or Para-Samoyedic language. As for Chukotko-Kamchatkan, Fortescue no longer considers it Uralo-Siberian.

So what we are left with are IE, Uralic and the three (Micro-)Altaic families, and while both Indo-Uralic and Micro-Altaic do not look all that bad (but still a long way to go), they are only very vaguely connected by places of articulation of the key consonants in the pronouns, which is not much.

The more I look at these things, the less optimistic I am about it! There are tantalizing resemblances between these languages, but they are too vague to get a handle on anything. And when a hypothesis makes less sense the more research one invests in it, it is probably wrong.

[Pabappa]

I may continue to believe in Mitian even if you abandon it. I know Eskimo-Aleut has /m/ and /t/ for 1st and 2nd person, which is the core of the definition. There is very little lexical overlap, but with a polysynthetic language, that may be expected .... the various branches of Eskimo-Aleut don't even have that much vocab in common with each other.

[WeepingElf]

I am not sure yet about abandoning Mitian; it is at least plausible, though trying to demonstrate it will be hard, and way beyond my capabilities as an amateur in this field. When eight language families occupying a contiguous area share a set of pronouns like this, the hypothesis that they all inherited them from a common ancestor seems to be better than either borrowing (sure, Wanderwörter are a thing, but how likely is it that pronouns travel from language to language like this?) or chance resemblance.

And at any rate, such hypotheses are surely legitimate to explore in conlangs, and that's why I do this (though I am indeed curious about these matters!), and others here perhaps too.

Let me add, as you speak of what you "believe" in, that I don't believe in anything (apart from the decisively proven language families) when it comes to macrofamilies. It is just that I feel that some hypotheses are more plausible than others, but my feelings about this are sometimes subject to change. As to where and when Proto-Mitian may have been spoken, I think the most plausible assumption is that was spoken somewhere deep inside Asia, perhaps near Lake Baykal, at the end of the last ice age, when the climate zones shifted northwards by some 1,000 km within a few centuries, and people migrated at a grand scale as they followed their venison northwards. These grand migrations would have involved the spread of some very large language families in northern Eurasia, and Mitian may have been one of them.

[Howl]

Uralic looks somewhat conservative in having bisyllabic roots and a decent amount of root-internal consonant clusters, but the first can be found across Altaic groups and the second in PIE just as well. By similar logic e.g. initial clusters in PIE could also be considered at least partly archaic. The largeish vowel system in PU is almost surely not archaic; vowels have a half-life of like three centuries during the later development of most Uralic branches. (One of these days I should clean up and put online my conference presentation from last summer on the internal reconstruction of the PU vowel system.)

Fair. What certainly isn't archaic about the vowel system of Proto-Uralic is the reduction in non-first syllables, but the size of the 1st-syllable inventory reeks of some kind of innovation, too. But we don't know how many vowels pre-GVC PIE had - it seems to have at least */i e a o u/ and no front rounded vowels (otherwise, we'd have labialized front velars in PIE, which we don't), but it may have had more than three degrees of openness, two rather than one low vowel (though the paucity of unlabialized back velars in PIE seems to indicate that there was just one non-front unrounded vowel), etc. I would not be so bold to reconstruct vowel harmony for PIU or Proto-Mitian, though!

It would be trivial to turn the reconstructed Uralic 8x2 vowel system into a 4x4 vowel system by assuming umlaut from the second syllable. But more and more I want to have some evidence before reconstructing anything. And it seems that there are more than 8x2=16 vowel correspondence sets in Uralic already (e.g. the disharmonic stems, although that seems to be a mostly Finnic thing). I do think it is likely that Uralic had less vowels in the first syllable in an earlier stage. And I have always doubted that the funnel of the GVC was so small that it left only one vowel in PIE.

I don't know where the story comes from that PIE has few plain velars. But when I eyeball the data in LIV2, I get a different impression.

Afrasian is a much deeper thing than IE or Uralic, probably about as deep as Mitian (if the latter is a thing at all, of course). There are at least two mutually incompatible reconstructions of Proto-Afrasian (one by Ehret, one by Orel and Stolbova, both published in 1995) on the market, and nobody knows which one - if any - is right. As I have remarked earlier, that Afrasian is generally accepted and Indo-Uralic or Mitian is not, has to do with the fact that in African linguistics, the lumpers dominate the discourse, while in Eurasian linguistics, it is rather the splitters. That said, shared morphology makes it quite clear that Afrasian is a thing, and so is Niger-Congo (while Nilo-Saharan and Khoisan look much worse), but Indo-Uralic looks almost as good to me.

Almost any Nostraticist assumes that Eurasiatic is a valid node, but no one seems to take the effort to reconstruct it. The only thing we have are the lookalike sets in Greenberg's book. Most of Eurasiatic is based on lexical matches. But no one is going to take that as proof of anything unless there are at least some rough sound correspondences.

Typological similarity to "Ural-Altaic" means nothing.

I totally agree. I see typology as mostly an areal thing. Typology is probably the worst indicator to determine if languages are genetically related or not. Comparison of the lexicon is already better. And matching morphology is the best indicator there is.

Yes, Uralic is harder to reconstruct than IE because there aren't those nifty ancient literary languages which bridge half of the time in the case of IE. Basically, IE can be reconstructed as if it was just 2,500 years old, because we know how Indians, Persians and Greeks spoke 2,500 years ago.

I certainly agree with that. But what is also an issue is that there are much less people working on the reconstruction of Uralic than on Proto-Indo-European. PIE even has enough people to support different schools of thought. And this is a great thing, since it gives us different hypotheses to work with and to compare. When it comes to Uralic, any alternative school of thought like Tálos's 2-tier vowel system or Steinitz's vowel alternations just seems to peter out.

Also, the Neogrammarian model of regular, exceptionless sound changes is just a model which doesn't capture all detail; anyone who has ever perused a dialect atlas of whichever language knows that phonological isoglosses tend to vary (though usually not by much) from word to word. There is no reason to assume that prehistoric languages were any different in this regard.

However, the hypothesis of regular sound change is the only thing that allows us to look so far back in the past of a language family. So that is why it is always good to avoid any theory that depends on random sound changes unless there is no other reasonable explanation.

[Tropylium]

What IMHO also has a good chance of being archaic is Eskimo-Aleut (though the vowel system may have once been richer, as if EA had its own GVC).

Possibly, but is it archaic enough? Proto-Eskimo is only a few thousand years old. PEA could be older than has been usually thought (some recent work suggests that a proportion of the supposedly shared vocab are instead old loans into western Aleut), but still probably only maybe 6k years. There would be time for extensive restructuring during its way out of Central Asia.

Afrasian is a much deeper thing than IE or Uralic, probably about as deep as Mitian (if the latter is a thing at all, of course). There are at least two mutually incompatible reconstructions of Proto-Afrasian (one by Ehret, one by Orel and Stolbova, both published in 1995) on the market, and nobody knows which one - if any - is right.

There's a third, smaller one by now by Bomhard (most of it just from his Nostratic dictionary) which attempts building some compromise.

"Erythraean" (Berber-Egyptian-Semitic) looks real to me. Cushitic and also the big East Cushitic subgroup by contrast look to me fairly strongly defined by archaisms and not innovations. I've also taken a look at Agaw (Central Cushitic) a while ago, and it also looks divergent enough that in my nonspecialist impression, the clearest similarities with nearby East Cushitic could be areal influence and not common Proto-Cushitic inheritance. (Hetzron suggested an Agaw + Highland East Cushitic group which has not been accepted.) In this case we could be dealing with a third independent Afrasian group within "Cushitic".

But what regards the question of a relationship between Mitian and Afrasian, I don't see why these two entities should be especially close. There is almost nothing that seems to match, and what does seem to my just be chance resemblances.

Semitic and PIE have several seemingly good matches, but perhaps they're an old layer or two of loanwords (e.g. #ʔartɬʼ- 'earth' seems like a word that's likely to be associated with the adoption of agriculture). I've got a project stewing to dig up the most promising cases from Bomhard's data and seeing just what language groups they really connect.

Yukaghir and Chukotko-Kamchatkan. Their morphologies look nothing alike AND nothing like Uralic, IE or anything, not even Eskimo-Aleut, apart from those peaky /m/s and /t/s in the pronouns. The Uralic-Yukaghir hypothesis has been pretty much disproved by Ante Aikio, and the Uralic-looking words are probably just loanwords from a (possibly unknown) Samoyedic or Para-Samoyedic language.

So what we are left with are IE, Uralic and the three (Micro-)Altaic families (…)

I agree that cutting out CK seems like a good move for now, and that Uralic-Yukaghir is too poorly argued for, but I would only demote Yukaghir as far as an independent Mitian branch. The really interesting similarities are not that numerous, but they're quite often shared with not just Uralic but also IE.

[WeepingElf]

It would be trivial to turn the reconstructed Uralic 8x2 vowel system into a 4x4 vowel system by assuming umlaut from the second syllable. But more and more I want to have some evidence before reconstructing anything. And it seems that there are more than 8x2=16 vowel correspondence sets in Uralic already (e.g. the disharmonic stems, although that seems to be a mostly Finnic thing). I do think it is likely that Uralic had less vowels in the first syllable in an earlier stage. And I have always doubted that the funnel of the GVC was so small that it left only one vowel in PIE.

It may be true that something like what you suggest has indeed happened in Proto-Uralic, but there probably no way knowing without comparison with related languages, which have not been found yet. What regards the GVC, I assume that at least *i and *u, where they do not immediately precede a sonorant, have survived unscathed. Also, I am no longer sure that the GVC collapsed all the affected vowels into one vowel, though this seems to be the most parsimonious assumption. However, there are many unexplained irregularities in PIE ablaut which may make the assumption of more than one vowel besides *i and *u necessary. My knowledge of these matters is way too shallow to answer these questions.

Almost any Nostraticist assumes that Eurasiatic is a valid node, but no one seems to take the effort to reconstruct it. The only thing we have are the lookalike sets in Greenberg's book. Most of Eurasiatic is based on lexical matches. But no one is going to take that as proof of anything unless there are at least some rough sound correspondences.

The lexical lookalikes in Greenberg's book aren't worth much. This is his infamous technique of "mass lexical comparison" which reveals nothing, as long as there are no regular sound correspondences. At best, this technique can throw up ideas where to look for correspondences; at worst, it can be used to "prove" that anything is related to anything else.

Typological similarity to "Ural-Altaic" means nothing.

I totally agree. I see typology as mostly an areal thing. Typology is probably the worst indicator to determine if languages are genetically related or not. Comparison of the lexicon is already better. And matching morphology is the best indicator there is.

I concur with you on this. Typology mean nothing; lexical correspondences (as long as this involves regular sound correspondences and not just casual similarity) means a lot, but needs to be supported by matching morphology (which when used alone has the drawback of providing only small amounts of data). If a hundred words match but the morphology doesn't, you are probably dealing with a batch of loanwords. Case in point: Armenian, which was thought to be an Iranian languages due to its many lexical matches with Persian; the mismatching morphology revealed that these are just loanwords, and Armenian is not an Iranian (though an Indo-European) language. Yukaghir probably is a similar case.

What you need is a set of regular correspondences that involve a reasonable number of lexical items and a good part of the morphology, preferably entire paradigms. The former without the latter is vulnerable to loanwords; the latter without the former does not provide enough data to establish sound correspondences.

Yes, Uralic is harder to reconstruct than IE because there aren't those nifty ancient literary languages which bridge half of the time in the case of IE. Basically, IE can be reconstructed as if it was just 2,500 years old, because we know how Indians, Persians and Greeks spoke 2,500 years ago.

I certainly agree with that. But what is also an issue is that there are much less people working on the reconstruction of Uralic than on Proto-Indo-European. PIE even has enough people to support different schools of thought. And this is a great thing, since it gives us different hypotheses to work with and to compare. When it comes to Uralic, any alternative school of thought like Tálos's 2-tier vowel system or Steinitz's vowel alternations just seems to peter out.

Yep. There are many more Indo-Europeanists than Uralicists. What do you expect when the former family has about 100 times as many speakers than the latter? Linguists tend to be interested in their native language and its relatives, though there are of course so many exceptions to this "rule" that it is only a weak tendency. Yet, the body of scholarship is much larger in IE than in Uralic.

Also, the Neogrammarian model of regular, exceptionless sound changes is just a model which doesn't capture all detail; anyone who has ever perused a dialect atlas of whichever language knows that phonological isoglosses tend to vary (though usually not by much) from word to word. There is no reason to assume that prehistoric languages were any different in this regard.

However, the hypothesis of regular sound change is the only thing that allows us to look so far back in the past of a language family. So that is why it is always good to avoid any theory that depends on random sound changes unless there is no other reasonable explanation.

Right. As much as this is "just a model", we have no better one, and test cases such as Romance show that the method works excellently (though not perfectly). And all attempts at developing methods that could look deeper such as mass lexical comparison, glottochronology (which was originally meant for dating already established relationships, and failed at even that; yet, it was attempted to abuse it for finding new relationships), "system comparison", and whatever, have failed.

What IMHO also has a good chance of being archaic is Eskimo-Aleut (though the vowel system may have once been richer, as if EA had its own GVC).

Possibly, but is it archaic enough? Proto-Eskimo is only a few thousand years old. PEA could be older than has been usually thought (some recent work suggests that a proportion of the supposedly shared vocab are instead old loans into western Aleut), but still probably only maybe 6k years. There would be time for extensive restructuring during its way out of Central Asia.

Of course, Eskimo-Aleut had plenty of time to evolve away from Proto-Mitian. 8,000 years are more than enough to change a language almost beyond recognition, even if it spread across virgin territory and had no substratum.

Afrasian is a much deeper thing than IE or Uralic, probably about as deep as Mitian (if the latter is a thing at all, of course). There are at least two mutually incompatible reconstructions of Proto-Afrasian (one by Ehret, one by Orel and Stolbova, both published in 1995) on the market, and nobody knows which one - if any - is right.

There's a third, smaller one by now by Bomhard (most of it just from his Nostratic dictionary) which attempts building some compromise.

OK.

"Erythraean" (Berber-Egyptian-Semitic) looks real to me. Cushitic and also the big East Cushitic subgroup by contrast look to me fairly strongly defined by archaisms and not innovations. I've also taken a look at Agaw (Central Cushitic) a while ago, and it also looks divergent enough that in my nonspecialist impression, the clearest similarities with nearby East Cushitic could be areal influence and not common Proto-Cushitic inheritance. (Hetzron suggested an Agaw + Highland East Cushitic group which has not been accepted.) In this case we could be dealing with a third independent Afrasian group within "Cushitic".

I am no expert on anything like this, but the Berber-Egyptian-Semitic node seems reasonable to me, though it may in fact be a convergence area within Afrasian. Proto-Cushitic, Proto-Chadic and Proto-Omotic have, to my knowledge, not been convincingly reconstructed yet.

But what regards the question of a relationship between Mitian and Afrasian, I don't see why these two entities should be especially close. There is almost nothing that seems to match, and what does seem to my just be chance resemblances.

Semitic and PIE have several seemingly good matches, but perhaps they're an old layer or two of loanwords (e.g. #ʔartɬʼ- 'earth' seems like a word that's likely to be associated with the adoption of agriculture). I've got a project stewing to dig up the most promising cases from Bomhard's data and seeing just what language groups they really connect.[/quote[

This indeed needs more study. The Semitic-IE matches may just be Neolithic Wanderwörter. Also, it is a long way from the Upper Nile where Proto-Afrasian probably was spoken (pace Bomhard and most of the rest of the Nostraticists) to Central Asia/Southern Siberian where the Proto-Mitian homeland, if that's a thing, is most likely to be sought. It would be equally plausible to connect Mitian to Sino-Tibetan, Na-Dené, or whatever, on geographical grounds.

Yukaghir and Chukotko-Kamchatkan. Their morphologies look nothing alike AND nothing like Uralic, IE or anything, not even Eskimo-Aleut, apart from those peaky /m/s and /t/s in the pronouns. The Uralic-Yukaghir hypothesis has been pretty much disproved by Ante Aikio, and the Uralic-looking words are probably just loanwords from a (possibly unknown) Samoyedic or Para-Samoyedic language.

So what we are left with are IE, Uralic and the three (Micro-)Altaic families (…)

I agree that cutting out CK seems like a good move for now, and that Uralic-Yukaghir is too poorly argued for, but I would only demote Yukaghir as far as an independent Mitian branch. The really interesting similarities are not that numerous, but they're quite often shared with not just Uralic but also IE.

I would not dump these languages from Mitian yet; Yukaghir clearly has Mitian pronouns, and CK also seems to have personal morphemes of this kind. I am not sure how good Fortescue's 2011 paper is; I have read it, and it did not really convince me - and who says that Nivkh is not a Mitian language? Sure, its 1st and 2nd person pronouns are ńi and chi (my transcription), but these could easily have arisen from *mi and *ti by palatalization, though they could just as well have arisen the same way, for instance, from *ni and *ki, or whatever.

[Tropylium]

Also, I am no longer sure that the GVC collapsed all the affected vowels into one vowel, though this seems to be the most parsimonious assumption. However, there are many unexplained irregularities in PIE ablaut which may make the assumption of more than one vowel besides *i and *u necessary. My knowledge of these matters is way too shallow to answer these questions.

An "immediate" GVS with all pre-PIE non-close vowels merging into one sounds unlikely: e.g. even in Indo-Iranian, the *a/*o > *a merger probably happened first, the *e/*a > *a merger probably only later (after Law of Palatals at minimum). In pre-PIE, there were probably a few different rounds of vowel mergers. I would think the apparent uniformity in verb roots has to be in part analogical, including not just the (only partly complete) attraction of *i *u into *ei *eu ~ *oi *ou ~ *i *u ablaut, but also some parts of the basic *e/o ablaut.

I concur with you on this. Typology mean nothing;

My view is a bit more nuanced. Typological similarity means rather little, but typological dissimilarity can be a strong argument against any relatively close relatedness. Depending on the feature: e.g. SVO versus SOV means approximately nothing, but meanwhile the presence of noun gender is one of the more stable markers of Indo-European languages, and its absense is also 100% stable in the non-IE "Eurasiatic" languages.

So point being that, aside from geographic distantness, there's no a priori reason to entirely avoid comparing Dravidian to (parts of) Eurasiatic … while we can be sure that comparing anything Eurasiatic with e.g. any of the Southeast Asian families is not going to be a good idea. Whatever the real language tree of Eurasia looks like, it is not going to have surprize nodes along the lines of IE–Austronesian in it. It's also however possible e.g. that Dravidian is just as distant from Eurasiatic as Eurasiatic is from any of the "Gondwanan" languages of SEA, Australia etc.

— Or indeed, just as distant even from any of the sub-Saharan African families, now that it's starting to look like Homo sapiens sapiens actually rather evolved in western-to-southern Asia all along, and only backmigrated to Africa later on. This would make it more sensible how Africa seems to be covered by a few huge families while (Eur)asia still resists macro-grouping. If there was no African exodus, there is no reason to assume a recent Proto-Exo-African, while there could inversely have been a major "Proto-African".

(I kinda expect archeology to eventually show that archaic humans could have persisted in Africa even up to pre-Bantu times — e.g. Pygmy groups are highly genetically distinct, and this has usually been taken to indicate they adopted Bantu with a clean language shift, but who's to say they didn't use to be even more distinct before the Bantu expansion?)

This also makes me doubt the idea that Afrasian would have spread from a southern homeland. Sure the current center of diversity is along the upper Nile, but Africa is huge, and if it did used to be populated by archaic human (sub)species, then we expect to see a typical refuge zone pattern: numerous invasions, each of them enabled by a new layer of technology, coming in from the better-connected Eurasia (indeed, you could view the Scramble for Africa as the latest of these invasions). Afrasian is surely pre-agriculture, and after agriculture, the spread of Egyptian and Semitic would be expected to have wiped off any straggling "old Afrasian" groups from the Near East / lower Nile / Mediterranean coast.

[mèþru]

It's definite that all humans alive today are mainly descended from modern humans. No peoples, including the Mbenga, Twa and Mbuti, are "archaic" humans.

I have been looking at the literature after the major news on early humans, and no serious paper claims that this points to a non-African genesis of humans. Rather, it is now thought that humans had spread into the Middle East a long time before it was previously assumed. These migrations did not leave modern-day descendants unless if they migrated back; conditions eventually got too hot and dry for humans to live. The new assumptions on human evolution instead say that perhaps H. sapiens evolution happened as intermixing of various populations with various new mutations throughout Africa; a bit like how several different sound changes can start in different places and then spread to cover the whole continuum.

[WeepingElf]

Also, I am no longer sure that the GVC collapsed all the affected vowels into one vowel, though this seems to be the most parsimonious assumption. However, there are many unexplained irregularities in PIE ablaut which may make the assumption of more than one vowel besides *i and *u necessary. My knowledge of these matters is way too shallow to answer these questions.

An "immediate" GVS with all pre-PIE non-close vowels merging into one sounds unlikely: e.g. even in Indo-Iranian, the *a/*o > *a merger probably happened first, the *e/*a > *a merger probably only later (after Law of Palatals at minimum). In pre-PIE, there were probably a few different rounds of vowel mergers. I would think the apparent uniformity in verb roots has to be in part analogical, including not just the (only partly complete) attraction of *i *u into *ei *eu ~ *oi *ou ~ *i *u ablaut, but also some parts of the basic *e/o ablaut.

You are confusing a proposed pre-PIE sound change with an admittedly similar change within the family, it seems to me. With the GVC, I mean a change that transformed the probably richer Proto-Indo-Uralic vowel inventory into the */a i u/ system I propose(d) for the pre-ablaut stage of (Pre-)PIE. What happened in Indo-Iranian is a similar but much later and independent change. But I have to admit that the GVC is only a simplified model and what actually happened may have been much more complex, especially if the pre-ablaut stage had more than one non-close vowel.

I concur with you on this. Typology mean nothing;

My view is a bit more nuanced. Typological similarity means rather little, but typological dissimilarity can be a strong argument against any relatively close relatedness. Depending on the feature: e.g. SVO versus SOV means approximately nothing, but meanwhile the presence of noun gender is one of the more stable markers of Indo-European languages, and its absense is also 100% stable in the non-IE "Eurasiatic" languages.

What makes gender a stable marker in IE is not simply its presence but the fact that the gender markers are cognate. The Dravidian morphemes appear to be not cognate to the Uralic and "Altaic" ones. On the other hand, cases like Insular Celtic (with its radically transformed syntax and such oddities as initial mutations) or Armenian (which has lost gender and, in the modern language, shows an agglutinating nominal morphology) show that the typological profile of a language can change rapidly within a few centuries. Note that despite their typological aberrancy, in both groups the morphemes from which those "exotic" grammars are built are cognate to those of other IE languages. The Insular Celtic initial mutations, for instance, are traces of lost endings of the preceding words which turn out to be precisely cognate to the preserved ones in other IE languages such as Latin, as Bopp has shown.

Surely, IE is an oddball among the Mitian languages, with its fusional declensions, grammatical gender, tripartite velar series, ablaut, and other things. In my hypothesis, Early PIE was spoken by the Sredny Stog culture of eastern Ukraine, which formed from the merger of the intrusive Khvalynsk culture from the middle Volga and the autochthonic Dniepr-Donets culture. Khvalynsk was a neighbour of Proto-Uralic and may have spoken a related and structurally similar language, which was transformed under the substratum influence from the Dniepr-Donets language, which may have been related or typologically similar to NW Caucasian. This substratum theory is not my own invention; people like Uhlenbeck have proposed it decades ago.

So point being that, aside from geographic distantness, there's no a priori reason to entirely avoid comparing Dravidian to (parts of) Eurasiatic … while we can be sure that comparing anything Eurasiatic with e.g. any of the Southeast Asian families is not going to be a good idea. Whatever the real language tree of Eurasia looks like, it is not going to have surprize nodes along the lines of IE–Austronesian in it. It's also however possible e.g. that Dravidian is just as distant from Eurasiatic as Eurasiatic is from any of the "Gondwanan" languages of SEA, Australia etc.

Of course, one cannot rule out a priori a relationship between Dravidian and Mitian, Eurasiatic, Ural-Altaic or whatever. Yet, given the utterly different shapes of the Dravidian morphemes, the choice of Dravidian seems utterly random, not much better than, say, Dene-Yeniseian.

— Or indeed, just as distant even from any of the sub-Saharan African families, now that it's starting to look like Homo sapiens sapiens actually rather evolved in western-to-southern Asia all along, and only backmigrated to Africa later on.

Evidence? This is the first time I have heard of such a hypothesis!

This would make it more sensible how Africa seems to be covered by a few huge families while (Eur)asia still resists macro-grouping. If there was no African exodus, there is no reason to assume a recent Proto-Exo-African, while there could inversely have been a major "Proto-African".

Don't forget that "Nilo-Saharan" and "Khoisan" aren't families - they are wastebasket categories, into which those languages go which show no signs of being related either to Semitic or to Bantu, dependent on whether they have clicks or not. And even Afrasian and Niger-Congo are in fact macrofamilies on a par with Mitian. Thus, we have about 20 to 30 independent families in Africa - the only difference between Africa and Eurasia is that among Africanists, the lumpers dominate the discourse, and among Eurasianists, the splitters. Probably due to the fact that most linguists are Europeans (or North Americans etc.), and thus see Africa as a remote exotic continent where things look more similar from the distance than in their Eurasian homeland. No doubt an intellectual burden inherited from the colonial era.

(I kinda expect archeology to eventually show that archaic humans could have persisted in Africa even up to pre-Bantu times — e.g. Pygmy groups are highly genetically distinct, and this has usually been taken to indicate they adopted Bantu with a clean language shift, but who's to say they didn't use to be even more distinct before the Bantu expansion?)

Pygmies are not nearly as divergent as Neanderthals or Denisovans! They aren't a residue of an "archaic (sub)species" at all. In fact, their genetic peculiarities are part of the evidence that Homo sapiens originated in Africa. You really should read up on human genetics before making such statements. I'm sorry, as much as I respect you and your opinions, what you say about human genetics here is utterly wrong-headed.

This also makes me doubt the idea that Afrasian would have spread from a southern homeland. Sure the current center of diversity is along the upper Nile, but Africa is huge, and if it did used to be populated by archaic human (sub)species, then we expect to see a typical refuge zone pattern: numerous invasions, each of them enabled by a new layer of technology, coming in from the better-connected Eurasia (indeed, you could view the Scramble for Africa as the latest of these invasions). Afrasian is surely pre-agriculture, and after agriculture, the spread of Egyptian and Semitic would be expected to have wiped off any straggling "old Afrasian" groups from the Near East / lower Nile / Mediterranean coast.

Sure, we don't know where Afrasian originated, but archaic species of Homo certainly weren't involved here in any way - they were long gone when Proto-Afrasian was spoken, which probably was spoken around 10,000 BC, give or take a few thousand years. It is true that the centre of diversity may have been somewhere else than where it is now, yet we don't know, and the most parsimonious assumption is that it was not far from where it is now. (I don't know what genetics say about this case, though.)

[Pabappa]

My interpretation was that he was just using Indo-Iranian as an example, not actually saying the two shifts were related. I.e. if it happened stepwise in the example we know, maybe it happened stepwise in the example we don't.

[mèþru]

Personally, I prefer to think of languages as families only when a reasonable proto-language for them has been reconstructed.

[Tropylium]

What makes gender a stable marker in IE is not simply its presence but the fact that the gender markers are cognate.

I don't think you're following. The fact is that there is, across west-to-south-central Eurasia, a large belt of languages having some kind of a 2-or-3 noun gender system; a typological fact that can be observed without any further comparison. It also so happens that just about these languages belong in a single family (i.e. Indo-European), and inherit the gender system already from their common ancestor. (Also, suppose you were given just Norwegian, French and Hindi, and you'd have barely any chance to identify exact commonalities between the gender markers.)

Cases like Berber and Semitic nearby show that just this one typological marker still gives many false positives, and therefore isn't very good evidence for delimiting the family. But what is significant is that this gives very few false negatives; Englishes and Armenians are one in a hundred. Typological profiles can in some exceptional cases change rapidly, but the overwhelming majority of the time, they don't. For a few stable markers like these, not even across several millennia.

This is not an isolated example: the same also applies e.g. with Turkic and vowel harmony (not all vowel harmony languages are Turkic, but just about all Turkic languages have vowel harmony), or Sinitic and tone (not all tonal languages are Sinitic, but all modern Sinitic varieties are tonal), or ejectives and each of the Caucasian families (not all languages with ejectives are in a Caucasian family, but both all Kartvelian and all North Caucasian languages have ejectives).

Altogether: Sprachbund boundaries are very likely to be also family boundaries. Therefore e.g. none of the "Dene-Caucasian" families (prefixing, verb-heavy) are likely to be discernible relatives of any of the "Eurasiatic" families (suffixing, noun/verb morphology more balanced); and vice versa.

[Tropylium]

— Or indeed, just as distant even from any of the sub-Saharan African families, now that it's starting to look like Homo sapiens sapiens actually rather evolved in western-to-southern Asia all along, and only backmigrated to Africa later on.

Evidence? This is the first time I have heard of such a hypothesis!

As of earlier this year, we have evidence that modern humans were already about in multiple locations in the Middle East as many as 180 000 years ago. This shows that earlier discoveries with maybe a third of that age should have been treated with the Signor-Lipps effect in mind all along, not interpreted as evidence for some kind of a "catastrophic" spill-out from Africa.

Blogger Dienekes develops the detailed argument better than I can do here on my own:
The slow death of Out of Africa

The significance of the discovery of modern humans in Arabia >85kya is that it provides a second spot (other than Israel) were modern humans existed outside Africa long before the alleged 60kya blitz out of the continent. We now have modern humans outside Africa in roughly two locations (Israel and Arabia), and three time slices (~175-85kya) in Misliya, Shkul/Qafzeh, and Al Wusta-1. It is no longer tenable to claim that these modern humans "died out" to make way for the alleged 60kya OoA event.

Statistical Palaeoafricans

I have long maintained that the higher genetic diversity of extant Sub-Saharan Africans is the result of admixture between "Afrasians" (a population that spawned Eurasians and much of the ancestry of Sub-Saharans and which had "low" (Eurasian-level) of genetic diversity) [N.B. no especial relation to the family –ed] and multiple layers of "Palaeoafricans". It would seem that one such layer has now been discovered.

Out of Africa: a theory in crisis

A very short time ago, Ethiopia boasted the oldest AMH by 0.07My and now it's tied with the Levant and beaten by Morocco. It's a bit silly to argue for temporal priority based on the spotty and ever-shifting palaeoanthropological record.

Deepest Neandertal mtDNA split

Within the Modern Humans, Eurasians are a branch of a tree which is mostly African. This has been interpreted for decades as evidence for the Out of Africa hypothesis for the origin of Modern Humans. But, within the phylogeny as a whole, Modern Humans are a branch of the Eurasian tree. This has not (why?) in general been interpreted as evidence for Out of Eurasia for the common ancestor of Modern Humans and Neandertals.

---

Don't forget that "Nilo-Saharan" and "Khoisan" aren't families (…) and even Afrasian and Niger-Congo are in fact macrofamilies on a par with Mitian. Thus, we have about 20 to 30 independent families in Africa - the only difference between Africa and Eurasia is that among Africanists, the lumpers dominate the discourse, and among Eurasianists, the splitters.

So the current orthodox line goes, but I'm not so convinced anymore. African macrofamilies remain still poorly studied, and we cannot declare it known in advance that they would turn just as weak as the Eurasian ones. If anything, the fact that they're about equally contentious now, while still being vastly less studied, really kinda suggests that they're going to be on a lot firmer ground eventually. And maybe we're therefore wrong about the general time depth of language relationships on each continent. If Niger-Congo and Afrasian are both identifiable and somewhat reconstructible despite being 10 000 ish years old, then the idea that "Eurasiatic" would be around the same time depth is perhaps mistaken, and whatever real common inheritance we have there is actually from much further back; maybe 20 000, maybe 40 000, maybe 100 000 years old. We really don't have established detailed precedents on what a language family of 10 000 or 15 000 or 20 000 (etc.) years of age "should" look like.

Pygmies are not nearly as divergent as Neanderthals or Denisovans!

I know. By now we also know though that they have some archaic admixture (same way how Europeans have some Neanderthal admixture). So far this has been modelled by a single ancient mixing event, but more than one is conceivable: pre-Bantu-expansion "substrate Pygmies" could have been even more divergent than what they are today, with a larger proportion of archaic population inheritance.

[mèþru]

What about what I wrote (multiregionalism within Africa and several times humans went to the Middle East and died out there), which is also what I've seen in most of the literature I've read (published by professors within the field, not bloggers)? It seems far more simpler (Occam's razor) than refuting the origins in Africa.

[Tropylium]

What about what I wrote (multiregionalism within Africa and several times humans went to the Middle East and died out there), which is also what I've seen in most of the literature I've read (published by professors within the field, not bloggers)? It seems far more simpler (Occam's razor) than refuting the origins in Africa.

You need to unpack "humans" in a discussion like this, for starters. Genus Homo clearly and undoutedly arose in Africa. The more relevant question is however where Homo sapiens sapiens i.e. anatomically modern humans did.

I'd be interested in reading reasons why all humans in the Middle East dying out periodically should be assumed as the "simplest" solution, even though at least Homo erectus and the ancestors of Neanderthals spread out just fine.

(Of course, archeology and paleo-DNA research will keep going, and in some years we'll have a clearer picture still.)

[mèþru]

I mean Homo sapiens sapiens.

The solution is a logical consequence of the Sahara pump. I wouldn't be surprised if similar things happened to other species in the same time period and place (time period being essential, as the Sahara was not cyclical in some periods)