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

[mèþru]

I thought Dacian exists as a group of geographic, tribal and personal names that may or may not simply be a regional designation for Thracian.

I didn't say Italic is a sprachbund, I said that Italo-Celtic and some other languages in Western Europe seem to me to have gone through common areal features instead of inheriting all of the distinctive features common to both Celtic and Italic

TBH IDK about Thracian and Dacian's relations with Baltic, but I'm pretty sure Slavic is Baltic rather than a sister family

[KathTheDragon]

Full disclaimer: none of the ideas I'm about to suggest are really all that convincing.

There's some evidence that Italic and Celtic were the next to split off after Tocharian, namely that they did not participate in the innovation of introducing *y to the middle present. Of course, that doesn't have to mean anything - an alternative explanation is that the innovation started in the east and only got as far west as Greek and Germanic.

An alternative branching is that Germanic was third to break away, which could be supported by the simplicity of its verbal system compared to the complexities of Graeco-Aryan, which are also found in Italic. It's also very reminiscent of what we find in Anatolian. On the other hand, Tocharian also has a complex verbal system, and Germanic shows affinities with Balto-Slavic, perhaps even to the point of sharing a few sound changes. If so, Germano-Balto-Slavic could be a valid node.

[Richard W]

... but I'm pretty sure Slavic is Baltic rather than a sister family

No-one has yet disputed that in this thread today.

[Moose-tache]

"Dacian" is shorthand for whatever collection of place names you want to say are Dacian. The issue here is, it is impossible to know how many layers into the past you are looking, and since the only criterion you can use to determine if a certain name is from the layer you want to study is if it "looks Dacian," there is no way to establish a list of Dacian terms in a way that is not circular. There is one inscription that may be in Dacian, in the sense that it isn't written in any other known language, which consists of one word and two proper names. That's just enough information to determine that it was a language spoken by more than one person.

[Salmoneus]

I think the idea of Northwestern (Germanic, Italic and Celtic) seems to be surely more likely than not, given the shared inherited words, the shared loanwords, and the geographical proximity. Italic and Celtic can apparently be explained through shared sound changes, so I'm not sure why they're not considered a branch together anymore, but even if they were only "very closely related possible mutually intelligible adjacent dialects that shared grammatical and lexical innovations as well as sound changes", I'm not sure why that really matters - as I say, expect pure, isolated-on-different-islands tree development right from the fall of adam and eve is unrealistic. [Similarly, Albanian and Balto-Slavic share both lots of isoglosses and key sound changes]. Regarding the "but they only share a few soundchanges so it could be coincidence!" line of attack, which can be applied against ANY suggested family relationship - what do you expect? When you're talking about very close dialects that had only very recently diverged from the parent, you wouldn't expect more than a few shared sound changes.

Let's get a little historical perspective. By around 5,000 years ago, Tocharian, Anatolian and possibly Greek had diverged. But by 4,000 years ago, the Indo-Aryan, Nuristani and Iranian languages had broken off from that, and Greek certainly had only a few centuries later (though probably even earlier). It's possibly that the northwestern languages were already broken off only a few centuries after 5k ago, if you see them as remnants of the bell beaker expansion. Not much less than 3000 years ago we already certainly have Italic and Celtic as totally separate branches.

So of the maybe 5,000 years of non-Anatolian, non-Tocharian IE, we can say that the surviving branches had all broken apart at least 3,500 of those years, if not 4,500. So the length of time available for what we might call "divergent IE" - the period in which the surviving branches were breaking apart from one another, is a matter of only centuries, at most not much more than a thousand years most likely. How many unique soundchanges can we really expect in that period? If branches in that brief period shared a handful of sound changes, some grammatical innovations and some noticeable developments in the lexicon, isn't that enough to talk of 'grouping' those languages?

I don't doubt the perfect-isolation model of language families makes sense when you're looking at the bigger picture of families like IE. But when you're talking about the spread of just a couple of soundchanges across just a few centuries, it makes no sense to pretend that these things worked in perfect, simultaneous schisms, rather than clusters of developments within dialect groups.

[Tropylium]

Regarding the "but they only share a few soundchanges so it could be coincidence!" line of attack, which can be applied against ANY suggested family relationship - what do you expect? When you're talking about very close dialects that had only very recently diverged from the parent, you wouldn't expect more than a few shared sound changes.

When we're trying to generalize the lesson to other families, this is indeed the point: since minor similarities could be a coincidence, they cannot be used to decide on what to do with other features where we don't know about directionality. So e.g. distribution only in "Eastern Cushitic" or "Malayo-Polynesian" might not be a sufficient reason against reconstructing some vocabulary or piece of morphology already for Proto-Cushitic or Proto-Austronesian.

[Howl]

About a dozen of Bomhard's proto-consonants are only based on the evidence of one branch; a few of these are nominally continued in a larger number, but in practice the overlap between their evidence is too small to make any difference.
– alveolar vs. postalveolar sibilants *cʰ, *cʼ, *dz, *s | *čʰ, *čʼ, *dž, š (only distinguished in Kartvelian)
– velar vs. uvular stops *kʰ, *kʼ, *g, *kʷʰ, *kʷʼ, *gʷ | *qʰ, *qʼ, *ɢ, *qʷʰ, *qʷʼ, *ɢʷ (distinguished in Kartvelian, Central Cushitic [a subfamily of a sub-family!] and Chukotko-Kamchatkan)
– velar vs. pharyngeal fricatives *x, *ɣ | *ħ, *ʕ (only distinguished in North Afrasian plus Southern Cushitic; the latter only in Semitic)
There's even a *ʔʷ posited in just one word, that seems to be based on no evidence at all and could be immediately merged with *w.

You could easily rack PIE up to 40-ish consonants by this type of maneuvering: e.g. reinstate the voiceless aspirates, reconstruct a series of "half-palatalized velars" whenever Baltic has plain velars corresponding to II first palatals, set up an *ɫ just for unclear reflexes in Albanian or Armenian…

Yes, that is precisely why I am skeptical of such bloated protolanguage phoneme inventories. As some of you know, I consider it likely that the "Mitian" langauges (IE, Uralic, Turkic, Mongolic, Tungusic, Yukaghir, Chukotko-Kamchatkan, Eskimo-Aleut) are related to each other, while I am doubtful about the relationship of Kartvelian, Dravidian or Afroasiatic to them. And of the protolanguages of the eight Mitian families, PIE has the largest consonant inventory. The other MItian inventories are quite similar to each other: a row of voiceless stops at four points of articulation (labial, dental, palatal, velar; to which ChK and EA add uvular) with a matching set of nasals and of voiced obstruents which may be either stops or fricatives; in addition to that some sibilants and liquids. One would guess that the Proto-Mitian inventory was similarly structured. It seems as if PIE was the "odd man out" here, and the Proto-Indo-Uralic inventory (if that's a valid node within Mitian) more like the Uralic one.

Didn't we agree to do macrofamily talk in this thread?

There are many things that make me skeptical about Nostratic. Like the sound-correspondences for the consonants which are nearly identity correspondences except for the palatals, lateral affricates and laryngeals. And how it is possible that the language that the Berbers speak is from the same family as the language that the Inuit speak?

[Pabappa]

I dont see a problem with Berber and Inuit being related. After all, if you retract your reach inward just a tiny, tiny bit, you get Spanish and Russian, which are well known to be related.

But none of the Nostraticists have made a convincing case for their theory, so just because it *could* be true doesnt mean it is.

[Whimemsz]

.

[Znex]

And to Tropylium's points
<snip>

I'm reminded of a similar case concerning the Mandarin Chinese dialects: some dialects have a number of words with a nasal initial /n-~ŋ-/ where all other dialects have a null initial. People have been tempted to treat these as a new correspondence, but historical and comparative evidence of the Chinese languages as a whole reveal that these dialects all in common developed rhinoglottophilia and the expected null initial holds historically.

[WeepingElf]

Didn't we agree to do macrofamily talk in this thread?

Yes, we did. I was just about to propose to continue the discussion here, but you beat me to it

[Salmoneus]

To WeepingElf's points, I can add that there's been some similar shenanigans with the only linguist who tried fully reconstructing Proto-Algic (the ancestor of Proto-Algonquian + Wiyot + Yurok, which probably has a time-depth of something like 7,000 years and is, realistically, beyond our ability to reconstruct much of beyond a few words and grammatical morphemes, and some grammatical idiosyncrasies). [...](Mentioned briefly at the bottom of [url=https://miidashgeget.wordpress.com/2019 ... ian/#5four]this footnote

A fascinating monograph! Is it yours? [a shame that the blog seems to be inactive now]

I'm curious, though - and since this is the macrofamily thread and Algic is one in a way - why is reconstructing Proto-Algic so hard? A 7000-year timedepth makes it a relatively closeknit family, after all, similar to or slightly older than IE and Semitic. Is it just that the Ritwan languages are under-described?

(I know that we have advantages with IE, in the form of a few millennia of 'cheating' through ancient greek and sanskrit and eventually latin texts. However, while those undoubtedly helped, a swadesh list seems to make clear that the family would have been swiftly recognised and eventually reconstructed pretty well even without the ancient texts, just from the modern languages. Similarly, while the fact that the non-Algonquian languages only have two representatives, rather than two dozen, undoubtedly means that some details may be lost... but your skepticism seems relatively extreme. If we only had the modern Indo-Iranian languages, and, say, modern Russian and modern Spanish, I'm sure we wouldn't be able to come up with PIE as we think of it today, but I'd think we could do much better than reconstructing only 'a few words and grammatical morphemes, and some grammatical idiosyncracies'! Do you disagree with that, or do you think there's some feature of Algic that makes it much harder to reconstruct - or is it just a description issue?]

I'm also curious - given our lack of prehistoric American knowledge, what makes us think Algic is around 7,000 years old, rather than 3,000 or 16,000? Is there an archeological basis for this, or is it based just on glottochronology (formal or informal)?

Finally, reading that page (well, some of that page...), I'm struck that the proto-Algic reconstructions there seem extremely close to the proto-Algonquian*, and closer than any Algonquian language seems to be to proto-Algonquian. It seems implausible that this version of proto-Algic could possibly be, combining your proto-Algic date and that page's proto-Algonquian date, 4,000-5,000 years older than proto-Algonquian! So, do you put proto-Algonquian further back than that page does, or as I suspect do you disagree that page's proto-Algic reconstructions? Are they, indeed, an example of just pushing one daughter's features back onto the parent - they seem to just be proto-Algonquian with what I assume must be a few minor accomodations to explain the other daughters?





*
kʷet- > (ne)kwet(w)-
nyi·ḱ- > nyi·š(w)-
nekθ- > neʔθ(w)
nya·ʔw > nye·w-
wetapy- > watapy-
kleḱy- > kešy-
weθkʷen- > weθkwen-

These are less divergent than two neighbouring English dialects! They're virtually identical, and even some of the differences seem like they may just be notational (eg *kw vs *kʷ, where the site notes that the former may actually perhaps have been the latter still in PA; likewise, ḱ>š is an even smaller leap if we assume that the former may already be something like /tS/ (which also seems to match the Wiyot reflexes with /t/ just as well)). If this is really proto-Algic, presumably it must have been spoken at most a couple of centuries prior to Proto-Algonquian...

Indeed, looking up the Meskwaki reflexes of the numerals and comparing to proto-Algic:
kʷet- > nekoti
nyi·ḱ- > nîshwi
nekθ- > nethwi
nya·ʔw- > nyêwi
Bearing in mind that the interpunct symbolises length, and I assume the circumflex does likewise (or perhaps a tonal distinction derived from length?), and that -(w)i may be either a suffix or the result of the sort of rhyming influences one often encounters in counting it would seem that Meskwaki has barely changed in the last 7,000 years!! Compare that to /ainaz/, /twai/, /Tri:z/, /fedwo:r/, /fimf/ becoming /wVn/, /tu/, /Tri/, /fO/, /faiv/ in around 2,000 years (let alone /kwetwores/ becoming /fO/ in allegedly less time than it took /nekT/ to become /neTwi/ or /nya:?w/ to become /nye:wi/...). I know nothing about Algic languages, I freely admit (sadly, as they seem fascinating), but surely either that time-depth or that reconstruction must be wildly wrong...

[jal]

Compare that to /ainaz/, /twai/, /Tri:z/, /fedwo:r/, /fimf/ becoming /wVn/, /tu/, /Tri/, /fO/, /faiv/ in around 2,000 years

Far less than that, I'd say. /wVn/ instead of /on/ is only a few hundred years old, and so afaik is /tu/ for older /two/. So we have a few thousand years (more than 2000, since you seem to have used proto-germanic) of pretty little change, then a short period of major change. Which might be a normal development a la punctuated equilibrium.


JAL

[Salmoneus]

Compare that to /ainaz/, /twai/, /Tri:z/, /fedwo:r/, /fimf/ becoming /wVn/, /tu/, /Tri/, /fO/, /faiv/ in around 2,000 years

Far less than that, I'd say. /wVn/ instead of /on/ is only a few hundred years old, and so afaik is /tu/ for older /two/.

The /w/ in 'one' is apparently from around 1600 (after the raising of /O:/ to /o:/, but before the strut-foot split) and is irregular (though understandable). The loss of /w/ in 'two' is regular, and slightly later.

So we have a few thousand years (more than 2000, since you seem to have used proto-germanic)

PGmc is considered to be about 2000 years old or very slightly older. Spoken after 500BC, before around 400AD.

of pretty little change, then a short period of major change. Which might be a normal development a la punctuated equilibrium.

English certainly looks like that, yes, though I can't say for sure how representative it is. It seems rational, though - major changes spur other major changes to compensate.
I wouldn't say there was pretty little change, though - the big moment for 'one', at least, was in the fairly rapid development of Anglo-Frisian, when it quite suddenly went from /ainaz/ to /a:n/. (loss of final /z/ and of final /a/ (could be other way around, loss of unstressed final-syllable /a/ then of final /z/), and monophthongisation).

But yes, the GVS and its aftermath did wreak a lot of havoc quite quickly. Before the GVS, when everybody still lived somewhere in Somerset, it was /O:n/, /twO:/, /Tre:/, /fower/, /fi:v/. [ironically, the closest one to PGmc today, 'three', came indirectly, having at one point been as far away as /θrøː/.]

[Nortaneous]

Compare that to /ainaz/, /twai/, /Tri:z/, /fedwo:r/, /fimf/ becoming /wVn/, /tu/, /Tri/, /fO/, /faiv/ in around 2,000 years (let alone /kwetwores/ becoming /fO/ in allegedly less time than it took /nekT/ to become /neTwi/ or /nya:?w/ to become /nye:wi/...)

cf. the Austronesian words for two through five (arbitrarily taking languages whose names begin with M):

Code: Select all

PAn    | Malay | Malagasy | Maori | Maga Rukai | Mantauran Rukai | Marshallese | Mangaaba-Mbula
*dusa  | dua   | rua      | rua   | ɖusa       | nosa            | ruo         | ru
*təlu  | təlu  | telu     | toru  | turu       | toɭo            | jilu        | tel
*səpat | əmpat | efatʂ    | whā	  | patɨ       | patə            | emān        | paŋ
*lima  | lima  | dimi     | rima  | rima       | ɭima            | ļalem       | lamata

[Whimemsz]

.

[Salmoneus]

There just don't seem to be many readily identifiable cognates (unless you go Proulx's route and cheat). If all we had access to were modern Romance languages (with no attestations from before 1500 and limited attestations of most of the languages until the late 19th century), plus Armenian, plus Albanian (and also Armenian and Albanian are now both dead and all the documentation we have of them is three-four books written by linguists of varying quality, some texts, and some audio recordings, all from the 20th century [also none of the three branches share any loanwords with one another, and Albanian and Armenian lack any particularly distinctive dialects]) -- which I think is a better analog for what we have in terms of Algic languages -- I think it would be virtually impossible to reconstruct very much of PIE.

Really? I think we'd work it out. Well no - I think a lot of the grammar might be completely unrecoverable, but I think the basic sound system and much of the core vocabulary could be...

*looks at Albanian and Armenian Swadesh list...*

...OK no, you're probably right. And a lot of the cognates we'd find between Italian and Albanian would actually be loanwords from Latin...

Although it does seem a bit suspicious that we happen to only have the Armenian and Albanian of the family...

Peter Denny, the archaeologist mentioned in the blog post, who links the pre-Proto-Algonquian speakers on the Columbia Plateau with the Western Idaho Archaic burial complex, also suggests two older burial sites in the area as representing Proto-Algic speakers, namely the Cascade Phase burials (ca. 4800 BC) and DeMoss burial site (ca. 4000 BC), which supposedly share a number of features with Western Idaho Archaic (and were identified by a previous author as the antecedents of Western Idaho Archaic).

I'm a bit suspicious, though - I suspect that if you pick a time period, you'll be able to find a candidate culture...

What I think is far more significant, though, is just that some of these languages seem to be ridiculously more conservative than others, for reasons we can only speculate about. The given Proto-Algic reconstructions can perfectly account for not only PA, but also the Wiyot and Yurok forms, so there's no reason to think that, say, Wiyot and Yurok split off from an older language that later gave rise to PA -- the available evidence simply indicates that PA was extremely conservative, at lease phonologically

See, this makes me wonder whether part of the problem is with the reconstructions rigging the game more substantially.

Just because a given form is able to account for all the descendents doesn't mean it was the actual parent form. In particular, what I worry might have happened here is that this 'Proto-Algic' may actually be a form of Early Proto-Algonquian from which the sound values of Proto-Algic are still more or less "recoverable", such that it's possible to falsely derive Ritwan from it. Since many of these changes seem weird to begin with, it's hard to be certain of the direction of change in some cases. Particularly if all the daughter languages underwent similar chages.

But maybe that doesn't make sense.

Then there are some daughter languages of PA that are phonologically ridiculously conservative (Meskwaki, at least as spoken up to the beginning of the 20th century, is definitely the most conservative [and yes, the circumflex marks vowel length], but this category also includes languages like Ojibwe and western Cree), others that show a moderate degree of change like we would expect of a 2,000-3,000 year old family (e.g., Menominee and Massachusett), a few that show rather more dramatic but not insane changes (Mi'kmaq [Mi'kmaq is both fairly phonologically innovative and very lexically and grammatically innovative] and East Cree-Montagnais-Naskapi), and some (Blackfoot, Arapahoan, Cheyenne) that show a batshit level of innovation. Some of the batshit ones still retain a large amount of native Algonquian vocabulary (Arapahoan, Cheyenne), while Blackfoot has very few cognates with other Algonquian languages, and as noted, Mi'kmaq, though less batshit in terms of phonological innovation, has also had very significant lexical turnover.

I suppose my point again, to put it another way, is: there's an infinite number of possible parent languages that could explain all daughters in a family. Linguists pick the most 'likely' on the basis of aesthetic taste, citing subjective ideas like parsimony. But the most parsimonious from a given data set isn't necessarily the right one. In particular, I suspect that people are systematically biased toward older-established members of families and more numerous subgroupings.

So, in your PIE what-if, I suspect linguists, given only Romance, Albanian and Armenian, would reconstruct PIE as looking very like Latin, but with a few small changes to allow the other languages to be derived - leaving Romance looking very conservative indeed, and Albanian and Armenian looking as insane as it was possible to be while still being plausibly derivable.

[I mean, obviously, Albanian IS insane, and Armenian is half-insane and has a split personality due to overwhelming borrowing... but my point is that in this scenario we'd probably reconstruct them as looking insane even if they weren't...]

Having said that, obviously it IS possible for a language not to undergo major soundchanges - or, perhaps more accurately, not to undergo sound changes that become cumulative...

Maybe I just think that perhaps more expansive speculation may be justified when we know that we don't have much information and that something's odd somewhere. For instance, in IE, something like Cheyenne giving too completely different reflexes of *p might indeed be best shrugged off as a single-language oddity - after all, if there were an additional stop series or some weak cluster-forming spirant or whatever, surely it would be seen outside just one branch? But if you know that only a few branches have survived, and that Cheyenne is indeed relatively basal, AND that the crown languages otherwise seem bizarrely conservative, then maybe it's OK to be more adventurous in suggesting that Cheyenne preserves something tht the other languages have lost. Particularly when it comes to finding and explaining related languages, since positing proto-languages that are more generous is more likely to let you see links to other cousin-languages, or to explain the links you find, rather than by assuming the most parsimonious reconstruction at each stage...



Anyway, sorry for the ignorant questions and ramblings. The material you shared really is fascinating, thank you!



------



One small question: proto-Algonquian only had one stop series, I gather? Wikipedia says 'Ojobwe' has two, and that 'Algonquin', which may just be a dialect of Ojibwe, has three. Is Wikipedia talking nonsense, and if not, do you know where the additional series come from?

[Whimemsz]

.

[Znex]

<snip>

This is all so fascinating to think about! It makes me wonder about the various "experiments" that we've had here on the ZBB, and on the CBB, and elsewhere, where we've engaged in reconstructing the proto-languages of (designed) conlang families.

I must admit I'm not terribly familiar with the results of such endeavours, or whether there's a way to reliably quantify any results so they have scientific merit to the field of reconstruction. Does anyone recall what they've seen of these threads and the general trajectory of different reconstruction attempts?

[Tropylium]

*looks at Albanian and Armenian Swadesh list...*

...OK no, you're probably right. And a lot of the cognates we'd find between Italian and Albanian would actually be loanwords from Latin...

Although it does seem a bit suspicious that we happen to only have the Armenian and Albanian of the family...

That's probably just the worst-case scenario (one where I think very few people would accept the relationship), but you could get a fairly similar effect also by picking something like Romance + Tsakonian + Romani, or Germanic + Romanian + Bulgarian, or Slavic + Scots + Irish Gaelic, etc. It's probably doable to show them to be related (cf. that showing Hungarian and Sami alone to be related was done already in 1770) but you'd again really not end up getting a good picture of PIE out of it.

what I worry might have happened here is that this 'Proto-Algic' may actually be a form of Early Proto-Algonquian from which the sound values of Proto-Algic are still more or less "recoverable", such that it's possible to falsely derive Ritwan from it. Since many of these changes seem weird to begin with, it's hard to be certain of the direction of change in some cases. Particularly if all the daughter languages underwent similar chages.

This sounds right enough, it's in plenty of cases easier to establish a sound correspondence than to establish where the starting point between them was. Did Proto-Turkic have *z *š or did it have *rʲ *lʲ? Did Proto-Uralic have *ð or *d or *tɬ or *ɬ? Did PIE have *t *dʰ *d or *t *d *ɗ or *tʰ *t *tʼ…?