Subdivisions of Niger-Congo

[Xephyr]

Perhaps somebody on the board knows about this. So, Niger-Congo is probably a bust as far as it being a demonstrable genetic clade goes (or so I've heard). So what does that leave? What are the subdivisions formerly-within-Niger-Congo which actually are, in & of themselves, clades? Like, is Benue-Congo still a thing? Is Atlantic-Congo? Or is it really just Bantu + a hundred families each with a tiny handful of languages?

[zompist]

Have you been reading RMW Dixon? 'Cos Niger-Congo was one of his targets. He said he was so suspicious of Greenberg's Amerind that he went back to his African work to see if it was any better, and it wasn't. It was just put together with mass comparison, not the comparative method.

I don't know what the Africanists are saying these days, but you could get a vague picture by looking through my numbers page. I'd say it's not very much like Amerind. Here's my attempt:

Mande - looks good
West Atlantic - kind of a mess
Kru - looks good
Gur - good, can see resemblances to Kru
Adamawa-Ubangian - ditto
South Central - holds together, has similarities to above
Cara + Nyima - also looks good
Broad Bantu - hard not to relate to Bantu

So the worst case is that there's half a dozen or so families. But everything past Kru looks like it could well be related, with 2 as something like *ba, 3 as *ta(t), 4 as *na(n). (Note that 5-9 are generally derivations.)

Take this with a bag load of salt, as it's basically mass comparison

[Salmoneus]

My understanding, which is not from any professional-level study mind you, was that Atlantic-Congo was considered pretty robust.

Mande, Dogon, Ijoid, and various minor families and isolates would then be potential but undemonstrated members.

Apparently Dimendaal wants to take out Ubangian as a separate family, but says that Savannas (Gur plus Adamawa minus Ubangi) is proven beyond doubt.

Glottolog seems to accept 50 language families in Africa (excluding sign languages, pidgins, IE and Austronesian). Of these, Atlantic-Congo, Mande, Dogon, Ijoid, Heiban, Talodi, and Rashad are the ones that conventionally make up Niger-Congo. Lafofa, Laal, and Jalaa are isolates also often associated with NC.

Of AC, glottolog says: "The core is held together by regular sound correspondences in lexical items between subfamily languages, less systematic verbal extensions and noun class systems."

Nilo-Saharan, on the other hand, Glottolog rips into tiny, tiny shreds.

[vokzhen]

I don't know what the Africanists are saying these days, but you could get a vague picture by looking through my numbers page.

I heard somewhere, maybe on here, a "joke" about this. Show an IEist Niger-Congo and they'll wonder where the evidence is. Show an Africanist Nostratic and they'll wonder where the controversy is.

[sirdanilot]

Have you been reading RMW Dixon? 'Cos Niger-Congo was one of his targets. He said he was so suspicious of Greenberg's Amerind that he went back to his African work to see if it was any better, and it wasn't. It was just put together with mass comparison, not the comparative method.

The issue with Greenberg's amerind is not that he use mass lexical comparison. It's that he didn't use mass lexical comparison, but simply based a classification on his earlier work (which is probably loosely based on mass lexical comparison), and in his tome 'Language in the Americas' he decided to 'support' this with some cognates. However, any two languages usually have only like 5 cognates out of >200 words that are used in the comparison. And of course we know that his data have many, many problems of being inaccurate or sometimes plainly wrong. For example, he links Páez (commonly held to be an isolate) with the languages of the Barbacoan family based on an incorrect word list that has a mixture of both Guambiano (a barbacoan language) and Páez words. Probably either by a confused speaker, by two erroneously combined wordlists, a mixed language that is now dead or for some other reason (Note that both Guambiano and Paez are still living languages spoken in close proximity to each other). Go read Curnow, Campbell etc.

I cannot judge if this is so for his African work. I think that for african languages, much more data is available, and thus his african work is probably a bit better. But that of course doesn't allow us to blindly accept his classifications.

[Tropylium]

Niger-Congo presumably has the same problem as Sino-Tibetan — there is a solid core along the lines of "Wolof is related to Kwa is related to Bantu", but also hundreds of poorly researched languages whose affiliation has never been shown rigorously. Things like Bangime and Laal are probably only the tip of an iceberg. (The same problem also extends to Chadic and Cushitic to an extent, FWIW.)

The subgrouping contains much to work on too. "Atlantic" just means "anything northwestern that looks more Bantu-like than Mande does". If Mande is not a member, or if it is innovative, then Atlantic obviously becomes only an areal group. Kordofanian also only seems to be defined areally, kind of as if people attempted to cram Armenian and Ossetian together as "Caucasian IE".

[Salmoneus]

Glottolog does indeed have no 'Atlantic' node, although most Atlantic languages are within a 'North-Central Atlantic' node, one of five divisions of Atlantic-Congo.