Cladistics, the tree model, and bacterial evolution.

[Rhetorica]

There's a peculiar language spoken in Tanzania, Mbugu, the classification of which has apparently been a great deal of stress—it seems it was originally a language from the Cushitic family, but due to prolonged exposure with Bantu languages, actually lost all of its morphology, retaining only the Cushitic vocabulary. R. L. Trask's Historical Linguistics (1996) has this to say: "Indeed, some linguists have concluded that [Mbugu] now is a Bantu language, since grammatically it is virtually indistinguishable from any other Bantu language – though most historical linguists find it impossible to accept the suggestion that a language can actually move from one family to another through contact."

Apparently the debate's resolved—Ethnologue lists it as "mixed Bantu-Cushitic", but I can't help finding it weird that there was debate about this at all. Genaeological trees are, by definition, full of mergers, so why the insistence on assuming language contact is insignificant to lineage?

I suppose the obvious scapegoat is a fondness for imitating biology (which is totally cool with me because, as a biologist with linguistic inclinations, the analogy is often very good), but absolute lineage is a eukaryote's game, not a bacterium's, due to horizontal gene transfer¹. Some very worthwhile analogies have been drawn between horizontal gene transfer and borrowing of grammar and words—transferred genes even get assimilated over time, taking on something akin to more natural phonology so they can be pronounced more easily².

The thing is—Mbugu's peculiar story demonstrates something that's very recently being acknowledged in biology: it doesn't look like prokaryotes actually have a single common ancestor, but more likely a community of them. The same can, almost certainly, be said at every level of the tree of life within the superkingdom of Bacteria: whenever borrowing is possible, there's no limit on how much can be borrowed.

This is mostly inconvenient for us in biology, as it (potentially) means we now have to track the evolutionary histories for each gene separately. A bacterial strain is no longer a set corpus with slight variations, but a momentary ensemble of things that are constantly coming and going and being traded around, all depending on circumstantial utility. (I've made a quip or two about calling the stable ones "eigengenomes," but it looks like the term's just been swiped for a more abstract concept. Ho hum.)

However, for linguistics, it means it's time to really lay off on the certainties of the tree model. Mbugu may be relatively unique in how far it's gone, but I don't think it's appropriate to list, say, any modern European language as the product of just one family. Changes due to language contact need to be accounted for. Maltese, for example, with its profoundly non-Arabic vocabulary, shouldn't just be hanging off Siculo-Arabic given its significant borrowed English vocabulary, and Siculo-Arabic shouldn't just be hanging off Maghrebi given the heavy influence from Sicilian and Italian (which Maltese takes even further, so that's even more categories it needs to be added to.)

I get the feeling that at least some of the conservatism in the classification scheme comes from an imitation of eukaryote phylogenetics, which are much more boring and rarely involve much transfer of content.³ Perhaps bacteria are just really yucky, and that has been enough to drive people away from focusing solely on the correct analogy, or perhaps all of this stuff is just too recent and hasn't trickled over yet. On the whole this is pretty forgiveable, unlike some efforts to inspire theories in distant fields, but I really do hope this changes.

What do you think?

¹ Bacteria have a variety of mechanisms for scavenging bits of DNA and adding the contents to themselves.
² Literally speaking, the frequencies of letters, letter-pairs, and letter-triples shifts to match the rest of the chromosome, which can lead to increased efficiency in gene expression.
³ Although humans have a small number of bacterial genes imported by viruses, parasitism, and just dumb luck due to exposure.

[Drydic]

Awaken my child, and embrace the glory that is your birthright
Know that I am the Overmind, the eternal will of the Swarm, and that you have been created to serve me.

[Grunnen]

There's a peculiar language ...luck due to exposure

As a fellow biologist with linguistic inclinations, I couldn't agree more.

[Salmoneus]

Obviously the tree model doesn't always apply very well. Does anybody actually deny that? There are at least three reasons: a) mixed languages, b) dialect continua (even in PIE, people admit that 'PIE' was a continuum of related dialects and may not actually have been a single form of language spoken identically everywhere), and c) sprachbund effects.

However, the tree model is still useful because it is useful. That is, starting out by assuming a tree model enables us to do funky stuff we wouldn't otherwise be able to do - like, say, deduce the nature of PIE and of intermediary stages, based on a tree model with sequential branching. (Note that with Albanian and Armenian, say, two languages where we only have a single branch to work with, the intermediary stages are a lot more blurry). Thus it makes sense to try as hard as possible to assume (counterfactually) that everything follows a tree model, in order to be able to apply the comparative method, and only admit that it doesn't when the method conclusively fails.

I wouldn't imagine anyone's too obsessed with imitating eukaryotic phylogenetics - I'm not sure eukaryotic phylogenetics is really a main interest of most linguists. Plus, the tree model in linguistics dates from at least the 17th century - if anything, evolutionary cladistics was based on linguistics, rather than vice versa.

If there is any psychological barrier to the idea of a language switching families, I think it's a lot deeper than that. Ultimately, I'd identify two strong human instincts. The first is the belief that the past cannot be changed by future events, but remains constant. So, if a painting was painted in 1608 today, it'll still have been painted in 1608 tomorrow. Family affiliation is a statement about history - it's saying 'X evolved from Y'. So if X evolved from Y today, X will still have evolved from Y tomorrow. The second instinct is the belief that entities persist through time - the belief that, although many of my atoms are not the same as they were yesterday, I'm still the same person. Put those two instincts together, and you get a strong resistence to the idea of a language changing its origin over time.

To phrase the same situation in a different context: imagine I have a bicycle. Bits keep going wrong, so I keep replacing them, with parts I find lying around. Now, as it happens those parts were bought by you, and belong to you. So this is my bicycle, but more and more of it starts being made of bits that belong to you. Does that mean that one day if I replace the headlight suddenly it will be your bicycle and not my bicycle? Most people resist that idea, even if they know nothing of eukaryotic phylogenetics. They may admit that this bit of the bicycle belongs to you, and that bit, and this other bit, and that more than 50% of the bicycle actually belongs to you... but most people will at the very least have a strong instinct to say that the bicycle itself is not yours, even if you do have a right to demand the various parts back. Likewise, there is a strong instinct to say that a language remains descended from X, even if large parts of it have been replaced by spare parts from Y.

And if people are forced to a point where they can no longer claim that A is still A, once it has changed enough, then rather than saying that it has turned into B (in this case, that my bicycle has turned into your bicycle, or that a Cushitic language has turned into a Bantu language), they would rather say that it has become a new thing, C (in this case, that what we now have is a new bicycle composed of bits of my bicycle and bits of your bicycle, or that what we now have is a mixed language derived from both Cushitic and Bantu parents). It appears, while not inescapable, nonetheless a strong human bias to try to avoid thinking that things can just turn into other things willy-nilly. This is probably how we are able to navigate around the world at all - by positing persisting entities with distinct haecceities and immutable histories. The world gets very confusing otherwise!

[gach]

If we understand language ancestry as the path taken by gradual development, some form of a tree model sounds like a fair description of the ancestry of most languages. If a feature in a language, including a grammatical one, can be traced as a loan from another language, it's best to be understood describing the ancestry of something else than the receiving language as a whole.

Mixed languages require somewhat more care and I don't have the expertise to go deep into them. First of all, what do we even require from a language to be able to call it a mixed language?

b) dialect continua (even in PIE, people admit that 'PIE' was a continuum of related dialects and may not actually have been a single form of language spoken identically everywhere)

Unless you include some hypothetical case where unrelated languages converge to form some kind of a dialect continuum of mixed languages, I don't see how the existence of dialect continua would be against any reasonable variation of a tree model. As far as I understand, a tree model is best understood as a tree or a bush of speech varieties diverging from some common ancestor without any regard to the internal language borders.

[zompist]

Linguists, like God in Time Bandits, are not entirely dim. The wave model is an old alternative to trees. Tree models are of less use in analyzing dialects, and a Romance professor I knew, an expert in Sard, didn't put much credence in any sub-Romance groupings. Also check out R.M.W. Dixon's punctuated equilibrium model, which he finds works better for non-agriculturalists.

At the same time, there's good reason for the use of trees.

I'd underline Sal's statement: Family affiliation is a statement about history. No amount of borrowing will make English not come from Germanic, as it's a historical fact. Plus, it's synchronically meaningful— in its basic grammar, English really is more like German than French.

Anyone who really wants to make an issue of this should really, really read Thomason & Kaufman's Language Contact, Creolization, and Genetic Linguistics, which puts things into much-needed perspective. It's good to remember:

• Mixed languages like Mbugu are very, very rare. T&K only come up with about half a dozen examples worldwide. They're something to know about, but not something to upset the normal procedures of linguistics for.

• Large-scale lexicon borrowing, as in English, is quite common. It's impressive, of course, if you're a lexicologist. But lexicons don't give frequency information. The vast majority of the entries in the OED are Greco-Latin, but no English speaker, even a scientist, uses even a large fraction of those words. High-frequency words are more resistant to borrowing.

• To put it another way, borrowing is so common that it's not useful to rejigger categories because of it. We already have a way of saying that the English vocabulary has a lot of French in it; in technical terms, we say "the English vocabulary has a lot of French in it". Nothing is added by trying to make this fact into more than it is.

• English is not a weird outlier in the Germanic languages. All the coastal dialects (especially the nonstandard ones) have thrown away lots of morphology and have borrowed lots of French words. For that matter, German has reversed a lot of its own Romance borrowings. T&K have an extended case study on this.

[Terra]

Large-scale lexicon borrowing, as in English, is quite common.

[...]

High-frequency words are more resistant to borrowing.

Zomp, didn't you say just the opposite in your blog recently?

On the book: I’ve been revising based on readers’ reactions. Plus this weekend I added a bunch of German etymologies. I have to say, if German was a conlang, I’d have to say the lexicon creator had his head together. Lots of straightforward etymologies and calques. To put it another way, English is an outlier in how divergent the common and learned vocabularies are.

[zompist]

Heh, I shouldn't have said "outlier" in both statements.

They're aimed at different possible errors, though.

A beginning, English-speaking conlanger may assume that all languages have a learned vocabulary with roots unknown to the native speaker. That's what the statement about German is intended to address.

Other folks get worked up about the large foreign vocabulary within English, and assume that it makes English strange or unique, and it's neither.

[Hallow XIII]

It is worth noting that this aspect of German is a result of deliberate replacement of foreign words with native ones, be it by calquing, compounding or whatever.

[Rhetorica]

Just to clarify for Sal, I'm not saying that things suddenly stop being what they were—I'm saying all languages are mixed, and need to have their ancestry listed appropriately. It strikes arbitrary that we call Maltese a form of Arabic given that most of its vocabulary isn't. To return to the extreme example, Mbugu even has a whole second register consisting of a vocabulary that is almost pure Bantu; as I understand it, you can speak perfectly good Mbugu with hardly a single Cushitic word or inflection in sight.

Treating every case of language contact as a merger doesn't really impair the tests or metrics used with a tree model, also—it just makes the math a little hairier; at worst you can regard a language as a collection of distinct entities with different origins. (Hierarchical clustering is useless, of course, but given that we're arguing over labels you don't need that anyway.)

Mbugu shouldn't be treated an exception just because it doesn't fit the tree model nicely and seems to be nearly unique—that's cherrypicking to avoid difficult data, when the model is already stretched to the point of snapping around the nonlinearity of sprachbunds. You might as well refuse to theoretically acknowledge the composition of a baryon just because radioactive decay is rare. (Of course, that isn't rare at all, but pretend it's 1890.)

As for English—it genuinely has picked up bits of grammar and common words from Romance languages, particularly in the form of productive suffixes. These aren't neglible, and they aren't relegated to dark corners of academic journals. The language owes its expressiveness to these acquisitions and, to draw a bit of a Whorfian argument, Latin and French have influenced how a native English speaker views the world (to whatever minimal extent a non-strong Whorfian argument permits.) Sprachbunds seem to be the same thing; if word order has any implications for how people think, then these linguistic areas clearly have significant consequences.

The commonness of borrowing in general doesn't faze me. If anything, that's just more data being obscured by purism, even though it may be a very minor contribution. I would argue that being able to track influences actually has huge value, since language and dialect communities are an important proxy of cultures. (Or, at least, so says Nicholas Ostler in Empires of the Word.) By treating all parent languages with proportionate weighting, it's easier to understand how, for example, Romance languages and the cultures of their countries were affected by their native environments.

[Radius Solis]

Zompist already told you that linguists aren't entirely dim. They aren't, and neither are we. You are arguing against an extreme view of the tree model held only by crackpots. I do realize that you are not trying to be rude and that you are probably not entirely dim yourself; however, if you're going to try telling us what's what, I would recommend first getting some education in the topic at hand.

You can make a start by letting go of these laymen's ideas about what linguistic descent entails. It does not mean "where your words come from", nor "where your grammar comes from". It refers only to the process by which a past language, by means of gradual change that can be listed out development by development, became the present-day language in question. If you take Old English you can arrive at Modern English by applying one specific little change after another, hundreds altogether, or thousands if you're doing it for lexicon too. We don't normally go to all the trouble it would take to do that for lexicon, and the syntax part is tricky for other reasons, but writing out such a list of changes is everyday commonplace for a language's phonological history. In principle you could list every single lexical and syntactic change too, assuming you have all of the data.

So where do borrowed things fit into that? Well, let us ask: if you start from Middle French, say, can you make out such a list of developments that leads to English? No, you cannot. Instead, borrowings from French are simply entries on the list of things that happened to English. No matter what part of the list you look at, you are looking at things that happened in a prior stage of English; at no point does it become a list of the things that happened to Middle French. So when you follow the list back in time, you arrive at Old English, and then Proto-Germanic, and then Proto-Indo-European. And when we talk about a language's "ancestry", that relationship is all that we are referring to.

[zompist]

Radius's post is very good, and probably calmer than I could manage.

Before outsiders lecture linguists on language contact, it really is a good idea that they learn what linguists actually say. Though I've summarized some of the main points in T&K, you cannot refute a book by making one-paragraph responses to someone's summary. Read it.

[Rhetorica]

I'm not sure how to write this post. On one had, I'm very sorry for irritating you all and I absolutely want to promise I'll go off and buy and read T&K before I go any further, but on the other hand I don't feel that Radius's post addresses what I've been trying to say. I'm prepared to put my foot in my mouth and hobble off, but before I do so I'd like to make one more attempt to communicate.

To make a borrowing (such as French to English as in Radius's list of developments), a bilingual person has to make the decision to take a word (or other feature) from their personal idiolect of French and place it in their personal idiolect of English. In essence, they are combining two subsets of two idiolects and making a new, shinier one with the added word. (And then, of course, it propagates to the idiolects of others, but let's just keep it simple.)

As I understand it, the key distinction between this and an individual building a pidgin (suppose they're a very lonely conlanger) is that one of the subsets happens to not be a proper subset, but in fact the entirety of one language.

And that just doesn't strike me as a good distinction—what if a borrowing causes the deletion of a small number of native features or words, like in a Sprachbund-induced syntax shift? Then it's no longer a complete set being added to, and the outcome is more like a lopsided pidgin.

Middle French is a relatively minor ancestor of Modern English, sure, but you couldn't arrive at Modern English without contact with Middle French. It's still necessary to have it around for the new language to fully take shape, and I want to define parenthood as being that necessity.

A purely attributive model like this also scales better to cases of extreme borrowing, obviously, as it stops the historical classification of exceptional cases like Mbugu from being so disharmonious with their featural classification; we can just treat every language after every influence due to contact as a new pidgin and/or creole.

[Drydic]

You haven't irritated anyone. At least not to the point where you should worry about it or anything. We're just frank and at times a bit caustic with responses. Just look at how we bicker with each other.

[gach]

If you wish, you can consider the linear ancestry of a language merely a figure of speech. After all, it's just a model while we really should be concentrating on phenomena of the real world. It's just a pretty useful figure of speech and gives us insight into the prehistory of languages. Naturally this picture has to be supplemented with a discussion of convergence areas and lists of loanword layers.

[zompist]

A purely attributive model like this also scales better to cases of extreme borrowing, obviously, as it stops the historical classification of exceptional cases like Mbugu from being so disharmonious with their featural classification; we can just treat every language after every influence due to contact as a new pidgin and/or creole.

Which is exactly what we don't want to do. There is a discontinuity there, and there's reasons for that discontinuity--creoles do not form in the same circumstances as normal genetic descent. This is covered in T&K.

As it happens, you're reproducing the enthusiasm of some linguists from the 1980s, when pidgins and creoles received a good deal of attention. Many people had this reaction of "Aha! Everything's a creole!" Only, everything is not a creole.

[hwhatting]

Another thing - for family classification, not all elements are treated equally, for a reason. In cases like English, where we know the language history pretty well, it's relatively easy to say that there's a development starting with Old English, that there was Norse and Norman French influence, a bunch of loans from continental French, Latin, etc. So we can say that English started out as Germanic and, taking into account similar "analytic" tendencies in other Germanic languages, still is a Germanic language. Where we have languages with less attested history, we have to look at their synchronic features and try to reconstruct this history; with something like Mbugu, the question would be whether it's a Kushitic language that became Bantuized or a Bantu language with a lot of Kushitic loans. (This is just for illustration, I'm not going to discuss this question). What I'm getting at is that, based on the deveopment of languages the history of which we know more about, there is a loaning hierarchy, from easily loaned to rarely loaned:
(1) Lexicon, with specialized lexicon more easily borrowd than basic lexicon
(2) Derivative morphemes (nominal more easily loaned than verbal)
(3) Bound grammatical morphemes (again, nominal more easily loaned than verbal)
So, when in doubt about family assignment, linguists normally take the provenance of the basic lexicon and the bound grammatical morphemes as basis. For English, that again means "Germanic". Only in cases where the basic lexicon and the bound grammatical morphology is to a significant degree from two families one would start talking about mixed languages.

[Hallow XIII]

The big problem with this is that suddenly the ancestry lists of everything would explode and you could say with certainty maybe that if Proto-Human existed it is the definite ancestor. It doesn't serve any purpose, outside of those cases like Mbugu and Michif and whatnot, and also we *know* already that there was French influence in English. Putting "French" in the "ancestor language" field alongside Old English would be purely cosmetic and clutter everything up so what is the point.

[Radius Solis]

I'm not sure how to write this post. On one had, I'm very sorry for irritating you all and I absolutely want to promise I'll go off and buy and read T&K before I go any further, but on the other hand I don't feel that Radius's post addresses what I've been trying to say.

A moment's passing irritation on my end is of no consequence. But I thought you deserved fair warning that you were heading into territory that most often does not end well for the newbie. Note that you are perfectly welcome to argue a point! Just, beware that all the nicely constructed lines of reasoning in the world may still amount to making shit up, if you do not have a good understanding of a topic. And if you can't tell, which happens to me sometimes, I find it better to ask questions than risk going out on a limb that's easily sawed off. (I'm sorry to offer unasked-for advice in the middle of an L&L thread. I do so only in the hope it will help your time on the ZBB be more pleasant for you and for everyone.)

As for the remainder of my post, it was not meant to address all that you were trying to say. It was meant to address a particular point that seemed likely to be at the root of much of the issue. I should have been clearer about that.

Only in cases where the basic lexicon and the bound grammatical morphology is to a significant degree from two families one would start talking about mixed languages.

Right... sorta. I guess my prototype for a mixed language is something like Michif, which is famous for having taken all its basic nominal lexicon and bound nominal morphology from French and all its basic verbal lexicon and bound verbal morphology from Cree. And all of the usual top-billings in the "mixed language" category have the interesting property of having been created largely on purpose by a community subjected to very unusual circumstances, using linguistic spare parts that happened to be lying around within reach. I'm not sure these should be necessary conditions for qualifying as a mixed language, but Mbugu does appear to fail both standards.

[Rhetorica]

Alright. Thanks for humouring me, everyone. I'll try not to be too obnoxious in the future.

[hwhatting]

Only in cases where the basic lexicon and the bound grammatical morphology is to a significant degree from two families one would start talking about mixed languages.

Right... sorta. I guess my prototype for a mixed language is something like Michif, which is famous for having taken all its basic nominal lexicon and bound nominal morphology from French and all its basic verbal lexicon and bound verbal morphology from Cree. And all of the usual top-billings in the "mixed language" category have the interesting property of having been created largely on purpose by a community subjected to very unusual circumstances, using linguistic spare parts that happened to be lying around within reach. I'm not sure these should be necessary conditions for qualifying as a mixed language, but Mbugu does appear to fail both standards.

Well, yes, if Michif is your prototype, then Mbugu might not qualify. OTOH, there may be linguists out there who'd argue that Michif is Cree (because of the more "central" verbal morphology is clearly from Cree), with a strong French influence (as nominal morphology is more "peripheral"). These people might also argue that Mbugu is not a mixed language at all, if (as I understand) its bound morphology is all Bantu - they'd qualify it as a Bantu language with a strong Kushitic substrate (and making this a case where a language has indeed changed its affiliation). I assume that for such linguists, mixed languages probably don't exist at all.

[Hallow XIII]

Alright. Thanks for humouring me, everyone. I'll try not to be too obnoxious in the future.

What are you going on about, you're the best newbie we've had in a while.

[hwhatting]

Alright. Thanks for humouring me, everyone. I'll try not to be too obnoxious in the future.

What are you going on about, you're the best newbie we've had in a while.

Seconded.

[Drydic]

Alright. Thanks for humouring me, everyone. I'll try not to be too obnoxious in the future.

What are you going on about, you're the best newbie we've had in a while.

Thirded.

[sirdanilot]

The researcher that has worked on Ma'a/Mbugu, M. Mous, is at my university and I know him, so it's nice to see this pop up here.

This thread clearly shows that historical linguistics is such a dogmatic field, that ignores recent developments in other branches of linguistics. Mainly contact linguistics, that is. Historical linguists tend to stick very firmly to the classic top-down approach: language A branches off into languages B and C, and so on. Contact has some influence on languages mainly in lexicon and sometimes in other aspects of the language, but cannot have any influence on the genetical classification of the language. Creoles are something outside of the scope of historical linguists as they are way too problematic for their firm little models. There are works on the genetic classification of creole language (Thomason and Kaufman, Owens...), but they are mostly ignored by the hard core indo-europeanists.

Ma'a is a classical 'mixed language' which is typologically A but lexically B. In this case, Bantu and Cushitic. The genetic classification of this language is clearly Bantu, though if you are looking only at lexicon it's Cushitic.

People need to learn to accept the existence of mixed language and contact-induced phenomena, and get out of their dated Proto-Indo-European ivory towers. Linguistics doesn't work like that any more. In fact, genetic and historical development of a language may only be a fraction of what actually makes up a language, contact might have a much bigger impact than we have always thought.

This is also a reason that I am happy that the bachelors Indo European Linguistics and General Linguistics have been merged at my university; people now receive a broad linguistic knowledge before they specialize in Indo European comparative linguistics, rather than sit in their ivory towers from day 1. A very good thing, as they now also learn things about descriptive linguistics, contact linguistics, generative linguistics etc. (yes, bachelors are much more specified in europe than in America; I couldn't fathom such a narrow field being a bachelor major in the US)