Pan-Americanisms

[Richard W]

The OT discussion of the Amerind hypothesis raises a question. Is there a good discussion of Pan-Americanisms around? They sound like a good basis for a basis for a conspiracy theory - Greenberg was sloppy but right, but Campbell and others suppress the idea by denying an academic career to anyone who supports the idea of Amerind. If one avoids facts, one can claim that Amerind expanded and fractured into many groups early, and therefore one has no hope of recovering early branching. Indo-European seems to be an example of rapid expansion making subgrouping mostly difficult and contentious. Sino-Tibetan (non-ideological use of term) may be similar.

So, can anyone point me to facts, rather than rhetoric.

[Xephyr]

If one avoids facts, one can claim that Amerind expanded and fractured into many groups early, and therefore one has no hope of recovering early branching..

How early is early? Because if you want a "recoverable" Proto-Language, you would pretty much need to have the Proto-Amerinds migrate to North America 14,000 years ago and, despite having a vast, uninhabited region all to themselves, spend the next ~6,000 years sticking together in one small, cohesive group, like a little dot moving across the continent.

[zompist]

You can find good discussions in Campbell's American Indian Languages, Dixon's The Rise and Fall of Languages, and Larry Trask's Historical Linguistics. Ruhlen's A Guide to the World's Languages is an accessible account of the honorable opposition and has a good summary of previous work in the area.

Greenberg's "mass comparison" amounts to eyeballing a bunch of lexicons and guessing that certain languages are related. This in itself is normal and necessary. It is what happens before you apply the comparative method and attempt to rigorously prove a connection.

The controversy is not about whether mass comparison is useful, or whether Amerind is possible. It comes when Greenberg and Ruhlen deny that any other evidence is needed. (See Ruhlen, op. cit., p. 253: he describes it as a "serious error" to require "the presence of regular correspondences and the reconstruction of proto-forms", that is, the comparative method.)

The sad fact is that reconstruction takes a long long time. (Look at the fact that we have an active PIE thread: there are still controversies in this most studied of language families 200 years after the family was proposed!)

A lot of the families you see in overall classifications (including mine in the sci.lang FAQ) are not in fact supported by the comparative method, simply because no one's got that far yet. E.g. Greenberg's classification of African languages is generally accepted by Africanists, but has not been established by reconstruction. There are reconstructions of Bantu, but not Niger-Congo. There are reconstructions of Algonquian, but not Almosan.

As I mentioned in the other thread, something that seems obvious at first look— that is, when you apply mass comparison— may completely dissipate when you apply the comparative method. An example is "Quechumaran", a node intended to include Quechua and Aymara (and some remoter relatives of Aymara, such as Jaqaru). There is a large number of cognates... however, there is almost no variation between the Aymara and Quechua forms (there is more variation within Quechua), and the majority of words are spectacularly non-cognate. Grammatical endings don't match up at all. There's a good explanation for this: borrowing. One of the languages massively borrowed from the other, just as English did from French, Japanese did from Chinese, and Farsi did from Arabic.

(These remarks are based on my own attempt at reconstruction, and supported by e.g. Bruce Mannheim. Campbell also dismisses the surface level of cognates as borrowings, but believes he's found a deeper level that does point to a connection; but he gives no examples and I've never seen his data.)

Did Greenberg have some magic that allowed him to avoid such problems? No, he did exactly what I did when collecting numbers: he wrote down vocabulary items from dictionaries into big notebooks. (I'm pretty sure his notebooks are available at his university.)

You have to be particularly careful with Amerindian languages, because the data are often in bad shape. Lexicons always look nice in print, but they were often recorded by untrained people with English- or Spanish-trained ears. As just one example, Ives Goddard gives the example of Beothuk 'gathet' and 'yazeek', both words for 'one' from different sources, which have been compared to Proto-Algonquian *nekwetw- and *pe-šekw- respectively. But Beothuk is extinct and poorly recorded, and Goddard's best guess is that the two Beothuk "words" are two attempts at the same word, possibly something like /ɣażiʔ/. Specialists in the area can work out these things, but a generalist looking at published wordlists cannot.

As another example of how bad early work could be, Campbell mentions a work by Sapir (a perfectly respectable scholar) linking Beothuk and Algonquian with 16 words. Ruhlen discusses Sapir's speculations without mentioning how little basis some of them had.

If you want to dig into the fight, there is no conspiracy— go get the data and reconstruct a family that hasn't been properly treated yet. There's a huge amount of work to do in South America. In a couple hundred years maybe your successors can work up to Amerind.

[Xephyr]

A lot of the families you see in overall classifications (including mine in the sci.lang FAQ) are not in fact supported by the comparative method, simply because no one's got that far yet. E.g. Greenberg's classification of African languages is generally accepted by Africanists, but has not been established by reconstruction. There are reconstructions of Bantu, but not Niger-Congo. There are reconstructions of Algonquian, but not Almosan.

I actually think you're overstating to what extent "Almosan" is even a thing. I'm honestly not sure I've ever even seen it (or Aztec-Tanoan or Keresiouan or etc.) in "overall classifications", except as a footnote-- not even as an asterisked term-of-convenience like Niger-Congo or Altaic. The closest I've seen is Michael Fortescue, who iirc says that "Mosan" might represent a separate pre-NaDene migration if not a coherent linguistic grouping.

On the other hand, hopefully within 20 years or so Yuchi-Siouan will have been proven. And I've seen Muskogeanists state inequivocally that Natchez "is" related to Muskogean-- they just haven't yet started talking about a "Musko-Natchez" family yet.

As I mentioned in the other thread, something that seems obvious at first look— that is, when you apply mass comparison— may completely dissipate when you apply the comparative method. An example is "Quechumaran", a node intended to include Quechua and Aymara (and some remoter relatives of Aymara, such as Jaqaru). There is a large number of cognates... however, there is almost no variation between the Aymara and Quechua forms (there is more variation within Quechua), and the majority of words are spectacularly non-cognate. Grammatical endings don't match up at all. There's a good explanation for this: borrowing.

Yeah what's the deal with this btw?: apparently I've heard that Lyle Campbell is in favor of Quechumaran. Lyle Campbell, the splitter splittissimus of historical linguists? The guy who didn't even accept Na-Dené until 2010? Him?

As another example of how bad early work could be, Campbell mentions a work by Sapir (a perfectly respectable scholar) linking Beothuk and Algonquian with 16 words. Ruhlen discusses Sapir's speculations without mentioning how little basis some of them had.

16 words isn't a lot, but it depends on what the total attested corpus of Beothuk is. IMO you could in theory demonstrate a likely shared ancestry using an extremely small corpus. For example, Julian Granberry in his book on the Calusa argues (and I think he's right) that Calusa was a dialect of Tunica. This is based on an attested wordlist of 12 entries-- not a very long corpus obviously, except that 10 of them [or was it 9?] are more-or-less identical with their Tunica translations. Obviously Calusa could have easily had 10++ Tunica loanwords, but the language would have to be supersaturated with loanwords for it to be likely that 12 randomly chosen lexical entries comprise 10 words of Tunica origin.

[WeepingElf]

You can find good discussions in Campbell's American Indian Languages, Dixon's The Rise and Fall of Languages, and Larry Trask's Historical Linguistics. Ruhlen's A Guide to the World's Languages is an accessible account of the honorable opposition and has a good summary of previous work in the area.

Greenberg's "mass comparison" amounts to eyeballing a bunch of lexicons and guessing that certain languages are related. This in itself is normal and necessary. It is what happens before you apply the comparative method and attempt to rigorously prove a connection.

As I like to say about mass lexical comparison, the difference between it and the comparative method is that between asking a question and finding the answer. And the question mark won't straighten by itself. Greenberg's list of lookalikes doesn't say, "These languages are related", it says, "Are these languages related?". Of course, asking the right question is the first step towards finding the answer; but in the case of Amerind, we can't be sure that it is the right question. Of course, it is at least a legitimate question, but one may ask why Greenberg excluded Eskimo-Aleut (a well-established family) and Na-Dene (a grouping which is controversial but at least worthy of serious discussion). Probably, his reasons were that these groups largely coincide with populations that were shown to be late arrivals by means of genetics. But recent genetic studies and archaeological finds have drawn the "Clovis model", according to which all the other indigenous Americans descend from a single founder population that passed the Bering land bridge about 12,000 years ago, in question, and that undermines the Amerind hypothesis.

The controversy is not about whether mass comparison is useful, or whether Amerind is possible. It comes when Greenberg and Ruhlen deny that any other evidence is needed. (See Ruhlen, op. cit., p. 253: he describes it as a "serious error" to require "the presence of regular correspondences and the reconstruction of proto-forms", that is, the comparative method.)

The sad fact is that reconstruction takes a long long time. (Look at the fact that we have an active PIE thread: there are still controversies in this most studied of language families 200 years after the family was proposed!)

Yes. There are still a lot of controversial things in the reconstruction of PIE, despite IE being the best-studied family of all, and with the luxury of having ancient literary languages at disposal that bridge about half the time since breakup. With Uralic, things look worse, for instance, and with most other families, even worse still.

A lot of the families you see in overall classifications (including mine in the sci.lang FAQ) are not in fact supported by the comparative method, simply because no one's got that far yet. E.g. Greenberg's classification of African languages is generally accepted by Africanists, but has not been established by reconstruction. There are reconstructions of Bantu, but not Niger-Congo. There are reconstructions of Algonquian, but not Almosan.

Yep. Of Greenberg's four African macrofamilies, none has been reconstructed yet, though Afrasian and Niger-Congo show morphological similarities which suggest that these two at least are valid. But these resemblances aren't really closer than those between IE and Uralic! So there is a lot left to do here. It seems that Greenberg used the following algorithm:

1. If it looks as if it was related to Semitic, it goes to Afrasian.
2. If it looks as if it was related to Bantu, it goes to Niger-Congo.
3. If the above fail, and it has clicks, it goes to Khoisan.
4. If the above fail, and it doesn't have clicks, it goes to Nilo-Saharan.

So, two language family candidates, one grouping based on a single typological feature, and one wastebasket.

The Khoisanists at least seem to have realized that they have no valid evidence of their field of study being a single family.

At least, these four macrofamilies are somewhat useful for the purpose of classification. Amerind, in contrast, isn't.

As I mentioned in the other thread, something that seems obvious at first look— that is, when you apply mass comparison— may completely dissipate when you apply the comparative method. An example is "Quechumaran", a node intended to include Quechua and Aymara (and some remoter relatives of Aymara, such as Jaqaru). There is a large number of cognates... however, there is almost no variation between the Aymara and Quechua forms (there is more variation within Quechua), and the majority of words are spectacularly non-cognate. Grammatical endings don't match up at all. There's a good explanation for this: borrowing. One of the languages massively borrowed from the other, just as English did from French, Japanese did from Chinese, and Farsi did from Arabic.

(These remarks are based on my own attempt at reconstruction, and supported by e.g. Bruce Mannheim. Campbell also dismisses the surface level of cognates as borrowings, but believes he's found a deeper level that does point to a connection; but he gives no examples and I've never seen his data.)

Fine. If many open-class lexemes match but the morphology doesn't, one can be sure that one is dealing with loanwords. This was how Hübschmann established that Armenian was not an Iranian language in 1863. This is also what fuels the criticism levelled against the Altaic "family". The open-class lexemes are once again more similar than the inflectional morphology. (Though there are also similarities in the personal pronouns - which, however, also encompass IE, Uralic and a few others.)

Did Greenberg have some magic that allowed him to avoid such problems? No, he did exactly what I did when collecting numbers: he wrote down vocabulary items from dictionaries into big notebooks. (I'm pretty sure his notebooks are available at his university.)

Yes.

You have to be particularly careful with Amerindian languages, because the data are often in bad shape. Lexicons always look nice in print, but they were often recorded by untrained people with English- or Spanish-trained ears. As just one example, Ives Goddard gives the example of Beothuk 'gathet' and 'yazeek', both words for 'one' from different sources, which have been compared to Proto-Algonquian *nekwetw- and *pe-šekw- respectively. But Beothuk is extinct and poorly recorded, and Goddard's best guess is that the two Beothuk "words" are two attempts at the same word, possibly something like /ɣażiʔ/. Specialists in the area can work out these things, but a generalist looking at published wordlists cannot.

As another example of how bad early work could be, Campbell mentions a work by Sapir (a perfectly respectable scholar) linking Beothuk and Algonquian with 16 words. Ruhlen discusses Sapir's speculations without mentioning how little basis some of them had.

If you want to dig into the fight, there is no conspiracy— go get the data and reconstruct a family that hasn't been properly treated yet. There's a huge amount of work to do in South America. In a couple hundred years maybe your successors can work up to Amerind.

Amen!

[zompist]

As another example of how bad early work could be, Campbell mentions a work by Sapir (a perfectly respectable scholar) linking Beothuk and Algonquian with 16 words. Ruhlen discusses Sapir's speculations without mentioning how little basis some of them had.

16 words isn't a lot, but it depends on what the total attested corpus of Beothuk is. IMO you could in theory demonstrate a likely shared ancestry using an extremely small corpus.

Good point. It turns out that the Beothuk wordlists have 325 items. For more fun, none of the informants knew English, so the meanings were derived by "mime, drawing and pointing" (Campbell p 155).

[zompist]

(re Beothuk. Hit quote instead of edit.)

(For people who haven't studied Native American languages at all— this is an outlier; obviously we have languages that are still spoken and can be extensively documented. But there are a lot of languages that are either lost, or have never been studied better than when some random gringo wrote down what he could figure out. And the claim that Amerind is a valid genetic node does obviously made a claim about such poorly documented languages.)

[zompist]

I actually think you're overstating to what extent "Almosan" is even a thing. I'm honestly not sure I've ever even seen it (or Aztec-Tanoan or Keresiouan or etc.) in "overall classifications", except as a footnote-- not even as an asterisked term-of-convenience like Niger-Congo or Altaic. The closest I've seen is Michael Fortescue, who iirc says that "Mosan" might represent a separate pre-NaDene migration if not a coherent linguistic grouping.

I just grabbed it as an example. FWIW, Ruhlen and Greenberg accept it; Campbell discusses and rejects it; Voegelin and Lyovin mention the possibility; the Ethnologue doesn't list Beothuk; Wikipedia mentions that John Hewson promotes a connection.

[CatDoom]

You have to be particularly careful with Amerindian languages, because the data are often in bad shape. Lexicons always look nice in print, but they were often recorded by untrained people with English- or Spanish-trained ears.

Despite the extensive and oftentimes vicious destruction of indigenous cultures here in California, the data for many Californian languages is actually more complete than what we have for many extinct languages in other parts of the Americas, thanks in no small part to the work of J.P. Harrington. The guy was a machine when it came to fieldwork, but was more concerned with recording as much as he could than with organizing and publishing it. Linguists and native Californians have learned a lot from pouring over the hundreds of thousands of pages of notes he left behind, and the work of organizing and digitizing the collection is ongoing.

[Terra]

You have to be particularly careful with Amerindian languages, because the data are often in bad shape. Lexicons always look nice in print, but they were often recorded by untrained people with English- or Spanish-trained ears.

Despite the extensive and oftentimes vicious destruction of indigenous cultures here in California, the data for many Californian languages is actually more complete than what we have for many extinct languages in other parts of the Americas, thanks in no small part to the work of J.P. Harrington. The guy was a machine when it came to fieldwork, but was more concerned with recording as much as he could than with organizing and publishing it. Linguists and native Californians have learned a lot from pouring over the hundreds of thousands of pages of notes he left behind, and the work of organizing and digitizing the collection is ongoing.

Insert coffee, get data about nearly extinct language? Damn, that list gives 50 languages that he documented. I wonder how much time he spent in the field. How long is a usual stint of fieldwork? (if you want to document a language as thoroughly as he did.)

Edit: I was just gonna say that maybe his large output put on a strain on his marriage, which didn't seem to last that long, but apparently there's already a whole book (by his ex-wife) about that: http://www.amazon.com/Encounter-With-An ... 0826314147

[CatDoom]

Ethnographic fieldwork can go on for varying lengths of time, but I think it's safe to say that Harrington spent a lot more time in the field than just about any American academic does today. Then again, he wasn't working at a university, but rather for the Smithsonian's "Bureau of American Ethnology" (which no longer exists as such), a job which presumably carried different expectations.

I found this chronology, which gives some idea of how he used his time throughout his career, and also has some interesting quotes from contemporary figures in American Anthropology, including Franz Boas, Alfred Kroeber, and Edward Sapir.

[Richard W]

If one avoids facts, one can claim that Amerind expanded and fractured into many groups early, and therefore one has no hope of recovering early branching..

How early is early? Because if you want a "recoverable" Proto-Language, you would pretty much need to have the Proto-Amerinds migrate to North America 14,000 years ago and, despite having a vast, uninhabited region all to themselves, spend the next ~6,000 years sticking together in one small, cohesive group, like a little dot moving across the continent.

I wouldn't expect much - just a lot of strong possibilities, comparable to now-disbelieved equations such as English hound = Latin canis = Greek kyōn. (Yes, I deliberately mixed qualities here.)

[Richard W]

Of course, it is at least a legitimate question, but one may ask why Greenberg excluded Eskimo-Aleut (a well-established family) and Na-Dene (a grouping which is controversial but at least worthy of serious discussion). Probably, his reasons were that these groups largely coincide with populations that were shown to be late arrivals by means of genetics.

Well, in Classification of American Indian Languages: A Reply to Campbell in Language Vol 65 No 1 (March 1989), he writes:

I would like to emphasize the fact that my linguistic classification shows an almost exact match with genetic classification by population biologists and with fossil teeth evidence (Greenberg, Turner, & Zegura 1986). My linguistic classification was arrived at in total independence of this external evidence, and until recently I was unaware of the agreement.

[CatDoom]

I wouldn't expect much - just a lot of strong possibilities, comparable to now-disbelieved equations such as English hound = Latin canis = Greek kyōn. (Yes, I deliberately mixed qualities here.)

Since when are those disbelieved? I thought *ḱwō/*ḱ(u)wōn was a pretty well established PIE root. I mean, there are cognates in just about every branch of Indo-European, including several varieties of Anatolian and even a bunch of poorly attested languages like Thracian an Phrygian.

As for finding similar correspondences in a hypothetical Pan-American macrofamily, I wouldn't really expect there to be many. If every Amerindian language outside of Eskimo-Aleut and Na-Dene shares a common ancestor, one would presume that said common ancestor started to diverge somewhere between 13,000 and 16,000 years ago (assuming that certain older dates are discounted). With the exception of the paleolithic continuity hypothesis, none of the current theories about the Proto-Indo-European urheimat posit that the language was spoken before about 8,000 years before present, and most Indo-Europeanists posit that it can't be much older than 6,000 BP based on it having vocabulary related to wheeled vehicles.

Basically, reconstructing "Proto-Amerind" would be the equivalent of reconstructing the language that was to Proto-Indo-European what Proto-Indo-European is to modern English.

[Richard W]

I wouldn't expect much - just a lot of strong possibilities, comparable to now-disbelieved equations such as English hound = Latin canis = Greek kyōn. (Yes, I deliberately mixed qualities here.)

Since when are those disbelieved? I thought *ḱwō/*ḱ(u)wōn was a pretty well established PIE root. I mean, there are cognates in just about every branch of Indo-European, including several varieties of Anatolian and even a bunch of poorly attested languages like Thracian an Phrygian.

This wasn't meant to be a hidden trick. I thought the general opinion now was that Latin lacked a cognate. The suggestion is that canis might originally have meant 'puppy', and if Indo-European be derived from *ken 'young, to appear' as in Middle Irish cano, cana 'wolf cub', Welsh cenau 'wolf cub, puppy; rascal'.

As for finding similar correspondences in a hypothetical Pan-American macrofamily, I wouldn't really expect there to be many. If every Amerindian language outside of Eskimo-Aleut and Na-Dene shares a common ancestor, one would presume that said common ancestor started to diverge somewhere between 13,000 and 16,000 years ago (assuming that certain older dates are discounted). With the exception of the paleolithic continuity hypothesis, none of the current theories about the Proto-Indo-European urheimat posit that the language was spoken before about 8,000 years before present, and most Indo-Europeanists posit that it can't be much older than 6,000 BP based on it having vocabulary related to wheeled vehicles.

Basically, reconstructing "Proto-Amerind" would be the equivalent of reconstructing the language that was to Proto-Indo-European what Proto-Indo-European is to modern English.

Perhaps not so far from Indo-Uralic. Remember, what survives for millennia is resilient. Proto-Amerind should also have the advantage of multiple long-independent stocks, each capable of bearing independent witness to the common features.

[hwhatting]

This wasn't meant to be a hidden trick. I thought the general opinion now was that Latin lacked a cognate. The suggestion is that canis might originally have meant 'puppy', and if Indo-European be derived from *ken 'young, to appear' as in Middle Irish cano, cana 'wolf cub', Welsh cenau 'wolf cub, puppy; rascal'.

It hasn't become communis opinio - if you look at some up-to-date reference works, de Vaan in his "Etymological Dictionary of Latin and the other Italic Languages" (s.v. canis) derives it from PIE *k'won-, without even mentioning the alternative etymology, Mallory & Adams (Oxford Introduction) also list it as a cognate of the general IE word for "dog", NIL as well puts it under *k'won- without mentioning the alternative etymology, all assuming some mixture of irregular developments and paradigm levelling in order to explain the unexpected outcome in Latin. E.g. de Vaan assumes a Proto-Italic paradigm *ko:, *kwanem, *kunos, with the Nom. and the zero-grade oblique cases providing the Initial /k/ of Latin, and the Acc. providing the vocalism.
To me, all these contortions look a bit suspect, and I favour the alternative etymology that you quote.

[vokzhen]

As for finding similar correspondences in a hypothetical Pan-American macrofamily, I wouldn't really expect there to be many. If every Amerindian language outside of Eskimo-Aleut and Na-Dene shares a common ancestor, one would presume that said common ancestor started to diverge somewhere between 13,000 and 16,000 years ago (assuming that certain older dates are discounted). With the exception of the paleolithic continuity hypothesis, none of the current theories about the Proto-Indo-European urheimat posit that the language was spoken before about 8,000 years before present, and most Indo-Europeanists posit that it can't be much older than 6,000 BP based on it having vocabulary related to wheeled vehicles.

Basically, reconstructing "Proto-Amerind" would be the equivalent of reconstructing the language that was to Proto-Indo-European what Proto-Indo-European is to modern English.

Perhaps not so far from Indo-Uralic. Remember, what survives for millennia is resilient. Proto-Amerind should also have the advantage of multiple long-independent stocks, each capable of bearing independent witness to the common features.

Really quite far from Indo-Uralic. Indo-European languages are related at a time depth of ~6000 years (<5000 years for what we have the most substantial evidence for, i.e. Anatolian's clearly related but we at so far haven't been able to reconstruction a genuine PIE verbal system, just a post-Anatolian one). Indo-Uralic is based most strongly on a number of grammatical words (pronouns), morphemes (agreement markers) and quirks (fused case-number suffixes) at a time depth of ~8000 years afaict; almost all lexical data is controversial. American settlement, on the other hand, has a time depth of >13k years, which is approximately the same time depth as Nostratic and Dene-Caucasian, and depending on what theory you're working with (e.g. the first migrants were a refuge populations isolated for ~20k years before that) possibly much, much older.

The only two "pan-Americanisms" I've really heard of are a) n-m pronouns, which are above what you'd expect but still by no means common, and b) a propensity for polysynthesis. But that's very likely an areal feature rather than genetic, and from what little I know of it it seems like it would you useful to distinguish different strains of polysynthesis: Mayan, Algonquin, and Wakashan are all uncontroversially "polysynethic" but when you look at the details of how they work it's very different from one another, likely with different sources of grammaticalization and at different time depths.

[Zju]

The only two "pan-Americanisms" I've really heard of are a) n-m pronouns, which are above what you'd expect but still by no means common..

How reasonable it is to think that most languages with M-T and N-M pronouns each form a macro-family? Can chance resemblance explain the similarity in 1. and 2. pronouns across all those languages?

[linguoboy]

The only two "pan-Americanisms" I've really heard of are a) n-m pronouns, which are above what you'd expect but still by no means common..

How reasonable it is to think that most languages with M-T and N-M pronouns each form a macro-family? Can chance resemblance explain the similarity in 1. and 2. pronouns across all those languages?

At least some of them, given that Siouan is M-T rather than N-M despite having originated in America rather than Eurasia. (It's also polypersonal but not polysynthetic.)

[zompist]

Remember, what survives for millennia is resilient.

Such as what? Anything in language can be distorted by sound change. Endings and particles are often ancient, but they are also subject to innovation and replacement.

[Richard W]

Remember, what survives for millennia is resilient.

Such as what? Anything in language can be distorted by sound change. Endings and particles are often ancient, but they are also subject to innovation and replacement.

You'll have seen the constant, uniform rate glottochronological argument for a time depth limit for recovery. That limit is easy to underestimate, because some words, notably pronouns, usually get replaced far more slowly than other words. Of course, you can get situations where pronouns suddenly become unstable, as in Portuguese.

Simple gender systems hold up well, though marker perhaps not as well. Most Indo-European languages preserve a masculine-feminine distinction, even though a few branches have lost it. (OK, perhaps this isn't quite PIE - possibly just from Proto-non-Anatolian.) The main gender marker has held up well in Europe. The same pattern goes for Afroasiatic, with widespread survival of the /t/ gender marker.

[Richard W]

The only two "pan-Americanisms" I've really heard of are a) n-m pronouns, which are above what you'd expect but still by no means common, and b) a propensity for polysynthesis.

A specific lexical pan-americanism is "a word like _ma_ meaning 'hand'". From his diatribe againstreview of "Language in the Americas", Lyle Campbell seems to suggest that the form is rather _maki_.

Lyle Campbell's dismissal of a connection proposed by Haas dismissed 23 of the proposed cognates as 'pan-Americanisms'. 23!

[linguoboy]

The only two "pan-Americanisms" I've really heard of are a) n-m pronouns, which are above what you'd expect but still by no means common, and b) a propensity for polysynthesis.

A specific lexical pan-americanism is "a word like _ma_ meaning 'hand'".

You mean like Lakota nape or Osage šáake?

(I don't know where Siouan is really from, because it seems based on these criteria it can't be America.)

[vokzhen]

The only two "pan-Americanisms" I've really heard of are a) n-m pronouns, which are above what you'd expect but still by no means common..

How reasonable it is to think that most languages with M-T and N-M pronouns each form a macro-family? Can chance resemblance explain the similarity in 1. and 2. pronouns across all those languages?

At least some of them, given that Siouan is M-T rather than N-M despite having originated in America rather than Eurasia. (It's also polypersonal but not polysynthetic.)

Yea. While certainly not as strong as the connection of mama, nana, baba, and dada to parents and food, there's probably quite a bit going on with things like m-t and n-m pronouns just because they're some of the easiest, first sounds that can be pronounced. There are certainly some due to genetic relationship, Indo-European and Uralic for example, but positing genetic relationships based purely on pronominal systems, without other grammatical support, seems very premature.

And I meant to imply that even the "pan-Americanisms" I've heard aren't particularly pan-American. N-M pronouns are a minority, and I'm not sure even a significant one, but a tantalizing one. Level of synthesis seems to be as much an areal feature as a genetic one.

[zompist]

Remember, what survives for millennia is resilient.

Such as what? Anything in language can be distorted by sound change. Endings and particles are often ancient, but they are also subject to innovation and replacement.

You'll have seen the constant, uniform rate glottochronological argument for a time depth limit for recovery.

So far as I know glottochronology is not taken seriously any more; it doesn't even give good results for IE. See the discussion in Trask's Historical Linguistics, for example.

That limit is easy to underestimate, because some words, notably pronouns, usually get replaced far more slowly than other words. Of course, you can get situations where pronouns suddenly become unstable, as in Portuguese.

Third person pronouns seem to be regularly innovated; they can't even be reconstructed for PIE (according to Lehmann at least).

It's hard to look at the mess many European language have made of the second person without being skeptical of how long-lasting it is. Here we obviously have PIE forms, but V forms are often replaced, and the T forms were replaced in English and Brazilian Portuguese. In Mandarin we have the very weird *njidx > er. And these changes are all from the last 1500 years or so, 1/10 of the time depth of the Americas.

The first person does seem to more resistant, though we only have to look over to French for a(n ongoing) replacement in the plural, and to Spanish for some unexpected accretions.

As so often, the fact that we have 3000 years of texts for PIE makes reconstruction much more solid. If we had only a dozen IE languages, and all spoken only-- a situation typical of many Amerindian families-- the pronouns might not be so easily reconstructed.

(I started to collect pronouns along with numbers but was never very consistent. Still, just a glance at my bookshelf gives Nishnaabemwin gii for 2s, and Quechua ñuqa for 1s, which don't fit the N-M thing.)

(Also, for 'hand', Jaqaru has 'ampara'.)

Simple gender systems hold up well, though marker perhaps not as well. Most Indo-European languages preserve a masculine-feminine distinction, even though a few branches have lost it. (OK, perhaps this isn't quite PIE - possibly just from Proto-non-Anatolian.) The main gender marker has held up well in Europe. The same pattern goes for Afroasiatic, with widespread survival of the /t/ gender marker.

This probably isn't the best example, as the typical m/f/n IE system is innovated; thus it's an example not of how resilient gender systems are, but how they can be completely innovated (and destroyed) in a 5000 year period.