So here we go.

The tail does not wag the dog

The primary finding of the Pagel et al. study is this: When comparing lexicons from different languages, meanings that shared a common word in an ancestral language change over time more slowly if the word in question is used more often in day to day speech. This finding was found to be consistent enough that the authors call this a “law-like” property of language.
́

Greek speakers say ”oura”, Germans ”schwanz” and the French ”queue” to describe what English speakers call a ‘tail’, but all of these languages use a related form of ‘two’ to describe the number after one.

You can do this yourself. Here is the English “horse” translated into two closely related and one more distantly related Indo European languages:

Dutch: paard
German: Pferd
French: cheval

Not a lot of overlap, though a linguist would see the Dutch and German as similar, I suspect. Here, in contrast, is the English word “hand”

Dutch: hand
German: hand
French: main

The three Germanic languages are identical, and maybe that French word is not so different. Now let’s try for some more anatomy, with the English word “penis.”

Dutch: penis
German: penis
French: penis

Wow. According the purported law like properties of language change … oh never mind, no way to draw any hard and fast conclusions at this point I suspect. (I’ve left off the accents and the pronunciations are more different than they look here.)

(Above results all obtained using Google Translate.)

Pagel et al. estimated the rates of change among vocabulary words for 200 different meanings across 87 Indo-European languages. The number of different cognates (words that are linguistically the same) ranged from one to 46. From this analysis they calculated that the half life of a word, on average, was probably a bit over five thousand years, with a very skewed distribution.

Our findings, based on a sample of fundamental vocabulary items, identify a general mechanism of linguistic evolution, which is expected to operate across all languages and timescales and makes predictions about rates associated with specific meanings. To the extent that the structure and everyday functions of human verbal communication mean that some words will tend to be used more frequently in all languages, we expect these words to evolve slowly, and vice versa for infrequently used words. Combined with parts of speech, this simple factor allows us to account for about 50% of the variance in rates of lexical replacement throughout the 6,000- to 10,000-year history of Indo-European languages. Given the many social, cultural and cognitive factors that can influence language, it is striking that word-use frequency alone can explain such a large proportion of the historical variation in rates of evolution. The generality of this influence is suggested in the finding that estimates of the rate of lexical replacement in Indo-European languages are correlated with rate estimates in Bantu10, Cushitic and Malayo- Polynesian.

A Tale of Two Disciplines

This research is partly based on, and partly demonstrates the validity of, the assumption that language change over time can be modeled as a tree-like pattern, much like genetic change over time is modeled to create species (or population) trees. (I hasten to add: I will be using terminology here that may annoy hard core cladists. I love annoying hard core cladists.) However, linguists have come to believe in recent decades that such research, beyond relatively simplistic grouping of very closely related languages that have diverged recently, is not worthwhile. Most linguists active today simply believe that the idea that time-deep language phylogenies can be built with any degree of reliability is utterly discredited.

The work by Pagel et al. seems to prove these linguists wrong, but the culture of incredulity is strong and seemingly unshakable. But I’d like to ask you to imagine what it might be like if things were just a little different in recent history.

( … harp music as everything becomes blurry, and the scene changes to a 1960s era lab with the large and furry figure of Charles Sibley holding four liquid filled test tubes in one hand, up to the light, gazing at them… Nearby, Jon Ahlquist is re-ordering a series of IBM punch cards that just got scrambled when they fell out of the box on the way back from the Batch Window at the computer center … )

“This is never going to work,” says Sibley. “This whole idea of using DNA to make a family tree of living species has too many problems. True, we came up with a number of plausible phylogenies, but the quick work of our colleagues in the fields of biogeography and morphology sure made quick work of our quick work!”

“I wish you would stop with the stupid puns,” intoned Ahlquist. “But as usual you’re right. This hybridization stuff kinda works but the results are not sufficiently resolved to sort out either really closely related species or very distant relationships. As for this in between scale of relationships, we can SEE those. We don’t need this extra expense. What are you doing with those colored liquids in those test tubes, anyway?”

“New martini. I call it “The Sarich,” replied Sibley.

“Meh.”

( …. scene becomes blurry again, with harp music, and refocuses on a group of graduate students and a junior prof type sitting around a table in Nick’s Beef and Brew on Mass Avenue in Cambridge. These researchers are attached to Harvard’s Phylolinguistic Research Center, a new facility just built on the foundation of the recently torn down Peabody Museum…)

“So what if they don’t think it works!” said the one named Merritt. “We’ve been using Pagel’s phylogenetic method on languages for decades, and no one has questioned our ability to make deep phylogenies going back more than half way to the origin of human speech! All we’re trying to do here is to apply the same exact methods to the phylogeny of the mammals, using genes instead of words. Of course it will work!”

The group was interrupted as the waiter, Irv, came by with a large tray and efficaciously dealt out a half dozen Double Cheeseburger Specials as though they were mere playing cards. “Which one of you gets double tops….” he said as he glanced around. Then he noticed Big Tim, and remembered ….. right, double tops…. “Here you go. Enjoy.”

After a few minutes of passing around of the ketchup and adjusting the French fries, the conversation resumed. Just then, the door opened and in came Mark, the group’s statistician. Whenever the door opens in this place, a mighty wind blows across all the tables in the general direction of cook’s grill, where a 93,000 BTU open flame is constantly in use making more and more hamburgers, converting several cubic meters of oxygen into oxidized beef per minute.

(One day, a few years after this conversation, it just happened to occur that no one went in or out of Nick’s for a full hour and ten minutes. All of the oxygen was burned up at the grill and the entire retinue of diners, employees, and Nick himself suffocated, in what would later become known as the Great Snuffing Out on Mass Ave.” But I digress…..)

Pagel sat down with the group and they started to talk again about the application of proven phylolinguistic methods to genetics.

“The problem with genetics,” someone said, “is that the are under selection, unlike words.”

“Another problem,” someone else said, “is that we’re looking at genetic change across vastly different animals, with different metabolic rates and generation times.”

“… and in some cases” someone else jumped in, “Different systems of reproduction…”

“… right, and not even the same number of chromosomes across species, so linkage effects may be different….”

“Don’t worry.” Pagel spoke those words and took a bite of his meal. “Oh, did someone order beer by the way?”

Someone handed Pagel a beer to wash down his cheeseburger.

“Cheers,” Pagel said, “You don’t have to worry about most of that stuff. Most genes are highly conserved across organisms. The plurality, anyway. And other bits of DNA seem to change fairly quickly. You couldn’t find a better system than genetics to try the phylogenetic methods on. It will work better than with language, and it works pretty well with language.”

“Why didn’t they … the biologists … why didn’t they, I mean, shouldn’t they have… um, how come…” sputtered the one called Greg, just starting on his second cheeseburger and not quite sure if he was ready to speak up yet.

“Why don’t they get it? Why did they give up on this sort of thing fourty years ago?” Pagel clarified. “Because their first few attempts used a technique that sucks, and because they had no idea how the numbers worked statistically. Now, with genetic sequencing we have excellent data, and we understand the numbers. This will be easy. You guys go collect the data and bring it back here. I’ll run it on my Android and we’ll have he paper out by dinner time…”

( … scene goes blurry, ad all six of the scholars crowded into the high-backed wooden booth in Nicks simultaneously chomp on the last bit of their cheeseburgers …. )

Well, I doubt it would have happened quite that way, but my point should be clear. Linguists gave up the ghost on phylogentics when they ran into a number of problems. The method became “discredited” and no further work has been done with it. Meanwhile, in another discipline in which this sort of method can be used (genetics, in the real world) the approach continued to be developed. And now, practitioners of this method will be happy to apply these ideas to language, and teach the old boys a thing or two.

(Clarifications: 1) In “real life” the “phylogenetic method” was invented by Pagel and Harvey, but this is not the method being used to do language phylogenies. It is a wholly different thing. 2) No one ever really died of suffocation in Nick’s. 3) Irv would not have been that good of a waiter.)

Mark Pagel, Quentin D. Atkinson, Andrew Meade (2007). Frequency of word-use predicts rates of lexical evolution throughout Indo-European history Nature, 449 (7163), 717-720 DOI: 10.1038/nature06176

It’s like an elephant and a mammoth meeting up in the Twilight Zone. Close enough to know there is a similarity, yet different enough to be a bit freaky.

This is from the work of Mark Pagel, of Reading (England) and his team. And it isn’t quite as simple as I’ve characterized it above. As Pagel told me in a recent interview, “… when I say ‘I’ or ‘two’ are very old, I mean that they derive from cognate (homologous) sounds . Every speaker of every Indo European language uses a homologous form of ‘two’ such as ‘dos,’ ‘due,’ ‘dou,’ ‘do,’ etc. It is an amazing thought because there are billions of Indo European speakers and hundreds of thousands of ‘language-years’ of speaking across all the unique branches of the phylogeny of these languages. In all that time ‘two’ has remained cognate. Cognate does not mean identical … it is a bit like my hand being homologous but not identical to that of a gorilla.”

Pagel acknowledges that may linguists are ‘upset’ with the assertion that there are numerous cognates that share a common ancestor …. which is also a cognate … that must be over 10,000 years old. But he indicates that this dislike for the proposed reconstruction is more of a misunderstanding of this concept of homology than anything else.

Indeed, most linguists reject the idea of even being able to begin to think about maybe planning in the most preliminary way to even maybe consider doing something like what Pagel and his team have done. And these days, the main reason that linguists give for not being able to reconstruct either individual words or linkages between languages and language groups is something like “… You can’t do that because it is long discredited.” But in fact, this is alchemy. Most modern linguists, in my experience, can not provide an actual coherent reason for this discreditation.

Linguists long ago rejected the very methods that were used in the old days (back when linguists thought they could and should reconstruct language phylogenies). There are almost no living linguists trained in this area. The previous generation, which did engage in this activity, were using methods that at the time were cutting edge but today are outdated. So, Pagel is using updated methods for working with words in a similar way that we work with genes, and getting results that are statistically valid.

As with a genetic study, the reconstructed phylogeny is complex. There are meaning-sound links that go back to a certain time period, but not before, because of a change at that node. There are some that are perhaps 40,000 years old (based on an estimate of cultural divergence, which in turn becomes less certain as one goes farther back in time) and others that are only a few thousand years old. As has been demonstrated in other research projects, words that are used frequently are more likely to stay relatively unchanged than are rarely used words. Also, according to Pagel, nouns change more slowly than verbs, and verbs more slowly than adjectives.

So, the phrase “colorless green ideas sleep furiously” uttered in the far distant future might be “blifnork orgonst idears sloop firooslnitch.” According to me, not Pagel. (Pagel refused to comment on that question.)

But seriously, I’m glad to see the linguistic phylogeny challenge taken up again, despite the naysayers, and I’m especially glad that Pagel is doing it because he’s got the methodologies necessary to make this work.

Background.

Although many animals communicate vocally, no extant creature rivals modern humans in language ability. Therefore, knowing when and under what evolutionary pressures our capacity for language evolved is of great interest. Here, we find that our closest extinct relatives, the Neandertals, share with modern humans two evolutionary changes in FOXP2, a gene that has been implicated in the development of speech and language. We furthermore find that in Neandertals, these changes lie on the common modern human haplotype, which previously was shown to have been subject to a selective sweep. These results suggest that these genetic changes and the selective sweep predate the common ancestor (which existed about 300,000-400,000 years ago) of modern human and Neandertal populations. This is in contrast to more recent age estimates of the selective sweep based on extant human diversity data. Thus, these results illustrate the usefulness of retrieving direct genetic information from ancient remains for understanding recent human evolution.

The authors actually get more specific regarding the role of FOXP2 in language:

Although language and speech are clearly genetically complex phenomena, the only gene currently known that has a specific role in the development of language and speech is FOXP2. The inactivation of one FOXP2 copy leads primarily to deficits in orofacial movements and linguistic processing similar to those in individuals with adult-onset Broca’s aphasia

While the paper by Krause et al is an important contribution because it involves allele-level comparison of nucleic genetic material between hominid groups and across living and extinct forms, the role of FOXP2 and the characterization of the genetics of language may be misleading, if not simply very very wrong.The paper describes how nuclear DNA was recovered from Neanderthal material, and demonstrates that the FOXP2 variant found in modern humans is also present in this ancient material. The extraction process is very carefully done, there are piles of controls and checks for contamination and there is no particular reason to believe that the results are not real. However, these results are so sensitive from an interpretive perspective, and recovery of ancient DNA is so new, it is necessary to replicate this work with additional recovery attempts.The authors conclude:

… the current results show that the Neandertals carried a FOXP2 protein that was identical to that of present-day humans in the only two positions that differ between human and chimpanzee. Leaving out the unlikely scenario of gene flow, this establishes that these changes were present in the common ancestor of modern humans and Neandertals. The date of the emergence of these genetic changes therefore must be older than that estimated with only extant human diversity data, thus demonstrating the utility of direct evidence from Neandertal DNA sequences for understanding recent modern human evolution. Whatever function the two amino acid substitutions might have for human language ability, it was present not only in modern humans but also in late Neandertals. Ongoing in vivo and in vitro experiments should help to delineate these functions.

The news that Neanderthals have the same FOXP2 gene as modern humans, indicating that they may, therefore, have had modern human speech and language, has been misinterpreted in my view. I’d like to make the following points:1) It is not true that this evidence can be used to draw this conclusion.2) The FOXP2 gene is not a gene for speech and language.3) Science reporting has always used framing, this is an example of framing, and this is an example of framing that sucks. This may not mean that all framing sucks, but it does demonstrate how framing can suck.The role often attributed to the FOXP2 Gene in both the scientific literature and the science press can be summarized in this set of phrases from a recent report in the New York Times:

“Neanderthals, an archaic human species that dominated Europe until the arrival of modern humans some 45,000 years ago, possessed a critical gene known to underlie speech … a gene called FOXP2 which is associated with language … The human version of the gene differs at two critical points from the chimpanzee version, suggesting that these two changes have something to do with the fact that people can speak and chimps cannot.”NYT:”Neanderthals May Have Had Gene for Speech”

So clearly, it is on the table … in the nation’s premier daily science news print outlet, that FOXP2 is thought (but not with total certainty, of course) to be a speech and language gene. Is it necessary for me to demonstrate that lesser news outlets make this conclusion seem even firmer? No, but it would be fun. The following are phrases from such sources, not attributed here because I don’t want to embarrass anybody.

• Modern speech gene found in Neanderthals
• Neanderthals had same version of FOXP2 “language gene” as modern humans
• It’s the only gene known so far that plays a key role in language. When mutated, the gene primarily affects language without affecting other abilities.
• Neandertals, humans share key changes to ‘language gene’

Let’s see, that’s one blog, two major news outlets, and the journal Nature. I’ll let you guess which source uttered which phrase.But what is the FOXP2 gene really?

Language is a uniquely human trait likely to have been a prerequisite for the development of human culture. The ability to develop articulate speech relies on capabilities, such as fine control of the larynx and mouth, that are absent in chimpanzees and other great apes. FOXP2 is the first gene relevant to the human ability to develop language. A point mutation in FOXP2 co-segregates with a disorder in a family in which half of the members have severe articulation difficulties accompanied by linguistic and grammatical impairment. This gene is disrupted by translocation in an unrelated individual who has a similar disorder. Thus, two functional copies of FOXP2 seem to be required for acquisition of normal spoken language. We sequenced the complementary DNAs that encode the FOXP2 protein in the chimpanzee, gorilla, orang-utan, rhesus macaque and mouse, and compared them with the human cDNA. We also investigated intraspecific variation of the human FOXP2 gene. Here we show that human FOXP2 contains changes in amino-acid coding and a pattern of nucleotide polymorphism, which strongly suggest that this gene has been the target of selection during recent human evolution. Enard W, Przeworski M, Fisher SE, Lai CS, Wiebe V, Kitano T, Monaco AP, Pääbo S.. (2002). Nature. Aug 22;418(6900):869-72. Epub 2002 Aug 14.

It is arguably true that language is a prerequisite for the phenomenon of “human culture” in the sense of social ontogeny, but not individual ontogeny and not as the chicken as in “chicken before the egg” (or the reverse if you prefer) to human culture as egg. In other words, language does not precede human culture in evolutionary time, but rather, language is part of human culture. The neural differences between a hypothetical pre-cultural human (and that is very hypothetical) and a post-“dawn of culture” human would be those brain differences that we see as distinguishing between the non-cultural (arbitrarily defined as such for the present purposes but obviously a falsehood) chimpanzee-like ape ancestor and, say, you or me.So when we speak of human intelligence, we may be speaking about the FOX2P gene. Is this the gene that Klein refers to in his somewhat insightful but mostly misguided “one gene theory” ? (Yes, he would likely say, … or at least “Yea, this really could be the gene.”)OK, that was the first sentence of the FOX2P abstract of Enard et.al. Onward.We have now switched topics, from language to speech. This happened in the reverse direction in the verbiage cited from the more recent NYT piece on FOXP2 and Neanderthals. One of the great findings of the Language Research Boom of the latter half of the 20th century was this, now pretty much undisputed: Language operates independently of modality. Speech is no more language than getting to work is my car. Now, I quickly add that the authors of the above cited texts do not explicitly tell us that they are conflating speech and language. They are doing it implicitly and tactfully, to avoid the obvious error. But they need to make this leap … and it is a leap across a fairly narrow but very, very deep chasm … in order to link the physical (genetic) trait that they have in hand with the grand concept of that one feature that makes us human.Then, we learn about a broken family. It is not clearly articulated in the paper cited here, but this information came out in later reports as well as under friendly interrogation at conference venues by Terrence Deacon and others: The FOXP2 family — the grammar-error correlated with a base pair mutation family — presents a wide range of abnormalities. If FOXP2 caused the consistent grammatical goofiness than it also caused these other attributes.Many traits work that way. There is a single, widespread phenotype and a smaller number of “mutants” each the expression of a simple mutation that causes some protein to either not be functional at all or to fail in function in a number of contexts. Quite often such individuals are not so dysfunctional that they can’t survive, but there is likely overall selection against the variant most of the time. So, each time such a mutation arises, it makes the turn towards that old exit door we call “fixation” and pretty soon everyone in that “deme” is reduced to the more typical and widespread variant.Sometimes this tells us about a gene that relates to a particular system. Many examples that come to mind have to do with pigment. This is because we humans, as primates, fixate on the visual so we “get” pigment, we see pigmentation. Where I live there are many pockets of all-white squirrels (always in urban areas) and a few pockets of nearly all black squirrels (they can be anywhere) and mostly gray squirrels. If I went out and sampled the squirrels, I could probably put together a pretty good story of pigmentation variation involving a single gene or maybe two for the grey-very dark grey variants (I’m thinking one gene because it does not look like a smooth range to me) and maybe a set of common mutations in any one of a handful of genes that produces the white variety. I would test the hypothesis that the white squirrels have lower individual fitness owing to problems other than pigment loss that arise from this mutant allele, as well as, perhaps, being white all summer (instead of doing what other critters around here do, turning white in winter and back to brownish in summer). And so on.When I’m done with this research, I’d have a story of metabolic process mainly but not entirely having to do with pigmentation. In fact, I’d have small set of metabolic pathways sharing their “roots” with many other metabolic pathways, but their “branches” largely having to do only with pigmentation. So alleles that “block” the pathway later on cause visible changes in pigmentation, and other secondary changes that also have to do with pigmentation (negative effects on eyesight, increased chance of skin cancer?, being visible to the predators, etc.). Alleles that block the pathway early on also block other pathways, so the direct genetic effects analogous to the simple lack of production of a pigment also occur in other pathways, some of which may be developmental. So you end up not with a white squirrel that may be more likely eaten by a hawk, but rather with a half formed embryo.The FOXP2 mutation seems to be a non-deadly allele (but we can’t be sure of that … there may be prenatal synergistic effects. What is the rate of miscarriage in FOXP2 women? We don’t know) causing a broad-spectrum mutation that has, among many other things, a negative effect on neural development, which is manifest in several ways, one of which happens to be related to linguistic production.The affected individuals have a range of cognitive problems, but they are all very minor, but one happens to be dyspraxia and thus of great interest. These individuals have difficulty articulating speech and performing certain semi- or fully-automated tasks that are centered in Broca’s Area and one or two other spots. FOXP2 in mammals is involved in development of a number of organs, including brains and lungs and esophagus in mice and neural systems related to echolocation in bats. This does not mean it is not involved in lung formation in bats … that experiment has not been done. In birds, the same gene is involved at least in neural development focusing on the avian cortex, but we don’t know about its other functions. The same gene is found in exemplars of most if not all of the vertebrates.The reason why the gene works this way is because FOXP2 is, you will not be surprised to know, one of the FOX genes, a set of transcription factors. These are genes that turn on during development and ultimately affect the transcription of other genes that are part of that developmental process. An error in a FOX coded transcription factor is likely to have broad effects, most of which will be deleterious.There are genes that vary a lot (have a lot of alleles) and this makes sense. In some systems, variation is associated with positive fitness (like the MHC genes). Other genes vary very little, and this is generally in the systems where variation is almost always bad. This is probably the case with FOX genes in general, and FOXP2 in particular.Damage Brains, Damaged TheoriesMuch of the early understanding of how the human brain works was based on the study of damaged patients. Strokes or trauma would cause an aphasia or other behavioral effect. You study the effect until the patient dies, then you look at the brain to see which part was damaged, and eventually, you get a model of the distribution of functional areas within the brain itself.Subsequent studies from patients in surgery and MRI’s or similar technology indicate that while there is a rough correspondence to the earlier ideas of brain function, things are in fact very different than first expected. There are more areas involved in any given function, with individual areas often having antagonistic subdivisions. So damage to a particular area of the brain, at the gross level, could have very opposite effects each time it happens to a different individual. Naturally, many of the possible kinds of damage that might have revealed something about the function of the brain could not do so because damage to those areas typically resulted in death. For these reasons, the damage-based models were very under determined and very inaccurate.Think of it this way: Say you knew absolutely nothing about how a car worked inside, but you wanted to find out so you could at least describe how a car works (accurately) and maybe eventually build one from scratch. So you go out and get a bunch of cars You never open the hood, never look at the guts of the car. Instead, you take a shot gun and fire the shotgun at one of the cars and see how it affects performance. Record your results. Fire the shotgun at this car and keep testing the effects, recording them, until the car utterly stops working. If you happen to “kill” the car on the first shot, then you won’t get much data from that car.Now move on the the next car and do the same thing. And so on.You will eventually get a model of how the car works but it will be pretty coarse, and you will probably have a lot of misconceptions that make your model useless. Especially if you had a mixture of engine-in-front and engine-in-back cars (as would the the case of human brains, studying a mixture of males and females, French-speaking vs. Chinese-speaking, etc. .. as they would have differences in the distribution of linguistic and other functions owing to various hormonal and social effects while growing up).The linkage between FOXP2 and language is like this. A broken FOXP2 gene is probably almost always fatal early in development. The few alleles that result in a human growing to maturity are going to have strange effects that will not really make much sense. Taking these effects to the next level … to attributing prior allelic evolution in this particular gene the role of evolutionary foundation for language … is misleading and inappropriate.There are probably hundreds of genes like FOXP2 that have to be working a certain way or language function will be impaired. If they are developmental genes, there will be very few alleles. Most alternate alleles will result in inviable embryos or otherwise catastrophic effects that ultimately provide no information to us.Most or all of these genes probably have a variant in chimpanzees that is also conservative in chimps, but owing to an accumulative 10 to 16 million years of evolutionary time, they will be variant. If you go looking in Neanderthals for these genes, you will almost always find the human variant rather than the chimpanzee variant. This pattern is exactly what we would expect under a basic neutral model, the most likely to apply in this case.In the broad picture, FOXP2 is very interesting. In the context in which it is usually discussed these days (as the “language gene”) it is little more than annoying._______________________________Krause, Johannes, Carles Lalueza-Fox, Ludovic Orlando, Wolfgang Enard, Richard E. Green, Hernán A. Burbano, Jean-Jacques Hublin, Catherine Hänni, Javier Fortea, Marco de la Rasilla, Jaume Bertranpetit6, Antonio Rosas and Svante Pääbo. (2007) The Derived FOXP2 Variant of Modern Humans Was Shared with Neandertals. Current Biology. doi:10.1016/j.cub.2007.10.008