Proto-World and the Language Instinct
By Mark Rosenfelder
The question has been reopened recently, on several fronts:
Not all these claims are accepted; but many linguists feel confident that there is a language instinct (whether or not it’s what Chomsky claims to be describing) and that there was a single ancestral language (whether or not it’s what Ruhlen and Greenberg claim to be reconstructing).
I will argue below that both propositions are supported by scant evidence, and that the contrary propositions-- that there is no language instinct, and that there was no single ancestral language-- are as well or better supported.
The usual case for this proposition is, I believe, as follows:
Point 1 is based on pure speculation, for which no evidence is offered whatsoever. No descriptions of differing "language organs" are ever offered; we are never given an example, much less a proof, of how a language based on one language organ could not be learned by someone with a different kind.
More importantly, all three of the points assume that language developed in just one step and in one population. Here, for instance, is Bickerton on the one-step hypothesis:
The evidence surveyed above indicates that language could not have developed gradually out of protolanguage, and it suggests that no intermediate form exists. If this is so, then syntax must have emerged in one piece, at one time-- the most likely cause being some kind of mutation that affected the organization of the brain. Since mutations are due to chance, and beneficial ones are rare, it is implausible to hypothesize more than one such mutation. (p. 190)
I sometimes parody the Proto-World view as "The language organ evolved on Tuesday; language was invented on Wednesday; and everyone else in the world was eliminated on Thursday." This neatly summarizes the improbable assumptions of the theory.
As evidence, some point to the observed historical development of sign languages by the deaf; the usual cited case is Nicaraguan Sign Language, developed in two stages after the first schools for the deaf were established in 1979 (Pinker, p. 36f). (Of course, isn’t the development of a language from scratch evidence against monogenesis?)
Such stories are intended to help us picture the same thing happening 100,000 years ago, when the language organ developed. But there is an important difference between the Nicaraguan children and the first mutants with a language organ: the Nicaraguans were born into a population where everyone already had the mutation. The sign language could thus be developed cooperatively.
Who would the first mutant talk to? It wouldn’t do him or her much good to refine the pidgin of his community into a true language if there were no one else around who could understand it.
It’s commonly thought that children absorb languages effortlessly; but language acquisition studies make it clear that this is not so (cf. François Grosjean’s Life with Two Languages, 1984). Language learning takes effort even for children, and children seek to minimize that effort by learning only those languages they absolutely must, in order to communicate. For instance, once children of immigrants discover that their parents speak the national language, they may speak it with them, and retain only passive knowledge of the parents’ native language. Full language ability is likely to continue only when there are people in the child’s life that cannot understand the national language.
The Nicaraguan children had other deaf children to sign with. The primeval mutant did not, and is not likely to have continued on the unrewarding path of language development, any more than deaf children reared in isolation by hearing parents develop full sign languages.
Perhaps the original mutant didn't develop language, but his or her children did. Under this scenario, the siblings provide enough stimulation to each other. Now, we're rather out on a limb here. Siblings in hunter/gatherer populations tend to be widely spaced out (four years or more); the oldest child will be 5 or 6 by the time the next is doing more than babbling -- and mutations are not always passed along to all of the children in a family. And the other children in the tribe don't share the mutation. This is hardly the boisterous language-developing environment that we found in the Nicaraguan school for the deaf.
(How about twins? Possible; but twins only comprise 1/100 of births... and twins are severely disadvantageous in a hunter/gatherer environment.)
But fine, let's suppose that language develops. We have not yet shown that monogenesis would follow; we only have a few children with a language organ and a language, not an entire species. For the whole species to have true language, the mutation must spread throughout the population, by making its bearers reproduce more successfully than non-bearers.
At this point we usually hear some handwaving about the obvious benefits of language. But we must distinguish language from the language organ. The language organ doesn’t offer any evolutionary benefits in itself; only speaking a language does. And speaking a language requires other speakers to communicate with.
Are our siblings more likely to reproduce than other humans? This seems highly questionable to me. They have enhanced communication skills, but only with each other; I see no reason why this would make them more attractive as mates. It might as easily make them less attractive. People who are standoffish, different from everyone else, and mutter incomprehensibly might easily be feared and shunned.
Perhaps the siblings will marry each other and found the modern race of homo fabulans? This seems unlikely; incest is generally covered by strong taboos, and in fact most hunter/gatherer societies have elaborate rules on who can marry, with the general result that intermarriage between bands is common. The twins would very likely end up in different tribes, or if not, their children will.
The mutation is not likely to stay in one family or band; within a few generations it will spread widely to other families and bands, through intermarriage-- and other eventualities; there are plenty of ways a carrier of the mutation could be separated off, allowing his or her children to develop new languages. (Did language-using men never have sex with a woman from another tribe?) The mutation is of little use if the child has no one to communicate with; but isolated bearers can still pass the mutation on to their children. (We assume that the next generation inherits the mutation; we’re not interested in the cases where it did not.)
Eventually the mutation will be widespread enough that marriages between bearers become possible, and the children can turn the local pidgin into a language. But now we’ve lost monogenesis. The scattered mutants, living in various small bands, cannot cooperate to generate a single Proto-World; they will each develop the differing local pidgins into a language.
We arrive at the same conclusion, by the way, if we forget the siblings and simply suppose that the mutation benignly spreads until it’s common enough to trigger language creation. Once it’s common enough to produce language-generating siblings, those siblings will exist in scattered locations, generating different languages.
This theory clears one hurdle, in that if the population was tiny enough, the prohibition on incest would no longer hold-- there may not be any potential mates besides close relatives. The mutation could thus spread faster; and the small population would make it likely as well that its bearers would encounter each other and cooperate in generating a single Proto-World.
Still, the linked chain of assumptions is once again getting tottery. Is it likely that the fully formed language organ appears just at the moment when a genetic bottleneck appears, lifting the incest taboo? Perhaps we can invoke the anthropic principle, and state that if this didn’t happen we wouldn’t be here-- only creatures where this eventuality occurred can ever be in a position to wonder about it. However, while the story may be plausible, plausibility is not proof. The theory simply becomes one possible origin story, not knowledge about how language arose.
In addition, monogenesis depends crucially on the timing and conditions of the story. If the population grew and spread out quickly enough, different languages might still have developed. If a bearer of the mutation found their way to a non-bearing population, he or she (or rather their descendants) would potentially be the focus of another language-generation process. If a mutation-bearing band conquered a larger non-bearing group, children (including the inevitable hybrids) might grow up learning the conquered population’s pidgin, turning it into yet another new language.
This is not really support for monogenesis, however. The timing and worldwide scale of this expansion make it clear that we are not dealing with a single population speaking a single language. (Modern-looking skulls appear 100,000 years ago in Africa, and parts of the expansion, such as the colonization of Australia, occurred 50,000 years ago. That’s enough time to develop the Indo-European language family, with all its divergences from English to Russian to Hindi… five times over.)
There are still many questions about what happened during this expansion, and the evidence of complete extermination is not unequivocal (e.g. there is some evidence for hybridization). Furthermore, Neanderthals, who according to this story didn't have the language organ, buried their dead and used stone tools and fire-- and had brains larger than ours. This doesn’t seem compatible with an ape-level command of not-quite-language.
The hunter-gatherer mutants with the language organ are in the same situation. Scattered about a sea of non-mutants, they will develop new languages when they can grow up together, and won’t do so when they’re isolated.
The code is an impressive achievement, especially as performed by a creature with a brain a cubic millimeter in size. (Admittedly, that’s pretty large for an insect.) The code indicates the direction of a find relative to the sun, its distance, and its quality. In the European and Indian honeybees, the direction of the sun is indicated only indirectly, as an offset from vertical, as the bee dances on the vertically oriented honeycomb in the dark hive. Bees can dance for several hours, and the indicated direction is even updated as the sun’s position changes. And the dance can be used not only to indicate food sources, but sources of water, or possible new nesting sites. In the latter case, there is even a consensus mechanism, as the scouts investigate each of the sites nominated, and those who are dancing to indicate inferior sites switch to better ones.
Investigation of the dance in different honeybee species has led entomologists to reconstruct at least three steps in the development of this code:
An even earlier stage (or perhaps an independent development) is represented by the stingless bees, where the dancer does not indicate direction, but rather recruits other bees to follow her; in some species the recruitment involves a distance cue, though not one of direction.
Certain other insects also execute stereotyped circling or waggling movements after finding a food source, which correlate to its quality and distance.
(Descriptions here are summarized from Friedrich Barth’s Insects and Flowers: The Biology of a Partnership, 1985.)
What’s relevant here is that something relatively simple, as compared to human language, is nonetheless seen to have developed in at least five stages. The appearance of human language full-fledged in one step, or even Bickerton’s two steps, does not match this biological model, nor that of the development of other complex capabilities, such as vision or flight.
One argument for the language organ, for instance, is the poverty of stimulus argument-- that the language available to children is too little in quantity and too suspect in quality to serve to construct a grammar. Sampson points out that research into child language acquisition makes the ‘quality’ argument hard to sustain-- that in fact the evidence is that ‘Motherese’ is significantly more grammatical than conversation between adults (p. 39).
The case for low quantity is no better. One of Chomsky’s frequently repeated claims, for instance, is that sentences "rarely arise" which would teach children that the verb to be fronted in yes/no sentences must be the main verb, not the first verb (Everyone who is coming will raise their hands --> Will everyone who is coming raise their hands?, not *Is everyone who coming will raise their hands?; see also Pinker p. 40). But Geoffrey Pullum found such samples to form 12% of the yes/no sentences in a particular corpus (Sampson, p. 42). (A corpus of written English, admittedly; but at least it’s based on examining actual data; Chomsky’s claim so far as anyone knows is not.)
One of Pinker’s most convincing examples along these lines is the black Southern community that "do[es] not speak to their prelinguistic children at all" (p. 40). Sampson points out (p. 87), however, that Pinker has exaggerated the claims in his source, Shirley Brice Heath’s 1983 study Ways with Words. The children hear plenty of language, much of it addressed to them (e.g. by older children, or by adults including them in a group).
Another pillar of the language organ is the window of opportunity argument-- that language ability must develop in a relatively small span of years, or it will be highly deficient at best. This seems reasonable to many people; but there has never been much evidence for it, and what there is-- chiefly, children raised without language by madmen-- is equivocal; the difficulty of language acquisition can hardly be separated from the cognitive and emotional difficulties caused by the way the children were raised.
We do have a lot of evidence, of course, about second-language (2L) acquisition, and it is a commonplace that children learn languages easier than adults do. Backing up this commonplace assumption, however, turns out to be surprisingly difficult. In their review of the literature, Ellen Bialystok and Kenji Hakuta (In Other Words : The Science and Psychology of Second-Language Acquisition, 1995) find no good evidence that 2L acquisition becomes impossible at any given age. Children do better than adults in the sense that children tend to achieve fluency but not all adults do, but there is no level of fluency that seems to be denied to adults, and there is no sharp age cutoff. David Singleton in Language Acquisition: The Age Factor comes to about the same conclusion.
It’s worth contrasting the ‘language instinct’ with more incontrovertible biological abilities, such as the ability of newborn foals to walk within hours of birth. Children are certainly not born knowing any language; it takes them two or three years to acquire a basic proficiency, which is then refined over a period of more than a decade. And there is excellent evidence (see Grosjean again) that children find language difficult, and in multilingual situations avoid learning languages they don’t absolutely need..
Why do children learn languages fluently, and adults don’t always do so? The simplest answer is surely that children have several years of concentrated time to learn languages, and are more strongly motivated. Immigrants learn the national language too when they must-- and particularly when they must speak it at home as well as at work. But many immigrants don’t really have to learn the national language-- the required level of national language competence in their work is often minimal, and they may live in an enclave (a family or a neighborhood) where it is not needed at all. Children, by contrast, have no alternative language to fall back on (nor one that will distort their speech with an accent) and must learn language to understand and affect their world.
Finally there’s the argument from linguistic universals. Pressed for details, however, linguists tend to come up with things that are unexciting, vague, not very strict, or simply wrong. For instance, Chomsky suggests that there could be no word "similar to ‘limb’ except that it designates the single object consisting of a dog’s four legs". As a counter-example, Sampson suggests French rouage, which can be used to refer to (just) the wheels of a vehicle (p. 109); I would also suggest the baseball term battery, which refers to the catcher and pitcher. (And could we really not invent a term 'limb system' to refer to the dog's limbs?)
Implicational universals are interesting, but not much good as evidence for a universal cognitive constraint. For instance, modifiers tend to consistently precede the modified word-- or to consistently follow it. The fact that English is a glaring exception (e.g. we put adjectives before and relative clauses after a noun) surely casts doubt on this being a biological necessity. We don’t need to invoke biology to explain why a language could develop some internal consistency.
There’s no doubt that all languages show such structure. However, even if one doesn’t accept everything in Sampson’s critique, he more than adequately demonstrates that syntactic structure is not an argument for the language organ. Following an argument by Herbert Simon originally applied to economics, Sampson argues that any complex entity produced by an undirected, fitness-rewarding process will result in tree structuring-- simply because building things up out of smaller components is the most efficient way (short of directed design) of achieving that complexity.
Thus, plants and animals are built out of organs and ultimately out of cells; human societies are built out of smaller units (provinces, corporations, families); and language is built out of phrases.
(Simon's example relates to watchmaking. He contrasts two watchmakers, each of whom makes watches out of a hundred parts. One, however, makes the watches out of ten subassemblies of ten parts each; the other makes the watch out of the hundred parts with no intermediate structure. Other things being equal, they'll assemble a watch in the same time. But other things are not equal: the possibility of errors and interruptions ensure that the first watchmaker will be much more productive.)
An intelligent designer may create things differently: like the second watchmaker, she may create a vast unitary architecture which cannot be divided into semi-autonomous components. But even the designer is well advised to follow the tree model. Thus, machines are built out of subassemblies, computer programs are built out of object classes, and symphonies are built out of movements, musical phrases, and repeated motifs.
Simon’s argument might be supposed to support a biological origin for language’s tree structure; but it can equally well support a model of language acquisition Sampson calls Popperian, modelled on the way sciences develop-- making and testing hypotheses, rejecting ideas that don’t work, understanding one aspect of the field at a time, and building up a complete picture in stages.
Another approach to this point might be to try to seriously imagine an alternative to a tree-structured language. The usual Chomskyan example is that instead of inverting subject and verb (which may be complex components with their own subject and verb), we might form a question by fronting the first verb: Is everyone who coming will raise their hands?
But how could such a rule develop? Surely both in evolution and in language acquisition, simple sentences (Is everyone here?) preceded complex ones; an inversion rule must have first developed, and must first be acquired, applied to simple sentences. The very power of a tree structure is that rules and operations can apply not only to simple but to complex components. Once we’ve learned that the components of a sentence can be entire substructures, we don’t need to modify the inversion rule at all-- we can still simply invert subject and verb, even when the ‘subject’ has a complex structure of its own. The proposed ‘non-human’ rule would require a reanalysis of the inversion rule-- in effect, explicit instructions to disregard the tree structure and treat the sentence, rather unnaturally, as a single string of words. The proposed rule is more complex; there is no need to invoke biology to explain why languages prefer simple rules.
Other alternatives to tree structuring are generally not throught through. For instance, the structured rule is sometimes compared to a rule that forms questions by reversing the order of words in the sentence (Hands their raise will coming is who everyone?). But how could such a rule work?
These are not merely awkward facts, but obstacles to learning: how could a rule that doesn’t apply to the simplest sentences be learned or developed?
Such discussions do not establish that biology has chosen a rather quirky mechanism, tree structures, to base language on. Thought through, they show that tree structures are a natural and effective mechanism, simpler than any of the alternatives anyone has yet come up with, and consistent with how other complex entities without a conscious designer have developed.
However, in the spirit of Occam, we should require very good evidence before multiplying mutations, and see how far we can get with none at all.
For instance, we know that babies pay attention to their parent’s speech even as newborns. An instinct which tells the baby to pay attention to speech doesn’t seem outlandish. On the other hand, an instinct that tells the baby to pay attention to whatever the parents are doing explains just as much and, being more general and less human-specific, is more plausible. (It also better explains why infants are equally good at learning spoken or signed languages.)
It may be a useful exercise to consider how language might have developed without any language-specific mutations at all.
We must start with the apes. Chimpanzees are expressive and social creatures, and have both a wide repertoire of communicative signs and an admirable inventiveness in getting others to do what they want. Bickerton considers the results of ape language experiments to be the same as human pidgins. But compare an ape utterance like Nim Chimpsky’s
Give orange me give eat orange me eat orange give me eat orange give me you (Pinker p. 340)
to the Russenorsk pidgin utterance
If you buy-- please four pood. If you no buy-- then goodbye. (Bickerton p. 121)
Bickerton argues that the Russenorsk sentence is "hardly more complex" than the ape sentence; but this is hard to credit. The ape sentence is repetitive babbling; every word in the Russenorsk sentence is doing useful work. Nim’s sentence has no discernible internal structure; the Russenorsk sentence fits into a typical tree structure and seems to follow consistent rules: conditions before consequences; positive before negative conditions; verbs before objects; numbers before nouns. The existence of grammatical markers (if, then) is notable, as is the fact that they consistently precede their subclauses.
There does seem to be a qualitative difference between even this simple human utterance and Nim’s sentence, said (Bickerton p. 109) to be the longest recorded from an ape. Some may resist this conclusion; I can only quote David Berlinski: "No distinction in kind rather than degree between ourselves and the chimps? No distinction? Seriously, folks? Here is a simple operational test: the chimpanzees invariably are the ones behind the bars of their cages." Humans are brighter than chimps, have a greater flexibility with tools, are better adapted to a much wider variety of habitats-- and can handle a more complex language.
The most striking difference between humans and chimps, besides the hair, is probably our bigger brains. Could humans be able to handle language simply because they’re smarter?
I don’t mean to suggest that ‘intelligence’ is a unitary, fungible phenomenon. Quite the opposite: I would maintain that intelligence is a compendium of simpler capacities-- a cognitive toolbox. (Marvin Minsky’s The Society of Mind describes just such a toolbox.) We seem to have a bigger toolbox, and very likely more kinds of tools, than our ape cousins.
The difficulties in imagining the evolution of a language organ don’t apply to the evolution of the cognitive toolbox. There is no need for the toolbox to evolve all at once; there may have been a long series of mutations adding tools, making tools more efficient, or growing the toolbox. Some of the tools might even be language-specific (e.g. a lexicon partially centered in Wernicke’s area); but a tool that’s generally useful would be even more adaptive, since it could offer more than linguistic benefits (an encyclopedia partially centered in Wernicke’s area?).
A positive feedback loop would emerge driving the development of cognitive tools and language. New or improved cognitive tools (e.g. better memory; better analytical tools; better control over the vocal tract; better acoustic analysis) would allow the creation of ever more powerful languages; the existence of more powerful languages would also provide a rewarding environment for the enhancement of cognitive tools.
The use of tree structures would allow a great increase in language complexity at little cognitive cost. This might have been a cognitive change (perhaps explaining why music, sign language, and other human constructs are also tree-structured), or a cultural innovation.
There is no support for Proto-World under this scenario, because no group would experience a sudden jump in language skills that would allow them to develop a language from scratch or exterminate their neighbors. More likely, a riot of languages would develop, getting more complex as the cognitive toolbox improved. The languages would exchange vocabulary and features through contact with other humans, even as the genes underlying new tools spread through intermarriage.
Gradual development of language also seems to me to better fit what we know of early Homo sapiens, including the Neanderthals. Their material achievements (e.g. shaping stone tools, using fire) are more than apes have achieved, but less than Cro-Magnons; this (as well as the belief system implied by burial of the dead) fits in with a substantial but not perfected use of language. The Cro-Magnons’ advantage in language may have been not cognitive but physiological (better articulation) and cultural. Nor should we rule out the possibility that there was no linguistic advantage at all; Spanish is not more advanced than Quechua.
Why isn’t this process continuing-- why don’t we see minds and languages becoming even more complex? Of course, perhaps it is continuing, and we simply don’t notice over the mere milennia of human history. But if it’s stopped, there may be several reasons. Perhaps we simply ran into diminishing returns: more memory or other cognitive improvements stopped improving the lot of hunter/gatherers enough to provide a reproductive differential. Perhaps the population increased to the point that genetic changes tended to be damped out. Perhaps only competition with other hominid species provided enough rewards for increasing cognitive capacity. Perhaps any further increase in brain size would be energetically inefficient (brains take a lot of oxygen to run) or couldn’t fit through the pelvis.
I don't, by the way, claim that the cognitive tools are necessarily genetic. It's possible that the genes for bigger brains involve little more than instructions to let the brain grow for a longer time; everything else might be developmental or cultural.
The story in support of monogenesis and the language instinct, however, is no story at all. The evidence for the language instinct is weak, and the major supporting point-- the tree structure of language-- is useless, inasmuch as it supports a cultural or cognitive-toolbox origin of language just as well. And the belief in monogenesis rests only on a set of shaky, unexamined assumptions.
We will certainly hear more of all these ideas, as the origin of language is a compelling focus of interest. We must be on our guard, however, to analyze the stories we hear with a heavy dose of skepticism.
There are about 10 billion neurons in the brain, and 100 trillion interconnections. Meanwhile, an estimated 10,000 genes (1% of the gene pool) directly contribute to the architecture of the brain... the entire brain, not just language. The vast majority of these will be identical in humans and chimpanzees (and not too different in sheep).
There's just no way that a fraction of these genes can encode a complex linguistic program (to say nothing of the encoding of thousands of specific meanings, a form of innatism Chomsky indulges in some of his darker moments).
What could be done with a minimum of genetic impact that would nonetheless explain the differences between us and the chimps? As Harold Klawans puts it in Defending the Cavewoman, "those few genetic differences which make us Homo sapiens must relate primarily to prolonged or extended postnatal expansion and development of the brain."
Neoteny, in other words. At birth, our brains are about the same size as chimps' (about 350 cc). By adulthood, the chimp's brain expands about 28%. Ours expand 300%, to about 1400 cc. Most of our brain, then, develops after birth, and is subject to environmental influence.
The brain develops in an eccentric way, entirely unlike the way that computer programs are developed: the infant brain has trillions of extra pathways; the ones that are used get reinforced, and the ones that aren't atrophy and disappear. This is one reason for the 'window of opportunity for the development of language and many other cognitive functions. Child brains are qualitatively different from adult brains.
Michael Gazzaniga likes to emphasize the hard-wired nature of brains. Brains are not entirely general thinking machines, and if one part of the brain is injured there is not always another part that can take over. That seems like support for the 'language organ'. But Gazzaniga's picture is not that different from the 'cognitive toolbox' I discussed above; and it has to be supplemented by the very limited space in the genome for detailed specifications.
The question is still open, really. There's no question that our brains can develop all sorts of skills evolution never foresaw, from writing to video gaming to quantum mechanics. Some of these things even occur in predictable areas of the brain. But we don't try to shove these things into the genome.