William H. Calvin and Derek Bickerton, Lingua ex machina: Reconciling Darwin and Chomsky with the human brain (MIT Press, forthcoming). See also

This is a précis of the forthcoming book; links to the individual chapters are at the end. Some photographs of Bellagio can be found here.
Webbed Preprint Collection
This 'tree' is really a pyramidal neuron of cerebral cortex.  The axon exiting at bottom goes long distances, eventually splitting up into 10,000 small branchlets to make synapses with other brain cells.
William H. Calvin

University of Washington
Seattle WA 98195-1800 USA

Précis of

Lingua ex machina

Reconciling Darwin and Chomsky
with the Human Brain


William H. Calvin and Derek Bickerton

A machine for language? Surely not, say the people who associate machines with stolid reliability - language, after all, is the basis for metaphor, up there near the heights of creativity. Machines aren't creative (and if they try, we get them repaired).

Surely yes, say the neurophysiologists, busy studying the language specializations of the human brain and trying to identify their antecedents. What else could it possibly be?

Linguists like Noam Chomsky talk about machinelike 'modules' in the brain for syntax, that language is more of an instinct (a complex behavior triggered by simple environmental stimuli) than it is an acquired skill like riding a bike or playing the piano. That's because many potential systems for syntax are never used in any human language, making you wonder what's common to the ones actually used (a common neural substrate, in all likelihood -- and that suggests that genes might have something to do with it).

But structured language seems to present the same evolutionary problems as forelimbs for flight: you need a lot of specializations to fly even a little bit. So how are you going to get them, if evolution has no foresight and the intermediate stages don't have intermediate payoffs? Many argue that the same long-time-until-useful objection applies to the structured aspects of language. The wonderful Darwinian scheme for species self-improvement cannot, they say, explain our most valued human capability, the one that sets us so far above the apes, language itself.

How, then, do we reconcile the facts arrayed around 'Darwin' and 'Chomsky' with the obvious, namely that the human brain has the capacity for structured language? That children routinely discover its principles in the first few years of life, inferring them from the sound chaos that surrounds them?

Fortunately one can avoid invoking miracles by simply taking a less cartoonlike view of evolutionary mechanisms. Charles Darwin himself was careful to emphasize that natural selection need not be for the same thing, all along the evolutionary path to a given function: the usefulness of a structure could be initially for one function, and then the structure might become useful for some very different function. Darwin's example was that a swim bladder might have been converted into a primitive lung, for venturing out onto land. Forelimb feathers might have been under selection for thermoregulation and, when there were enough for gliding, the selection might have shifted to emphasize airfoil efficiency. Something paid for in one coin might, when reaching some critical mass, become useful for a new function, with further improvements paid for with a new coin. Foresight might have accomplished this more quickly, but it really isn't required for the result. 'Preadaptations' suffice.

But it has been difficult to identify nonlanguage predecessors of structured language. Most books on language evolution seem to dissolve into a series of vague generalities when it comes to deal with specific issues, such as how language acquired its universal characteristics, or what novel processes in the brain made language possible. The linguists haven't known the relevant neurology or evolutionary theory, and the brain researchers haven't known enough linguistics to realize what features required a detailed explanation. And there is, in addition, the requirement of continuity: How do your proposed mechanisms create an easy series of improvements or conversions of function, stages that span the whole distance, starting with chimpanzeelike utterances, working up to words and protolanguage, and finally to the phrases and clauses that enable you to say "I think I saw him leave to go home" with its nested embedding?

We do it here with about two-and-a-half preadaptations, yielding both argument structure and phrase structure or, if you prefer, the newer minimalist grammar. The coin in which these improvements were purchased was that of reciprocal altruism, the cognitive categories handy for detecting cheaters, and of ballistic movement planning, handy for toolmaking and hunting.

And, of course, once structured sentences were possible, natural selection could operate on improving it. Since planning ahead and chains of inference are likely to use the same neural mechanisms as syntax, their usefulness can also drive improvements in syntax mechanisms. But the invention of such structured symbolism can be a sidestep in evolution, one of Darwin's conversions of function in anatomical continuity.

The mental categories needed for reciprocal altruism, a well-established phenomenon in many primate species, show a strong resemblance to those needed for understanding a structured sentence. The detection of freeloaders, important in any primate society, became crucial with the emergence of dyads that cemented their alliances by mutual grooming, food-sharing, and so forth. If such stable arrangements were not to collapse into an anarchy of freeloaders, it became necessary to detect cheaters. But this in turn required a cheater detection method that would keep a rough score of transactions via a template for events which contained slots for Action, Agent, Theme (or Patient) and other abstract roles such as Beneficiary.

The categories and event structure required for cheater detection are precisely the thematic roles and argument structure that later form the core of universal syntax. It is envisaged that this cheater detection mechanism anticipated the earliest forms of language by millions of years, and that it co-existed for perhaps two million years with a primitive protolanguage until a crucial threshold in brain development was crossed that allowed the two quite independent developments to mesh smoothly with one another. Once the core structures and properties of syntax (including recursivity) had thus been laid down, novel selective pressures would have come into operation and a series of 'Baldwin effects' would have quickly formed and fixed other universal features, such as the use of grammatical morphemes for government and case assignment, and an algorithm for assigning reference to unspoken arguments (the equivalent of Chomskian 'chains' and the 'Empty Category Principle').

Another preadaptation for structured sentences can be seen in the movement planning needed for hammering and throwing. First came two developments that were related to an increase in the efficiency of aimed throwing, essential if hominids were to survive through temperate-zone winters when food-sources other than meat were rare or nonexistent. One of these was an increase in the number of neurons required to reduce timing 'jitter' as improved range and accuracy called for an ever narrower launch window, a factor that helped to drive brain expansion. The other was an ever more detailed planning of the complex sequences of segmented movements -- upper body, shoulder, arm, wrist, etc. -- that precision hammering and aimed throwing require. Each rotation is embedded in the ones higher up, as when wrist rotation has to be planned on the basis of an lower arm that is itself accelerating, and elbow planning is itself embedded in upper arm movement, and so forth.

These developments in cognition and movement planning, apparently quite unrelated to any 'higher process,' produced unforeseeable consequences for all of higher intellectual functions. The process of 'training up' abstract and intricately embedded movement sequence commands would serve as an important preadaptation for the nested embedding of phrases and clauses, once other developments had made it possible to produce long and complex sentences.

But perhaps the most crucial event was the achievement of coherence in the transmission of long-distance corticocortical messages, the equivalent of the fiber optic bundles used in endoscopes for viewing ulcers. The anatomy is incoherent (axons do not consistently remain neighbors over long distances) but physiology can temporarily correct for it. This arose in part as a result of ballistic movement planning. It made possible the large cell assemblies necessary not merely to establish consistent codes for nominal and verbal concepts, but to handle the composite codes that resulted as word-codes were merged to form phrase-codes and phrase-codes were merged to form codes for clauses -- all to be transmitted, coherently and without any learning prerequisite, to the motor organs of speech that would execute the resultant sentence. The increase in size of the cell-assemblies that could be recruited for each of these superposed, coded messages likely drove an increase in the number of fibers in the arcuate fasciculus, one of the main pathways of the brain (second only to the corpus callosum), which links the prefrontal and temporal lobes most crucially involved in language production, and this increase in turn helped made it possible to achieve the coherence threshold required for accurate transmission at the high speeds required by speech.

While all these developments were, of their nature, gradual and incremental, they were not of a type that, individually, would have brought about much improvement in language; indeed, they would not likely have taken place if language had driven them rather than throwing. Structured messages in the brain that are not fully coherent are not partially coherent: they are completely incoherent, and with anything less than a coherent message, the brain could not produce a long sentence. Thus the final crossing of the coherence threshold would allowed language to pass in a single step from a structureless protolanguage to something that would have been recognizable as a modern human language, although it might at first have lacked some of the streamlining features that Baldwinian evolution would later fill in.

For many decades, we have faced a dilemma. On the one hand we have had the thesis of Chomskian nativism, with its unimpeachable evidence. On the other we have had the antithesis of Darwinian evolution, with its unimpeachable evidence. The whole problem, not just with language evolution but with the understanding of all higher intellectual function, has been that these two bodies of unimpeachable evidence seemed to point in contrary directions, and countless millions of words of print have been devoted to warfare between them -- a warfare doomed to futility, since the work on each side contained profound truths which partisans of the other felt obliged to denigrate, deny or simply ignore. Lingua ex machina shows that the apparent opposition between these bodies of thought is illusory, and presents a synthesis showing how a universal, innate language faculty could have been produced without violating any of the basic principles of evolutionary theory. As such, it presents a constructive message that should gain the attention of all who concern themselves with our place in nature.


William H. Calvin is a theoretical neurophysiologist at the University of Washington in Seattle. He is the author of nine books, mostly about brains and evolution; his research on neocortical circuits, The Cerebral Code, published in 1996 by MIT Press is part of a long-term interest in how Darwinian processes in the brain can operate in mere seconds to shape up novel plans, such as sentences to speak aloud. He is also the author of The Atlantic Monthly's January cover story, "The Great Climate Flip-flop," which will become a book about abrupt climate change and the role it played in human evolution. His web pages start at

Derek Bickerton is an emeritus professor of linguistics at the University of Hawai'i in Honolulu. He is the author of Roots of Language, Language and Species, and Language and Human Behavior. His research on how children exposed to pidgins convert them into creoles provides one of the key lines of evidence for universal grammar. He has studied the evolution of language for two decades and one of his major concerns is to resolve the apparent conflict between formal, nativist accounts of universal grammar and the evidence from paleoanthropology and neurology.

copyright ©1998 by William H. Calvin and Derek Bickerton

June 1998 DRAFT

Send those comments to:
Email Calvin

Email Bickerton


1. Stage Setting

2. What Are Words? (DB)

3. Of Trains and Pyramids (DB)

4. Bigger than a Word, Smaller than a Sentence (DB)

5. Language in the Brain (WHC)

6. How Are Memories Stored? (WHC)

7. Hexagonal Mosaics and Darwin Machines (WHC)

8. A Common Code: The Brain's "Esperanto" Problem (WHC)

9. Protolanguage Emerging (DB)

10. Why Language Birth Doesn't Have to be Gradual (DB)

11. Reciprocal Altruism as a Predecessor of Argument Structure (DB)

12. Role Links for Words (DB)

13. The Word Tree as a Secondary Use of Throwing's Segmented Movement Planning (WHC)

14. Corticocortical Coherence Promotes a Many-voiced Symphonic Sentence (WHC)

15. The Pump and the Slingshot (WHC)

16. Darwin and Chomsky Together at Last (DB)

End Notes


Email Calvin || Email Bickerton || Calvin Home Page || June 1998