To the editors:

As long as there is language, there will be argument—especially about language. This state of affairs was recognized by the nineteenth-century Parisian sages that Robert Berwick and Noam Chomsky cite in their essay. After all, an entire century and a half later, there is still remarkably little agreement even about what language is, let alone about how we modern human beings acquired this unique and remarkable apomorphy. Whatever language may be, some observers discern its roots deep in primate vocal and even gestural communication, whereas others see it not only as strictly a property of modern humans, but even, at least in origin, as unrelated to communication.¹

A major difficulty here is that, as an abstract quality, language does not preserve directly in any material historical record. As a result, the use of language and of any of its putative precursors has to be inferred from indirect proxy evidence furnished principally by archaeology. The range of proxies for language that scientists have been willing to accept has been remarkably broad, adding to the confusion. Berwick and Chomsky sensibly settle on evidence for modern symbolic behavior patterns as the most reliable indicator of linguistic skill among extinct hominids. This conclusion allows them to situate the acquisition of this behavioral property about 100,000 years ago—within the tenure of our own species, Homo sapiens. They equally reasonably point out that the evidence for routine symbolic behaviors among the Neanderthals is thin at best, suggesting that, their large brains notwithstanding, those archaic hominids were in all likelihood nonlinguistic. Language, at least as we are familiar with it today, appears to be both uniquely a property of symbolic Homo sapiens and recently acquired. The quoted date of the divergence of the San from other modern human populations is, incidentally, irrelevant to this picture. Wherever in Africa language may have been invented, all that was required for its spread was that recipient populations had the potential to acquire and exhibit the new behavior. That potential had probably arisen in the neural rewiring that occurred as part of the radical developmental reorganization that produced anatomically modern Homo sapiens some 200,000 years ago. Language acquisition would almost certainly have been biologically possible for members of any structurally recognizable Homo sapiens population.

Berwick and Chomsky cite us in support of the view that looking for individual genes involved in language is excessively simplistic.² We reached this conclusion by employing recent genomic advances to scan archaic genus Homo genomes, including those of Neanderthals and Denisovans, to pinpoint Homo-specific or Homo sapiens-specific genes that might be relevant to language. We tried to enlarge the list of genes that might putatively be involved in language evolution by looking in the OMIM (Online Mendelian Inheritance in Man) and GWAS (genome-wide association studies) databases for matches between genes in those archaic Homo and modern sapiens genomes with single loci previously linked to language. The exercise resulted in very few matches. Only three candidate genes—CNTNAP2, GLI3, and WDPCP—were identified. Of these three, CNTNAP2 and GLI3 are involved in neural development and embryogenesis; it is thus conceivable that they might have roles in the evolution of the neural regulatory network that Berwick and Chomsky discuss in connection with SRGAP2C, another neural development gene. WDPCP does not seem to have any relevance to language or neural development. Our conclusion was nonetheless that fishing for genes using this approach most likely results, at best, in the identification of genes that might have the potential to produce a publicity splash, as was the case with FOXP2 and SRGAP2C, but that these genes would be only peripherally related to language acquisition.

More recently, we have examined the potential biological and genetic basis for several human meta-behaviors such as religious feelings, ethical behavior, our tendency to warfare and aggression, political preferences, sexual behaviors and identity, gravitation toward economic systems, and intelligence. Our analysis of these behaviors forced us to conclude that “attempts to associate human genes with particular complex behaviors have yet to get us very far … [and are] rather unlikely to get us much farther in the immediate future.”³

Like all organisms, we modern humans interact with our environments as the astonishingly complex beings we are, and not as carriers of particular behaviors. Eric Lenneberg also very cogently made this point in one of the passages that Berwick and Chomsky cite.⁴ It is conceivable that a single behavioral act might enhance the future reproductive fate of an individual, but we see this as the exception rather than the rule. In the long run, the evolutionary trajectory of a species counts immeasurably more than that of any individual, or of his or her features, or of genes that might explain only a tiny fraction of the behavioral variance within the species. Language was not one of the meta-behaviors we examined in our analysis, but we cannot help noting here that, as one of the most complex of human behaviors, it also is probably highly resistant to reductionist explanation.

In other contexts, the molecular geneticists Matt Rockman and Kenneth Weiss have made similar arguments about the applicability of genomic techniques such as quantitative trait loci (QTL) and genome-wide association studies (GWAS) to understanding the evolution of complex traits. After close scrutiny of both the QTL and GWAS approaches, they came to the rather depressing conclusion that genomic approaches are either using the wrong tools to approach key questions concerning complex traits, or simply asking the wrong questions. Rockman suggests that there is a “mismatch between question and method,” while Weiss and his coauthor Anne Buchanan argue that we should strive for a clearer goal.⁵ In the end, however superficially attractive those reductionist approaches might be, researchers seem to have reached a point at which they need to acknowledge that the traits we are most interested in are so complex as to be immune to simplistic atomization. Fortunately, we simply might not need fully atomized dissection of complex traits to make explanatory progress in evolutionary biology or medicine.

This appears to be the case with Berwick and Chomsky’s basic property (BP) of language, which they attribute to a “slight rewiring of the brain” that may have had extensive functional ramifications.⁶ As noted earlier, we are in full agreement with them on this point, with one caveat. Berwick and Chomsky see the BP, defined as “a finite computational system generating an infinite array of hierarchically structured expressions,” as “the expression of an underlying computational system” that is innate to humans. We would alternatively see the BP as simply one more manifestation, along with such other complex expressions as music and representational art, of a yet more generalized associative system that underpins the unique modern human symbolic cognitive mode. Symbolic cognition deconstructs the external and internal worlds into a vocabulary of discrete mental symbols that can be combined and recombined, according to rules, to make new statements about those worlds.⁷ It is the fount of our vaunted human creativity. If that thumbnail description makes our unusual cognitive style sound a lot like language, that is because language maps very closely onto symbolic thought; indeed, the linguist Wolfgang Hinzen has argued that thought itself is inherently grammatical.⁸

In our view, as well as in Berwick and Chomsky’s, the potential for modern human cognition was almost certainly born some 200,000 years ago with anatomical Homo sapiens. The archaeological indications are that this new potential lay fallow for upwards of 100,000 years, until it was activated by a cultural stimulus of some kind.⁹ The evolutionary phenomenon involved here is a routine one. The most plausible cultural stimulus was the spontaneous invention of language, which would then have been readily passed on among individuals and populations of this species that was already biologically enabled for it. This scenario is in complete agreement with Berwick and Chomsky’s requirement that “the final events leading to the BP must have been simple … a conclusion in agreement with the minimalist program.” But the scenario departs from their well-known contention that externalization came after internalized language.¹⁰ They bolster this position with Riny Huybregts’s recent conjecture that “the language faculty emerged with Homo sapiens, or shortly thereafter, but externalization in one form or another must have been a later development.”¹¹

There can be little doubt that Berwick and Chomsky are correct in viewing language as an essential portal to symbolic reasoning, as we think of it today; but to see its externalization as no more than a later afterthought deprives us of the incomparably best candidate we have for the cultural stimulus that incited a brain that was already structurally modern to switch from using the ancestral intuitive algorithm to operating in the modern symbolic one. Much more likely is that mutual reinforcement occurred between symbolic thought and spoken language as, amid the climatic rigors of late Pleistocene Africa, members of a small isolate of Homo sapiens possessing language-ready brains spontaneously began to attach specific meanings to strings of sounds, and to combine them into organized thoughts and utterances.

In conclusion, we return to the title of this piece, “Tough Luck.” These are the final words of the article by Richard Lewontin that Berwick and Chomsky approvingly cite.¹² Lewontin chose them carefully, because he saw no way in which human cognitive function could be atomized in a genetic context. Accepting tough luck is tough in itself. It is all the tougher when we allow it to lead us into chasing just-so stories.

Rob DeSalle and Ian Tattersall

Robert Berwick and Noam Chomsky reply:

We agree with much of what Rob DeSalle and Ian Tattersall write: basic to language is the system S, innate to humans, that yields the Basic Property (BP), where S is “a finite computational system generating an infinite array of hierarchically structured expressions.” These expressions have semantic interpretations as expressions of thought and can be externalized in one or another sensory-motor medium.

We further agree with their statement that “[t]he potential for modern cognitive function had probably arisen in the neural rewiring that occurred as part of the radical developmental reorganization that produced anatomically modern Homo sapiens some 200,000 years ago.” We also agree that the separation of modern humans began not long after their appearance, in the 150–200kya range. The date of separation is important, because it leaves little time for the neural rewiring. The crucial question is, what was this neural rewiring? Our proposal is that it was at least S, based on Merge, the simplest computational operation, and that S was available prior to the separation of modern humans. This would explain why it is shared among contemporary humans without known group differences. DeSalle and Tattersall instead propose that neural rewiring is a “generalized associational system that underpins the unique modern human symbolic cognitive mode.” What must be explained is how a generalized associational system yields S. We see no way at present to answer this question, and hence no way to explain why S is shared among human groups.

Until this question is answered, if it can be, it seems we must assume that neural rewiring includes S. We say “includes” because, while BP is at the core of the human language capacity, it does not exhaust it. Other properties of the human language capacity, which we have not elaborated, are being studied to discover to what extent they can be reduced to the language-independent principles of computational efficiency.