We wholeheartedly agree with Christiansen & Chater's (C&C's) central point that much of language evolution is likely to be adaptation of languages, via cultural evolution, to leverage species-typical properties of the brain. They are right to point out that while many language universals have been taken to reflect properties of a genetically evolved Universal Grammar (UG), they could equally well be universal culturally evolved properties of languages. This possibility is insufficiently recognized and should be a major focus of study.
What we take issue with, however, are the ideas that this is an “either/or” issue and that it is possible to rule out a priori the existence of mechanisms that evolved specifically as a result of their effects on language acquisition. Instead, how genes and culture interact to produce language, and whether domain-specific mechanisms exist for language acquisition, are empirical matters.
The biggest problem with C&C's argument lies in their attempt to rule out the existence of UG a priori by proposing a “logical” problem of language evolution: namely, that adaptation to “moving targets” is impossible. Unfortunately, this claim is simply false. “Moving targets,” in the sense of environments that vary over space and time, are the norm rather than the exception in biology. What matters for adaptation is not whether environments vary, but whether there are statistical regularities that natural selection could track. It is very likely that the world's languages do possess statistical regularities, some of which C&C point to.
In fact, it is not even necessary for the statistical regularities of the environment to be stable over time in order for adaptation to occur. One of the earliest ideas about this was Van Valen's (Reference Van Valen1973) “Red Queen” theory of antagonistic coevolution, in which different species, such as predator and prey, could be locked in ever-spiraling evolutionary arms races in which neither settle on a stable equilibrium. Nevertheless, these species can still evolve adaptations to the coevolving traits of the other, such as adaptations for fast running. Such coevolution is possible even when one party in the race can evolve “faster” than the other, as in host-parasite coevolution. This does not make impossible the existence of anti-parasite adaptations, such as the immune system (which C&C mention).
Perhaps the clearest and most relevant demonstration that genetic adaptation to culturally moving targets is possible can be found in the work of culture-gene coevolution theorists (Boyd & Richerson Reference Boyd and Richerson1985; Richerson & Boyd Reference Richerson and Boyd2005). The logic of these models is precisely that delineated by C&C: namely, that cultures vary substantially in time and space, such that precise genetic tracking of every cultural detail is impossible. Nevertheless, gene-culture coevolution models show that natural selection can favor genetically evolved mechanisms that facilitate the acquisition of local culture. Importantly, it is the statistical properties of cultural information that the genes track.
These models and associated empirical findings (Richerson & Boyd Reference Richerson and Boyd2005) show that culture and genes can and do coevolve. Although it is certainly possible to make models in which genes are “locked in place” and culture evolves to them, or vice versa, it is very unlikely that this is empirically what happens. Culture exerts selection pressure on genes, and genes exert selection pressure on culture, simultaneously. This means that questions like “which came first, the genes or the culture?” are inherently problematic. The “circularity trap” of C&C (sect. 3.2.1) is a problem faced by people who think in an either/or way; it is not a problem faced by the evolutionary process itself.
We suggest that the proper way to think about the gene pool of our species – and about the pool of cultural phenomena such as language – are as statistical clouds spread across space and time, each adapting to the other. C&C are entirely right that the properties of languages adapt to the statistical properties of the mind to make it more learnable and more easily understood. But there is, contrary to C&C's claims, no a priori reason why genes that do the exact same thing would not also be selected for. Indeed, if such genes existed, they would inevitably be selectively favored. It is a mistake to think either that language leaps fully formed upon the stage, and genes evolve to it, or that a genetic apparatus for language evolves, and then language sprouts from it. The process is likely to have been coevolutionary, all the way up.
What we expect, on this view, is a mesh, or fit, between genetically evolved mental mechanisms and language. The kinds of mechanisms we expect are ones that fit well with the statistical properties of language, and the statistical properties of language should fit well with them; how much of this fit has evolved on either side is an empirical matter. Language acquisition mechanisms can be seen as “prepared learning” mechanisms that reduce the frame problems inherent in any kind of learning by expecting certain kinds of regularities, or statistically present properties, to exist in the local language. These might include properties like long-distance dependencies; lexical types such as nouns and verbs; word order as a disambiguating device; hierarchical structuring; mechanisms for marking conceptual features such as space, time, causation, agency, and mental states; and more.
We recognize that many of these features might be argued to emerge from interaction with mental mechanisms that are not language-specific, such as conceptual mechanisms (although the language/conceptual interface could be a language-specific adaptation). Moreover, C&C and others (including, perhaps, Chomsky) might argue that these features should not be regarded as part of “UG” because they are not “arbitrary” or “non-functional.” However, we do not find it particularly useful to restrict UG to only “non-functional” features of language; among other things, it seems an odd way to carve up evolved structures. The important questions, for us, are twofold: (1) Do mechanisms exist that evolved because of their beneficial fitness effects on language acquisition? and (2) what are the computational properties of these mechanisms? We are happy to call these “UG,” though they might end up being very different from what Chomsky proposed. While we applaud C&C's efforts to draw attention to the culturally evolved properties of language that play a role in this evolutionary process, the authors have not convinced us that we should stop trying to look for the answers to questions (1) and (2).
