Christiansen & Chater (C&C) provide a thought-provoking account of language as a cultural adaptation to the minds of language learners: Language evolves through repeated cultural transmission, adapting to multiple pressures impinging upon its learning, representation, and use. This strikes us as a powerful explanatory framework, and we hope the target article will stimulate more work fleshing out the precise relationship between constraints on language arising from the brain and features of linguistic structure.

In fact, we believe that there are two respects in which the target article actually understates the full implications of this hypothesis. First, C&C focus on language as an adaptation to language learners' brains – but language must also adapt to constraints external to the human mind. Second, in discussing their account's implications for understanding biological evolution, C&C point out the “moving target” problem – but there are at least two further issues with the interaction of cultural and biological evolution they do not recognize. Taken together, these completely change the viability of accounts based on the adaptive evolution of a strongly constraining, domain-specific biological faculty for language.

Although there are, of course, fundamental differences between the account offered by C&C and the Universal Grammar (UG) account they contrast it with, it is useful nevertheless to focus on a point of similarity: Both argue that language is how it is because the mind forces it to be so (the crucial difference being the extent to which the relevant mental components are domain-specific or domain-general, and who does the fitting to what). Both postulate a good fit between observed properties of language and properties of the human mind: if we see some linguistic property in the world, we can reasonably infer that it reflects a property of our minds.

We have previously demonstrated, however (Kirby Reference Kirby1999; Kirby et al. Reference Kirby, Smith and Brighton2004), that mental properties cannot simply be read off from language universals, because the cultural process mediating between aspects of the mind and features of language distorts the underlying biases of human learners. For example, culture can amplify weak biases to give strong universal effects. This means that cultural transmission can obscure underlying differences in mental structure: for example, strong universal tendencies towards regularity can result from various levels of preference for regularity in individual learners, and (in extreme cases) the same observed distribution of languages can result from learners with widely differing strengths of preference for regularity (Kirby et al. Reference Kirby, Dowman and Griffiths2007).

This suggests that, while we can probably get an idea of the general flavour of the human mind from observable linguistic properties, making specific inferences is more risky. To draw appropriate conclusions from a given distribution of languages, we must understand the rather opaque relationship between mental properties and linguistic properties resulting from cultural transmission.

Furthermore, not all pressures acting on language during its transmission are mental: language is not only well adapted to its users' brains, but also to its medium of transmission. Computational modelling work highlights the importance of the learning bottleneck: learners must infer the structure of an infinitely large language from a finite subset. This bottleneck introduces a pressure for generalisation to which language must adapt, for example, by becoming compositional (Kirby Reference Kirby2001; Smith et al. Reference Smith, Brighton and Kirby2003a). Importantly, the bottleneck is not a property of language learners' brains, but rather of the medium through which language is transmitted, namely, a finite dataset. Linguistic properties that are eminently learnable (and therefore well adapted to the brain) may be disfavoured due to this transmission factor, or vice versa. Consequently, linguistic features represent a product of, or a compromise between, several pressures acting on language transmission, some (but not all) of which reside in the human brain.

There are therefore good practical reasons for taking cultural evolution seriously. However, cultural evolution is more than just a methodological hassle for cognitive scientists: the same arguments radically alter the plausibility of scenarios for the evolution of the language faculty. C&C point out one problem for biological evolution, which arises from cultural transmission –evolution is chasing a moving target. We raise an additional problem: Given the opaque relationship between mental and linguistic features, much of the human biological capacity for language will be shielded (Ackley & Littman Reference Ackley, Littman, Langton, Taylor, Farmer and Rasmussen1992) from evolution. For example, Kirby et al. (Reference Kirby, Dowman and Griffiths2007) show that the strength of prior preferences for particular structures in language arising from an individual's biology may have no effect on the strength of the resulting universals. Under this scenario, selection is completely blind to the strength of the prior biases of individuals and there is no selection for nativisation, nor for stronger, “more desirable” prior preferences for certain languages (Smith & Kirby Reference Smith, Kirby, Smith, Smith and Ferrer i Cancho2008).

Furthermore, cultural transmission may actually hinder the evolution of a priori functional domain-specific features of the language faculty: Evolving biases for communicatively functional languages is problematic for culturally transmitted languages, because the biases of individual learners require a critical mass of learners and a reasonable span of cultural time to make their advantageous properties felt (Smith Reference Smith2004). Indeed, based on such models, we expect that practically the only scenario under which biological evolution will favour particular language-learning predispositions is when those predispositions are domain-general, and selected for on account of their less problematic (acultural, non-linguistic) applications. While this is entirely consistent with C&C's argument, we believe it also follows naturally from taking cultural evolution seriously.

So if evolving a functional domain-specific set of language biases is problematic, how can we best understand the evolution of the human faculty for language? C&C force us to revisit the human uniqueness of language from a comparative perspective, and (we suggest) pose a fascinating, important new challenge for evolutionary biology: Which of their suggested domain-general pressures responsible for shaping language are unique to humans, and what can this tell us about the evolutionary origins of the species-unique trait of language? Are there pre-adaptations for the cultural transmission of linguistic systems (e.g., vocal learning capacity; intention-sharing ability)? If so, how much linguistic structure naturally falls out of the cultural transmission process once these pre-adaptations are in place?