Which aspects of the language faculty are uniquely human, and which are shared with other mammals? In this paper, Rodriguez and Granger try to address this fundamental question from an innovative neuro-computational perspective. They convincingly meet their challenging goal by developing two lines of argument.

The first line moves from the classical Chomskyan view of human language grammar as a set of combinatorial rules by which large linguistic units (called “phrases”) are put together to form sentences. In this framework, grammars describing “self-embedding” phrases, like in “The dog, I saw yesterday, ran away,” are known to be more complex than those describing “chaining” phrases, like in “Here comes the dog that chased the cat that killed the bird that I saw yesterday.” Accordingly, implementing the former grammars requires more powerful computational “machines” than implementing the latter. The general point here is that language machines must have the capacity to build phrases that contain other phrases, and, in order to do so, they must use memory stacks that contain other memory stacks.

Moving on to the neuroanatomical line of Rodriguez and Granger’s argument, they hypothesize that hippocampal operations in the human brain can compute stacks, and that loops of connections between the hippocampus and cortical areas implement potentially more powerful stack-using grammars. Because higher ratios of thalamo-cortico-cortical loops of connections can account for levels of embedding of increasingly larger linguistic phrases, it is tempting to conjecture that the language faculty evolutionarily arose from a quantitative leap in the ratios of connection loops among the cortex, the hippocampus, and the thalamus.

The intriguing conjecture makes the major contribution of this challenging paper, supported by allometric evidence of a continuously growing function between brain weight and brain component size in mammalians. Overall, reading the paper requires a non-superficial background in both computer science and neuroscience for its technical details to be understood in full. Nonetheless, its take-home message is at the same time straightforward, bold, and puzzling: language grammars have been evolving from general computational mechanisms for action coordination and navigation, adaptively exploited for communication purposes after a critical ratio of connection loops was reached. Structure embedding, which has authoritatively been held [1] as the sole species-specific tenet of the human language faculty (but see [2] for a different view), may in fact not be uniquely human.