The theory of formal languages, initiated by Noam Chomsky, introduces a fundamental difference between finite-state and context-dependent grammars: only the latter are capable of representing the embedded, recursive structures that underlie the mental representation of human languages. Hauser, Chomsky and Fitch's hypothesis is that only our species is capable of recursion. Numerous experimental studies have attempted to disprove this postulate by teaching various species of songbirds an artificial grammar known as " AnBn ", i.e. a series of n A sounds followed by n B sounds. This grammar has the advantage of requiring a context-dependent grammar, and its learning, in the human species, seems to involve Broca's area. Some have claimed that songbirds can learn this type of grammar, but many theoretical and practical arguments suggest that this research is not convincing : animal behavior is often unspecific, insufficiently evaluated, without analysis of individual performance, and too many alternative interpretations remain tenable.
Recently, however, using a very different approach, Liping Wang's team and I have demonstrated that macaque monkeys are capable of learning a " mirror grammar ", with sequences such as ABBA, ABCCBA, etc., where the second part repeats the first in a temporal mirror. They do this in a visual-spatial context, not an auditory one : their spatial memory enables them to repeat a sequence both forwards and backwards. Thus, the macaque monkey brain has the capacity to learn a supra-regular language, a grammar independent of context (but not necessarily all such grammars). At the very least, solving this task requires a " last in, first out " stack. The uniqueness of the human species may lie in the speed with which this type of rule is discovered : young preschoolers discover the rule in no more than five trials, and then apply it almost perfectly, whereas it takes macaque monkeys several tens of thousands of trials to achieve an imperfect performance. The human brain therefore seems biased towards rapid learning of rules and embedded structures.
