More recently, in a famous paper published in Science in 2002, Chomsky and his co-authors described a universal grammar that included only one feature, called computational recursion (although many advocates of universal grammar still prefer to assume there are many universal principles and parameters). This new shift permitted a limited number of words and rules to be combined to make an unlimited number of sentences.

The endless possibilities exist because of the way recursion embeds a phrase within another phrase of the same type. For example, English can embed phrases to the right (“John hopes Mary knows Peter is lying”) or embed centrally (“The dog that the cat that the boy saw chased barked”). In theory, it is possible to go on embedding these phases infinitely. In practice, understanding starts to break down when the phrases are stacked on top of one another as in these examples. Chomsky thought this breakdown was not directly related to language per se. Rather it was a limitation of human memory. More important, Chomsky proposed that this recursive ability is what sets language apart from other types of thinking such as categorization and perceiving the relations among things. He also proposed recently this ability arose from a single genetic mutation that occurred be­tween 100,000 and 50,000 years ago.

As before, when linguists actually went looking at the variation in languages across the world, they found counterexamples to the claim that this type of recursion was an essential property of language. Some languages — the Amazonian Pirahã, for instance — seem to get by without Chomskyan recursion.

As with all linguistic theories, Chomsky’s universal grammar tries to perform a balancing act. The theory has to be simple enough to be worth having. That is, it must predict some things that are not in the theory itself (otherwise it is just a list of facts). But neither can the theory be so simple that it cannot explain things it should. Take Chomsky’s idea that sentences in all the world’s languages have a “subject.” The problem is the concept of a subject is more like a “family resemblance” of features than a neat category. About 30 different grammatical features define the characteristics of a subject. Any one language will have only a subset of these features — and the subsets often do not overlap with those of other languages.