(This is now cross-posted at my personal blog here)
As I write this I am sitting in the Linguistics Department lounge at the University of Toronto. Grad students and Post-doctoral researchers are working, chatting, making coffee. Faculty members pop in every now and then, taking breaks from their work.
It’s a vibrant department, full of researchers with varied skills and interests. There are those who just got back into town from their summer fieldwork, excited to dig into the new language data from indigenous Canadian, Amazonian, or Australian languages. There are those struggling to find a way to explain the behaviour of some set of prefixes or verbs in Turkish, or Spanish, or Niuean. There are those designing and running experiments to test what young children know about the languages they are acquiring. There are those sifting through massive databases of speech from rural farmers, or lyrics of local hip-hop artists, or the emails of Enron employees, to hopefully gain some understanding of how English varies and changes. And there are those who spend their time thinking about our theories of language and how they might be integrated with theories of Psychology, Neurology, and Biology. What unites these disparate research agendas is that they are all grounded in the hypothesis, generally attributed to Noam Chomsky, that the human mind contains innate structures, a Universal Grammar, that allows us to acquire, comprehend, and use language.
According to a recent article in Scientific American, however, the community I just described doesn’t exist, and maybe couldn’t possibly exist in linguistics today, because the kind of work that I just described has long since shown the Universal Grammar hypothesis (UG) to be flat-out wrong. But such a community does exist, and not just here at UofT, or in Chomsky’s own department at MIT, but in Berkeley and Manhattan, in Newfoundland and Vancouver, in Norway and Tokyo. Communities that collectively groan whenever someone sounds the death knell of the UG hypothesis or the enterprise of Generative Linguistics it spawned. We groan, not because we’ve been exposed for the frauds or fools that these pieces say we are, but because we are always misrepresented in them. Sometimes the misrepresentation is laughable, but more often it’s damn frustrating.
Articles like the one in SA are frustrating because they are usually wrong, both about the history and culture of UG and about the idea itself.
The picture of UG’s history, as painted by the SA article is one of Chomsky coming up with a nice theory of language based on a handful of English sentences and passing that theory down to the academics who accept it without question or debate. The academics, then, play around with his theory and make grander and more absurd claims, all without leaving the safety of their ivory towers. Meanwhile the real linguists are out in the world collecting data from languages all over the globe, data which is devastating to UG and therefore stifled by Chomsky and his acolytes, the generative linguists.
This picture is false on two counts. First, Chomsky and UG have always faced strong and vocal opposition. The philosophers and linguists who were Chomsky’s contemporaries in the 50’s and 60’s objected to the very notion that language could be used to scientifically investigate the mind. Many of his earliest students developed competing theories of grammar which are still worked on today. And that’s not even counting the linguists who largely agree with Chomsky’s picture of language but disagree with some of the technical details he proposes.
Second, far from being hostile to language data gathered by fieldwork (or experimental work, or corpus work) generative linguists often seek out new language data to test predictions made by their theories. Some generative linguists, such as Keren Rice and the late Ken Hale, are known not only for their contributions to generative linguistics but also for their commitment to fieldwork and advocacy on behalf of the marginalized communities in which they do/did their fieldwork. And the interest in a wide variety of languages extends beyond fieldworkers. A glance at the program for virtually any generative linguistics conference will demonstrate that. Because UG is such a solid theory, throwing language data at it doesn’t kill it, but makes it better.
Which brings us to the mistaken view of the actual theory of UG that article such as the SA piece present. Daniel Everett, a field linguist and former evangelical missionary, gained prominence in 2005 when he claimed that Pirahã, a language spoken by a remote Amazonian tribe showed properties that categorically refuted the UG hypothesis. Central to UG theory, in Everett’s estimation, is embedding, the ability of a language to place, for example, a clause inside another clause (as in “I heard that Maura laughed”). Pirahã, it seemed, was unable to embed.
On the surface, this does seem like a knockout punch to UG, but there’s just one problem with it: Everett is mistaken about what UG is.
The source of his confusion seems to be the term recursion, which is the central concept of modern Generative linguistics, not embedding. This might be confusing. I confess I don’t know that I understood the distinction until I was well into grad school. When we think of recursion, what comes to most people’s minds is the Droste effect, the effect of a picture appearing within itself.
The Droste effect is an example of embedding, and it is a demonstration of a possible effect of recursion, but it is not the definition of recursion. The term recursion, when used by Generative linguists, refers to a property of functions. Functions are relations between a set of inputs and a set of outputs. A function is recursive if its output can also serve as its input. The +1 function is recursive when applied to integers (1+1=2, 2+1=3, etc.). A function that converts numbers into their Roman numeral representation is not recursive (7 → VII, VII → ??). For generative linguistics the recursive function is Merge, which combines two words or phrases to form a larger structure which can then be the input for further iterations of Merge. Any expression larger than two words, then, requires recursion, regardless of whether there is embedding in that expression. For instance the noun phrase “My favourite book” requires two iterations of Merge, (Merge(favourite, book)= [Favourite book], Merge(my, [favourite book])= [my [favourite book]]) and therefore is an instance of recursion without embedding.
The confusion between recursion and embedding, is due to the fact that only recursive functions are capable of generating embedded structures, not all recursive functions generate embedded structures. The relationship between recursion and embedding, is like the relationship between precipitation and snow. If we see snow on the ground, we know there’s been precipitation, but if we don’t see snow, that doesn’t mean there has been no precipitation. So, for Everett to say that Pirahã’s alleged lack of embedding means it lacks recursion is equivalent to him saying the Amazonian climate lacks precipitation because you never find snow on the ground there.
I’ll admit this is a subtle distinction. The average person, even the average linguist, doesn’t need to worry about the distinction. But here’s what makes Everett’s claims and that SA article (written by two academics, Paul Ibbotson and Michael Tomasello) so frustrating: They should know better. Not because their day-to-day research projects require them to know better, but because when you claim that someone said something incorrect, you had better know what it was they said. What’s more, when the recursion/embedding distinction is mistakenly blurred by those who reject the UG hypothesis, Generative linguists are quick to correct the mistake.
So, being trained academic researchers, Everett, Ibbotson and Tomasello should have researched the claims they intended to rebut. But even if they missed or misinterpreted something in Chomsky’s writing, a number of Generative linguists have already clarified their mistake. Why, then, do they persist in misrepresenting the claims they are trying to rebut?
I’m certain there are more irritating articles and books proclaiming the death of UG coming, but in the meantime, in vibrant communities like the UofT Linguistics Department, Professors, Post-docs and Grad students will continue to investigate UG, with every tool at our disposal. We’ll gather data from remote villages and diaspora communities in our cities. We’ll run statistical analyses on corpora and develop formal models. We’ll present our findings and debate proposals. And in doing so, we hope to continually better our understanding the deep properties of our species’ unique ability to acquire, comprehend and use language.
Edit: I’ve removed the assertion that that mathematicians and computer scientists share the definition of recursion given here.