June 11, 2007

The enveloping Pirahã brouhaha

Back at the end of April, there was a conference in Normal, Illinois, on Recursion in Human Languages. Yesterday, the Chicago Tribune got around to reporting on it, and the article's adrenaline level is so high that it makes you wonder why they waited so long (Ron Grossman, "Shaking language to the core", Chicago Tribune, 6/10/2007):

To get some idea of the brouhaha currently enveloping linguists, occupants of a usually quiet corner of the ivory tower, suppose a high-school physics teacher found a hole in the theory of relativity.

Students of language consider Noam Chomsky the Einstein of their discipline. Linguistics is a very old science, but beginning in the 1950s, Chomsky so revolutionized the field that linguists refer to the time prior to his work as B.C., or before Chomsky.

They may have to add another marker: A.D., after Dan.

This is clearly a Hot Story -- if only it were true. Actually, it's a great story anyhow, but the way that Grossman tells it is, well, kind of misleading. At least, it seems that way to me. I'm not a syntactician, so you should take all of this with a grain or two of salt, but here's a discussion of the issues as I understand them.

Let's start by following the Trib's article a little further:

Daniel Everett, a faculty member at Illinois State University, has done field work among a tiny tribe in the Amazon. He reports that their obscure language lacks a fundamental characteristic that, according to Chomsky's theory, underlies all human language.

The ideas behind it are fairly basic: Some birds squawk and some animals grunt, alerting winged or furry compatriots to danger, but only humans can share complex thoughts.

A Scottish professor illustrated that at a recent gathering with a nursery rhyme: "This is the cat that chased the rat that ate the malt that lay in the house that Jack built."

In those lines, the word "that" is what linguists call a recursive device. Recursion allows humans to link various parts of our experience: to direct others to not just any cat, but to the one that chased the rat.
The device enables humans to pool knowledge and skills, share hopes and ambitions, build sophisticated societies and elaborate technologies.

Everett, however, fired a volley straight at the theory when he reported that the Brazilian tribe he was studying didn't use recursives. "For a long time, I said to myself: 'Maybe if I just hang around the tribe long enough I'll find it,'" Everett said. "But after 30 years, I don't know how much longer I'm going to be able to hang around."

OK, hold that thought. Now read the abstract of Rachel Nordlinger's paper "Spearing the Emu Drinking: Subordination and the Adjoined Relative Clause in Wambaya", Australian Journal of Linguistics, 26(1) 5-29, April 2006. As you do, keep in mind that Ken Hale, whose 1976 paper is so prominently cited, was Chomsky's colleague at MIT for several decades. This includes the period 1972-1975, when I was a graduate student there and Ken was working out the ideas presented in the 1976 paper.

Studies of subordination in Australian Aboriginal languages have been heavily influenced by Hale's foundational paper on the `adjoined relative clause'—a non-embedded, multifunctional subordinate clause type found in Warlpiri and a `large number of Australian languages' [Hale K 1976 `The adjoined relative clause in Australia' in RMW Dixon (ed.) Grammatical Categories in Australian Languages AIAS Canberra: 78–105 at 78]. Since this paper, almost every Australian grammar makes some reference to this clause type, presenting a general picture of structural homogeneity across subordination structures in Australian languages, and leading to the general perception that Australian languages typically don't have syntactic embedding. In this paper I present an analysis of subordinate clauses in Wambaya, arguing that these share many features of Hale's `adjoined relative clause' while still being clearly subordinate. The differences between subordinate clauses in Warlpiri and Wambaya show that complex constructions in Australian languages can be structurally dissimilar while sharing many of the properties of the `adjoined relative clause' type. I argue, therefore, that clause-combining in Australian languages may be more structurally heterogeneous than is traditionally assumed, and that a single analysis for complex sentences across a majority of Australian languages is quite likely inappropriate. This has implications for both the analysis and description of subordination in Australian Aboriginal languages, and for their relationship to the typological literature on subordination more generally. [emphasis added]

So Ken Hale's 1976 paper proposed that "Australian languages typically don't have syntactic embedding", and this idea has been widely accepted for the past 30 years. Ken was saying this sort of thing openly at MIT in 1974 or so, in the classes that I took from him. If this fundamentally challenged the foundations of Chomsky's theory, why weren't there any fireworks? Why isn't Grossman writing about AK instead of AD?

Well, this is partly because Ken Hale was not at all a rebellious or combative sort. But it's mostly because the idea that some languages completely lack clausal syntactic embedding (and thus a fortiori lack recursive clausal syntactic embedding) was perfectly compatible with Chomsky's theories -- until recently.

Through the 1980s and early 1990s, there were versions of Chomskian theory, countenanced within the range of speculation acceptable to his acolytes, in which you could set one or more parameters of the hypothetical Universal Grammar machinery, and get a syntactic system of the type that Hale considered Warlpiri and other Australian languages to have.

This began to change with Chomsky's "mimimalist program" in 1995, which jettisoned the idea of any well-defined layer of syntactic structure in between meaning ("logical form") and sound ("phonological form"), and therefore also discarded many of the parametric switches and knobs available in his earlier theories. And the change became important in the early oughts, when he decided that the human language faculty "only includes recursion". [See "JP versus FHC+CHF versus PJ versus HCF" (8/25/2005) for a discussion of some back-and-forth between Chomsky and others on this point, along with links to the original papers if you want every sanguinary detail.]

By the time that the "only recursion" idea became Chomskian orthodoxy, Ken Hale was dead. His response might have been to reconsider his analysis of Warlpiri, I don't know; but instead, other linguists are reconsidering it for him, and the confrontation between the "only recursion" idea and the analysis of Australian languages is being carried out quietly and implicitly in the discussions of these relatively obscure papers. I predict that this discussion will take a while to come to a consensus. For one thing, it's often not easy to decide whether or not structural embedding is involved in a particular class of examples. I explained some of the reasons in a post from last year: "Parataxis in Pirahã", 5/19/2006. And for another thing, there are many specific constructions in many different languages to consider.

Curiously, Dan Everett's path has been exactly the reverse of this reconsideration. In his 1986 paper "Pirahã" (pp. 200-325 in Desmond Derbyshire and Geoffrey Pullum, eds. Handbook of Amazonian languages), he presented a detailed analysis of Piraha morphosyntax that included many examples of recursive structures. In his recent work, he's reconsidered these ideas and offered new, non-recursive analyses. As he explains in "Cultural Constraints on Grammar in PIRAHÃ: A Reply to Nevins, Pesetsky, and Rodrigues,

As a matter of potential historical interest, I did wonder about my own analysis as early as 1984. Everett (1986) was actually written in 1982 in Portuguese, appearing initially as Everett (1983) and later as Everett (1990). So since I was a Visiting Scholar at MIT during this time, I talked to Chomsky about my idea that there seemed to be very little evidence for embedding of any kind in Pirahã, apart from these –sai examples which I was beginning to question. We discussed it briefly and Noam gave me some ideas for further testing the idea. Mark Baker, writing his PhD under Noam at the time, mentioned to me one day as we were having lunch that Noam was really intrigued by the idea that a language might not have embedding (Mark said something like 'You really got Noam's attention with what you told him about Pirahã', or some such).

[See "Dan Everett and the Pirahã in the New Yorker", 4/19/2007, for some explanation of the context of this rebuttal of a rebuttal.]

So Grossman's analogy -- "suppose a high-school physics teacher found a hole in the theory of relativity" -- is wildly off the mark.

In the first place, Dan Everett is no high-school teacher. He's been a well-established academic researcher for more than 20 years, with numerous widely-cited publications, and previous faculty positions at the University of Pittsburgh and at Manchester University. In the second place, the specific theory that Grossman focuses on ("recursion only") is no theory of relativity. It's a relatively recent idea -- dating only to 2002 in its current form -- and it has never been universally accepted, with prominent rebuttals from Ray Jackendoff and Steven Pinker, among others. (It's true that Dan is challenging the idea of Universal Grammar more broadly -- but without trying to diminish the interest of his arguments, it's fair to say that he's joining a long and distinguished list of challengers, stretching back to the beginnings of the UG idea in the 17th century.)

All the same, there is certainly a brouhaha centered on Dan. See Geoff Pullum's post "Fear and loathing on Massachusetts Avenue", 11/29/2006, for a description of some of the symptoms, and some keen insight into the causes. And the brouhaha continues, including what seem to be some shocking intrigues in Brazil, designed to persuade the government to bar Dan's access to the Pirahã on the basis of trumped-up charges of racism.

[Let me mention in passing that informed readers are also likely to have a quiet chortle over Grossman's description of linguists as "occupants of a usually quiet corner of the ivory tower". If you'd like to get in on the joke, you could read The Linguistics Wars, a description of "the fierce, acrimonious controversies that have rocked linguistics since the 1950s", focused mostly on the Generative Semantics heresy, a form of what those familiar with earlier sectarian strife might call "premature minimalism".]

In other Everett-related news, the CBC radio program "And Sometimes Y" will air a show (on June 23 from 11:30 am to 12 noon) about Universal Grammar in the context of Dan's claims about Pirahã. In addition to Dan, the interviewees include David Pesetsky and Martha McGinnis, two fine linguists whose positions on the issues differ from Dan's in ways that you should find interesting.

Also, an interview with Dan is now up at www.edge.org.

Finally, I'm told that a consortium including the BBC, ARTE, the Australian Broadcasting Company, PBS Nova, and the Smithsonian Channel will be doing a film about linguistics, Dan, and the Pirahã (if the Brazilian government authorizes it).

[A list of other Language Log posts on Dan Everett and the Pirahã can be found here.]

[Update -- Dan Everett writes:

I would simply like to emphasize that the popular media's reporting on this controversy has been extremely hit or miss. The New Yorker did a reasonably good job, but I am really weary of pieces like the Chicago Tribune's that puts this all in personal terms. It is bad for me, bad for the field, and potentially bad for the Pirahas, because it makes them look somewhat freakish. I wrote a paper and published it in a peer-reviewed journal. Looking past all of the publicity and vitriol, there is really only one appropriate way to deal with these claims: design experiments, go to the field, and test them. On the other hand, Nevins, Pesetsky, and Rodrigues did do a very good and important bit of work in their long literature-based criticism of my paper and I welcome the opportunity that they afforded me to (attempt to) clear up some things. I think that not only Piraha but the Australian languages Ken Hale worked with, among others, all need to be studied more carefully, with a range of experiments and different methodologies and theories. These are important issues and I certainly never claimed to have the last work on any of them, not even for Piraha. Whenever we publish anything, all of us, we know that if it is important, people will want more data, they will want to replicate the experiments, etc. That is what I expected and that is what should be done.

There are several empirical claims here: Piraha lacks recursion, numbers, counting, etc. And there is a theoretical account I propose: the explanation is cultural. If I am right about Piraha, then this does not mean that, say, Australian aboriginal languages, if they also lack recursion, have the same explanation. I claimed in the Current Anthropology paper that Piraha was the only language known in which the claim was that they had no embedding. Ken Hale did in fact make a similar claim, but he never made a big deal of it and the press didn't pick up on it. But I think too that he was never categorical in saying that these languages lacked recursion, i.e. that the Australian languages he worked on were *finite* languages, which is what I claimed for Piraha. However, I have thought a lot about Ken's claims on non-configurationality and I think that those claims and facts could easily be recast in terms of non-recursivity, rather than non-configurationality. Peter Austin has told me, after listening to me talk about this, that what I say about Piraha seems very similar to what he knows about Australian languages. And Rachel Nordlinger (who was also invited to speak at the Recursion Conference at ISU last month but couldn't come because of scheduling conflicts) has said similar things to me.

Piraha is perhaps unique in the constellation of features associated with it and, if I am correct, the particular cultural explanation offered. But who knows? More fieldwork is needed.

One thing that is clear, though, from the conference at ISU, from Ken's work, from mine, and from the work of many others: Recursion and its manifestations are not well enough understood to support the claim that recursion is the core component of human language.

Finally, let me just say that I sorely miss Ken Hale. I wish so much that he were here now to engage in the debate. I am sure that his calmness, kindness, and brilliance would help all of us to see these issues much more clearly, as he would also pour oil on the waters that have been so disturbed in recent months.

Amen. ]

[Note: You may have to listen to the CBC And Sometimes Y program in real time -- past shows seem to be available for a limited period as RealMedia streams, for example a show on Clichés that I was on a couple of months ago -- but mostly they want you to buy collections of their shows on physical media, in the weird old-fashioned style that some publically-funded broadcast organizations have not yet abandoned. This is enough expense and trouble that I don't imagine that very many people bother, and it's been a mystery to me for years why these organizations don't take the obvious step here, and make individual shows available for download. If someone knows the explanation -- management is too sclerotic to consider new ideas until a few years they've become routine elsewhere? someone's college roommate ekes out a meager living running the business of selling CDs? -- please tell me.]

Posted by Mark Liberman at June 11, 2007 07:15 AM