June 24, 2007

Roeper on recursion

[This is a guest post by Tom Roeper, commenting on my earlier post "The enveloping Pirahã brouhaha" (6/11/2007).]


Mark Liberman’s claim that the Chicago Tribune piece was misleading was quite right---it was even misleading about the content and tenor of the conference. I was given one of the most minimal quotes one can imagine (one word, repeated (no recursion!)): “no, no” with no explanation.) Here’s what got left out.

I pointed out, and Terry Langendoen from NSF immediately agreed, that there were different kinds of recursion with important consequences.

First ---which I asked Dan about---is the fundamental and universal form (according to Chomsky) of assymetric Merge, which means you cannot put three words together without recursively merging them and letting one dominate. That’s what happens in compounds like:

bird migration computer manual trouble-shooting guide

which we can understand (perhaps in multiple ways) by this method. If this is correct (hard to see how it could be wrong), there is no more discussion. Every sentence involves recursion.

Perhaps the most interesting fact is that children grasp this asymmetry from the start. They say things like “Mommy shoe” but not *“Mommy Daddy”. Why? Because in “mommy shoe” there is a modification relation, “shoe” is the Head and “mommy” the modifier, but not where coordination is implied (Mommy is not a modifier of Daddy nor Daddy a variety of Mommy).

Second are language particular forms like recursive possessives (John’s friend’s mother’s hat).

In my new book The Prism of Grammar : How Child Language Illuminates Humanism (MIT Press---now out: publication date: May 22, 2007) there is an extensive chapter on merge and another on recursion with evidence that children have difficulty with language particular dimensions. Here’s an example. Only English, but not German, Swedish, or Dutch (which allow a single pronominal possessive), allows recursive possessives. They are not easy for children. Witness Sarah (Childes files):

MOTHER: What's Daddy's Daddy's name?
SARAH: uh.
MOTHER: What's Daddy's Daddy's name?
SARAH: uh.
MOTHER: What is it? What'd I tell you? Arthur!
SARAH: Arthur! Dat my cousin.
MOTHER: Oh no, not your cousin Arthur. Grampy's name is Arthur. Daddy's Daddy's name is Arthur.
SARAH: (very deliberately) No, dat my cousin.
MOTHER: oh. What's your cousin's Mumma's name? What's Arthur's Mumma's name?
SARAH: uh. oh.
MOTHER: Thinking?
[Sarah nods]
MOTHER: And what's Bam+Bam's daddy's name?
SARAH: Uh, Bam+Bam!
MOTHER: No, what's Bam+Bam's daddy's name?
SARAH: Fred!
MOTHER: No, Barney.
SARAH: Barney.
MOTHER: What's his mumma's name?
SARAH: She's right here.
[ points to figure on Sarah's pajamas which have TV characters on them]

Try this out on your four-year-old and you will almost certainly have the same experience. (Some methods are suggested in the Prism book.)

Thirdly there is sentential recursion—what all the hot discussion is about. And this seems to be connected to how complicated our thoughts get. As Aravind Joshi in one paper and Bart Hollebrandse and I in another pointed out at the conference, it is in principle possible to get the same effect in discourse, but it seems to be prohibitively difficult. This is the subject of several papers and acquisition experiments (by Bart Hollebrandse, Kate Hobbs, Jill deVilliers and me---to be presented in at the GALA acquisition conference in Barcelona in September). Sequences like

John thought the earth was flat.
Fred believed it
But Bill could not believe that.

[=Bill could not believe that Fred believed that the earth was flat.]

However, note the inexactitude. The last that is much more ambiguous than the sentential counterpart: it could mean [that the earth is flat] or [that Fred believed the earth is flat] or [that Fred believed John thought the earth is flat]. (See also Liberman’s discussion of parataxis in Language Log ("Parataxis in Pirahã", 5/19/2006)] Discourse connections get too much, allowing any sort of inference---that’s the problem. If one wants to do further deductive reasoning, then only the sentential version is ideal, but the obscure discourse form is rich enough for complex human interactions to be understood. Hollebrandse and I (see forthcoming conference paper) have argued that recursion produces exclusive complementation under embedding.

Bart Hollerandse is in fact preparing materials to actually carry experiments like those used in acquisition with Dan Everett in Brazil with the Piraha together in hopes that one can find out exactly what linguistic mechanisms they use to approach complexity of this kind.

What real exclusive readings make more efficient can be found every day in the New York Times, where sequences like this can occur:

Gore believed that Bush’s belief that Gore had exaggerated his claim that he authored the internet forced Gore to defend himself against the claim that he was sleazy instead of his forcing Bush to defend himself against the claim that he had not acknowledged that has fund-raising was illegal.

The style is lousy but it gets across a complex set of dependencies that inferences across a discourse probably could not keep straight. [Clarification by Liberman: a search of the NYT archive suggests that this sentence is a hypothetical quotation, invented by Tom. For some real-world examples, see here, here, here, etc., though in fairness, one does not see these every day.]

Recursion is useful----which is not to say that the essence of our humanity depends upon it. Just like zeroes in math are very useful—much better than Roman numerals for multiplying-. Still the Romans managed to multiply anyway. Their society was still complex without having zeroes.


[Above is a guest post by Tom Roeper.]

[Update -- here is a comment by Dan Everett:

Tom's response on the recursion conference not only accurately conveys my own sense of the way things transpired at the conference, it also shows why his own contributions to discussions of recursion are extremely important. The paper by Bart and Tom at the conference really helped me to think more effectively about the issues.

Part of the cultural influence on Piraha's absence of syntactic recursion is the fact that the Pirahas purposely circumscribe the things they talk about, so that they have a much narrower range of discourse topics (again, by choice). As I understand at least part of what Tom Roeper and Bart Hollebrandse said in their contribution, recursion is a tool that does not define language per se but turns out to be extremely important in organizing complex utterances, a complexity which is at least in part a result of the complexity of the things most societies want to talk about.

At the same time, again as Tom accurately points out, Aravind Joshi and they agreed that it is in principle possible to get the same effects via discourse. This is interesting and, I believe, quite important because it is at the core of my own claim that Pirahas do have recursion in their discourse, but not in their syntax. What would be the source of this recursion? The brain. Humans think recursively. I think that this is a characteristic (though more study is needed) of our species. But as Ken Hale and I, and no doubt others if we look through extant grammars more carefully, have shown is that syntactic embedding is certainly not a universal characteristic of human languages.

Once again, when societies limit their discourse - not through any cognitive shortcomings (!) - because of their isolation or their cultural values (explicit or implicit), then the need for recursion, by my hypothesis, by Roeper and Hollebrandse's hypothesis, and by Joshi's work, is reduced, if not eliminated. Recursion is thus a tool for organizing our thoughts and our speech. It seems always necessary in the former, but not always necessary as a tool for speech. Once again, evidence for this in Piraha is recursion in discourse (without formal marking).

This issues and a large selection of the papers from the recursion conference at ISU will be the subject of a special issue of The Linguistic Review dedicated to recursion which I am guest-editing.

I hope very much that in the near future someone will be able to run Bart's experiments among the Pirahas.

Echoing what Tom said, it seems to me that we should start trying to use terminology more carefully in these discussions. You could have limited embedding without recursion; either recursion or embedding might involve clauses, or alternatively (for example) the various sorts of non-clausal structures involved in complex nominals; right- or left-branching structures are different from center-embedded structures; etc. Thus it would be clearer if Dan wrote "clausal syntactic embedding is not a universal characteristic", etc. Various other sorts of embedding, e.g. of a modified noun as a dependent of a verb, clearly are universal. Whether these embeddings are universally recursive is another question.]

Posted by Mark Liberman at June 24, 2007 09:58 AM