May 19, 2006

Parataxis in Pirahã

Take a look at this quote from Elmore Leonard's LaBrava:

"What're you having, conch? You ever see it they take it out of the shell? You wouldn't eat it."

The speaker is Maurice Zola, "five-five, weighed about one-fifteen and spoke with a soft urban-south accent that had wise-guy overtones, decades of street-corner styles blended and delivered, right or wrong, with casual authority".

"Wise-guy overtones", check; "casual authority", OK; but is this guy speaking a human language? Maybe not, according to Marc Hauser, Noam Chomsky and Tecumseh Fitch. If they're right, we need to wait and see whether Maurice comes up with any genuinely "recursive" syntax. Until then, the humanity of his soft urban-south way of talking is uncertain.

In a 2000 article in Science, Hauser, Chomsky and Fitch (HCF) argued that the "aspects of language that are special to language ... only [include] recursion". Steve Pinker, Ray Jackendoff and many others have disagreed. (A quick historical sketch of the debate, with links to the original publications, is here.)

This debate underlies the recent series of articles about whether monkeys and birds can learn "recursive" patterns. (You can read about the experiments on cotton-top tamarins here and the experiments on starlings here, and you can find a list of more than a dozen other relevant Language Log posts here. A short note from Ray Jackendoff, Geoff Pullum, Barbara Scholz and me about this stuff is here.)

By "recursion", HCF mean "computational mechanisms ... providing the capacity to generate an infinite range of expressions from a finite set of elements". "Recursion" in this sense goes beyond the simple combinations of modifiers and heads ("red" + "cow" → "red cow"), or subjects and verbs ("Joan" + "disagree" → "Joan disagrees"), or any other construction that doesn't involve embedding a complex element repeatedly inside another element of the same type. Non-recursive constructions (like modifier+head) are very useful, and such embeddings multiply the set of messages that you can make out of a finite set of elements, but they don't "generate an infinite range of expressions" unless they operate recursively.

And "recursion" in HCF's sense also excludes simply stringing together expressions in a sequence. Animal signaling, human or otherwise, is limited to a single item only if an unhappy accident immediately silences the signaler. Otherwise communication, like life, is just one thing after another; and if mere sequence is recursion, then bacterial signaling is recursive. To be syntactically "recursive", a message must involve structural embedding that goes beyond concatenation or juxtaposition.

That's why Dan Everett ("Cultural Constraints on Grammar and Cognition in Pirahã: Another Look at the Design Features of Human Language", Current Anthropology, Volume 46, Number 4, August-October 2005 -- free preprint) can plausibly think that

... the evidence suggests that Pirahã lacks embedding altogether.

even though he presents and discusses examples that he translates as "When I finish eating, I want to speak to you"; "If it rains, I will not go"; "I want the shirt that Chico sold"; "The woman wants to see you"; "He knows how to make arrows well"; "I said that Kó'oí intends to leave"; "There are two big red airplanes"; and so on.

Now, Dan is talking about recursive embedding, not simple modification or combining words into simple clauses. But even so, the most interesting thing about this claim, in my opinion, is that it imposes a lot fewer constraints on what the Pirahã can say than you might think.

Here are a couple of examples with his discussion:

ti gái -sai kó'oí hi kaháp -ií
I say -nominative Kó'oí he leave -intention
"I said that Kó'oí intends to leave."
(lit. "My saying: Kó'oí intend-leaves.")

The verb "to say" (gái) in Pirahã is always nominalized. It takes no inflection at all. The simplest translation of it is as a possessive noun phrase "my saying," with the following clause interpreted as a type of comment. The "complement clause" is thus a juxtaposed clause interpreted as the content of what was said but not obviously involving embedding. Pirahã has no verb "to think," using instead (as do many other Amazonian languages [see Everett 2004]) the verb "to say" to express intentional contents. Therefore "John thinks that ..." would be expressed in Pirahã as "John's saying that. ..."

A similar construction in Elmorese would be something like "My opinion, he's gonna leave".

English complement clauses of other types are handled similarly in Pirahã, by nominalizing one of the clauses:

kahaí kai -sai hi ob -áa'áí
arrow make -nominative he see -attractive
"He knows how to make arrows well."
(lit. "He sees attractively arrow-making.")

There are two plausible analyses for this construction. The first is that there is embedding, with the clause/verb phrase "arrow make" nominalized and inserted in direct-object position of the "matrix" verb "to see/know well." The second is that this construction is the paratactic conjoining of the noun phrase "arrow-making" and the clause "he sees well."

Dan gives some arguments (clitic agreement and so on) for the "paratactic conjoining" theory -- but what is this "paratactic conjoining" anyhow?

The American Heritage Dictionary says that parataxis (contrasted with syntaxis) is

NOUN: The juxtaposition of clauses or phrases without the use of coordinating or subordinating conjunctions, as It was cold; the snows came.
ETYMOLOGY: Greek, a placing side by side, from paratassein, to arrange side by side : para-, beside; ... + tassein, tag-, to arrange.

The OED's gloss allows "connecting words" in general to be left out, not just conjunctions:

The placing of propositions or clauses one after another, without indicating by connecting words the relation (of coordination or subordination) between them, as in Tell me, how are you?.

But there's a tricky point here: in parataxis, is the (contextually apparent) relation between the phrases really there, but just not overtly expressed? Or is it in some sense not there at all?

Another way to put this is to ask whether we're talking about the sentence structure or discourse structure. Most linguists think that the rhetorical structure of a coherent discourse is not encoded in the same way that phrasal structure is. On this view, a story may have a beginning, a middle and an end, but this is a different kind of structure from the organization of a clause into a subject, a verb and object. From the listener's point of view, you might say that syntactic structure is part of the evidence you use to make sense of a sentence, while rhetorical structure is part of the result you get when you've succeeded in making sense of a discourse.

There are a lot of different ideas about what rhetorical structures are like once you figure them out (see here and here and here for some examples), but everyone seems to agree that they can be recursive. For example, a story can contain another story, or an elaboration can contain another elaboration, as readers of this weblog have reason to know.

However, there's a sort of gray area in between sentential syntaxis and discourse parataxis. In English noun compounds like [[sickle cell] anemia] vs. [rat [bile duct]], we assume that there's really a (syntactic) structure there, even though there are no words or other signs that make the relationship explicit. (Well, there's a stress difference in this case, but never mind that for now.) So couldn't there be a similar implicit relationship between apparently "paratactic" words and phrases, at least in some cases?

The reported speech of Elmore Leonard's characters, especially the lower-class ones, is full of concatenated phrases where this question comes up, because the semantic connection between the phrases is clear in context, but is not made explicit.

Sometimes the implied connection is temporal ("When X, Y"), as in this example from Mr. Majestyk:

"We get here," Larry Mendoza said, "this guy's already got a crew working."

Sometimes it's conditional ("If X, Y"), as in this example from the same book:

"Listen," Renda said, "we get to a phone we're out of the country before morning."

There are also examples of juxtaposed noun phrases whose connection is left implicit, as in these two examples (again from Mr. Majestyk):

"All right, I call some more friends. They get us out of the the country, some place no extradition, and wait and see what happens."

"That goddam truck of his, he can go anywhere," Renda said. "He told me, he comes up here hunting."

This is all perfectly grammatical vernacular American English, in my opinion; I talk this way myself, in some kinds of casual conversation.

These examples feel similar to Everett's examples of Pirahã parataxis, sharing the property of implying, via juxtaposition, phrasal relationships that English in other styles encodes via explicit clausal or phrasal embedding. Here's how Pirahã does temporal clauses:

kohoai -kabáob -áo ti 'ahoai -soog -abagaí
eat -finish -temporal I you speak -desiderative -frustrated initiation
"When [I] finish eating, I want to speak to you."
(lit. "Eating finishes, I you speak-almost-want")

There is almost always a detectable pause between the temporal clause and the "main clause." Such clauses may look embedded from the English translation, but I see no evidence for such an analysis. Perhaps a better translation would be "I finish eating, I speak to you.

A lot like Elmorese, except that the Pirahã examples are often more explicit about the semantic relationships, as indicated in this case, for example, by the completive and temporal morphemes on eat.

So let's return to the example I started with. Maurice juxtaposes three clauses

[You ever see it] [they take it out of the shell] [you wouldn't eat it].

whose relationship might have been made explicit by adding an "if" and a "when"

If you ever saw [a conch] when they take it out of the shell, you wouldn't eat it.

Once the relationships are made explicit like that, we've arguably got a recursive sentence, since the structure is something like this:

What about the way Maurice said it? Are his three clauses organized in a recursive syntactic structure:

or just a paratactic juxtaposition:

I'm not sure. A lot seems to hinge on the answer, at least in the case of Pirahã. Is "paratactic juxtaposition" like stringing sentences together in a discourse, or is it like combining words and phrases in a sentence? Or are those two alternatives really just versions of the same thing seen from different disciplinary angles? Those are deep questions, to be answered by people who know more about syntax and discourse than I do.

In any case, I'm confident that HCF are now among those who make a sharp distinction between syntax and discourse structure, and I'm equally sure that "recursion", for them, is a matter of syntax. Chomsky has famously been skeptical for decades about whether discourse coherence is even a problem amenable to rational investigation, as opposed to one of the mysteries that "lie beyond the reach of the form of human inquiry that we call 'science'" ("Problems and Mysteries in the Study of Human Language," Reflections on Language pp. 137-227, 1975). And the recent animal experiments are all about learning "grammatical" constraints on short sequences of uninterpreted sounds, a situation where discourse structure simply doesn't arise. HCF are strictly concerned with recursion in the syntactic structure of sentences, not in the interpreted structure of discourses.

That's why Dan Everett's claim that Pirahã lacks (recursive) syntactic embedding is important. If some human languages (and Pirahã is not the only candidate) lack recursion, then it's hard to see how recursion could be the defining characteristic of human language. I tried to make this point in a humorous way last year ("Homo Hemingwayensis"). So Tecumseh Fitch is heading to Amazonia this summer, according to Elizabeth Davies' May 6, 2006 article in The Independent:

Professor Everett ... will head back to Amazon this summer with a bevy of enthusiastic young PhD students to try to introduce others to the Pirahã and to prove his theories. A mark of how seriously the linguistic world takes his studies is that accompanying him will be W Tecumseh Fitch, one of the three architects of the original theory of universal grammar along with Chomsky and Dr Marc Hauser. The expert is keen to see whether the tribe does indeed refute their long-established theory.

(Given that the theory in question (that human language in the narrow sense is only recursion) was first proposed in 2000, and immediately called into question by Pinker and Jackendoff among others, the modifier "long-established" is a bit of a stretch here. "Universal grammar" is a term with a longer and broader history -- but this post is not yet another critique of linguistic journalism...)

Anyhow, the article doesn't tell us what Fitch is going to be doing. Probably not analyzing the sentence and discourse structures of Pirahã, since that's not the kind of stuff he does. I'd guess that he'll be testing the Pirahã people on the sorts of acoustic novelty-detection tasks that Hauser and Fitch 2004 applied to cotton-top tamarins and Harvard undergraduates. If he can get them to approach the task the way he wants, I'll look forward to learning the results. But I'd also really like to know when parataxis is really covert syntaxis, and what sort of embedded linguistic structures the Pirahã actually use.

[See David Beaver's recent post "And people say we monkey around" for more on the question of concatenated signals from animals.]

[And yes, I'm aware that some of Elmore Leonard's paratactic juxtapositions could alternatively be analyzed in terms of "prosiopesis", Otto Jespersen's term for starting to talk without putting your mouth in gear, e.g. "[with] that goddam truck of his, he can go anywhere", "[in] my opinion, he's gonna leave." But I think there's plenty of evidence that this is not the right story, in general.]

[Update: Dan Everett writes that

This looks great. You have made this all much clearer than I have. So I will happily borrow from you in future discussions.

The New Yorker is going to be doing an in-depth article on this stuff. [...] What you have to say on this will be very helpful.

Actually, what Tecumseh will be doing is checking for recursive reasoning, which I am quite confident that the Piraha have. I see this as independent of their syntax, though, whereas he does not.

I'm encouraged that Dan didn't find any howlers, though I'm still not sure of my grasp of the logic of this question. His description of Fitch's project does suggest that my characterization of Fitch's views is wrong -- if "recursive reasoning" is the same as recursive embedding in syntax, then I suppose that rhetorical structure imposed on paratactic juxtaposition of phrases would count for Fitch as "recursion".]

[In a later note, Dan confirms that the Pirahã do have recursive narrative structures. A simple example would be interpolating a digression into the plot of a story. The same thing of course is also true of even the most paratactic writers. ]

Posted by Mark Liberman at May 19, 2006 12:04 AM