Smart Starlings

News of advances or challenges in the evolution of speech typically concern just one school of thought. Such was the case with a brouhaha that arose late last spring over the syntactical abilities of starlings. Many linguists believe syntax defines the distinctive nature of language. These syntax firsters were in an uproar. Most other people  could only wonder what they excitement was all about. Some birds learned to discriminate between meaningless sounds. Is that relevant to language?

The fuss began with Stephen Pinker’s book The language Instinct (1994). It argued that the brain has modules specifically for organizing words into sentences; however, doubters wondered, are we really supposed to believe that there was a strong selective pressure lasting for many generations that favored proper syntax when even today people regularly misspeak without suffering serious consequences?

Problems like these have forced a revision in ideas about the biology of syntax. Instead of looking for many language-only modules, many linguists, including Noam Chomsky, concentrate on the evolution of the ability to form recursive sentences.

Recursive rules repeat themselves. In a recursive procedure a system follows a rule and then can then follow the rule again. Take, for example, this bit of fundamental news:

A man bit a dog.

We can break that sentence into two parts:

• a noun phrase (a man) and
• a verb phrase (bit a dog).

We can break that verb phrase into two more parts:

• a verb (bit) and
• another noun phrase (a dog).

Recursion is the ability to keep on putting the same structural components (like noun phrases) into a single whole. We can say that sentences may be formed by generating a noun phrase + a verb phrase and we can make more complicated sentences without adding new rules. We can simply expand the verb phrase so that it contains yet another a verb phrase (verb + another noun phrase): A man bit a dog and banged his nose.

You can wonder if the infinitely long sentences that can be produced by such recursion are truly English, but Chomsky insists that recursive syntax is what enables people to utter an endless variety of new sentences and in 20002 he put that position in stone. The journal Science published an article by Marc Hauser, Noam Chomsky and Tecumseh Fitch titled “The Faculty of Language: What Is It, Who Has It, and How Did It Evolve?”(Abstract here.) The article distinguished between the “broad” faculty of language and a more “narrow” faculty.

The broad faculty covers many of the areas required for producing speech sounds, having a meaning to express, and organizing the sounds syntactically. It could have evolved in animals and predates language.

The narrow faculty is the part that is uniquely human. It consists of the ability to generate an infinite number of sentences from a finite set of recursive rules.

The Hauser-Chomsky-Fitch thesis was that this recursive ability is THE one feature of language that is unique to humans. Their article included an experiment in which humans heard examples of  meaningless“sentences” generated by two simple sets of rules. One of the rules was recursive, one was not. The human listeners were able to learn both patterns and recognize which rule set was being used. Cotton-top tamarin monkeys, however, did not notice mistakes in the recursive patterns.

The starling test that got so much attention challenged the Hauser-Chomsky-Fitch findings. The research appeared in an article published last April in Nature by Timothy Gentner, Kimberley Fenn, Daniel Margoliash and Howard Nusbaum, “Recursive syntactic pattern learning by songbirds.”

They tried to repeat the Hauser et al experiment by using starlings instead of cotton-top tamarins. It took a great deal of effort, over 12,000 exposures for even the quickest learning birds, but four out of eleven starlings generalized their responses well enough to respond appropriately to new songs generated by recursive rules while ignoring the new songs that did not.

A teenage blogger on MySpace.com asked Chomsky for a comment. Chomsky responded that the training had not involved recursion at all:

... not understanding the elementary mathematics, [the experimenters] concluded that starlings were learning context-free systems, hence on their way to language. Amazingly, an article with such an elementary fallacy made it to "Nature," a serious scientific journal.

Even so, the starling experiment has thrown the syntax-firsters on the defensive. Having surrendered so much ground to the “broad language faculty,” they have bet the bank on a narrow, recursive faculty. Meanwhile, others schools of thought roll untouched by the news. Dr. Gentner, the lead author of the starling paper, told the New York Times (available here($)), ''It's that interface between meaning and pattern where we humans really excel.'' It is how you mix meaning and grammar that makes language so intriguing.

Comments

[Jamaal]

Aren't you just perpetuating this story about starlings understanding recursive CFGs with this article? Chomsky stated the facts clearly. They discovered, yet again, that animals can 'count' (subitize). It might be interesting to see how long a bird can keep a number in its head, but not to people examining language. Just because you can understand one rule which can be implemented in a CFG doesn't mean you understand as a CFG.

BTW, great blog... I'm loving the way that this is starting.

-----------------

THE BLOGGER RESPONDS

Jamaal:

Thanks for the good words. You’re plainly an informed commenter. What do you think? Was the experiment in Hauser-Chomsky-Fitch unambiguous in its demonstration of recursive thinking among humans or might there have been some other explanation there too?

By the way, for visitors who are not as well informed on the jargon, CFG refers to context-free grammar. That’s a formal grammar that allows for the production of recursive sentences.

[tim]

I haven't read the Gentner et al. article, but an experiment that needs to use 12,000 trials to get an animal to do something seems somewhat dodgy to me - without knowing their procedures, the animals may well be picking up some other cue to recursion/non-recursion. Regardless, Chomsky's position would be that we are predisposed to detecting recursion (especially considering all that stuff about FOXP2, and in the article they point to the statistical learning paradigm of Jenny Saffran as a learning mechanism for recursion). That it takes 12,000 trials for a starling to distinguish recursion indicates that the starling is probably not predisposed in the same way that we are.

With that in mind, I do have problems with HCF's position. Firstly, they argued that recursion is not only unique to humans but unique to language, that only language uses that module. This something which is clearly negated by the presence of music and dance in humans, which are also recursive but not language. Something else that comes out quite clearly when you consider Pinker and Jackendoff's reply to the paper (and HCF's reply to that) is that HCF believe that recursion's uniqueness to humans is very good evidence of language being evolved. It could, however, simply be that it's not the presence of recursion as such that makes language language, nor the presence of certain grammatical formations (and so on, see P&J's reply for a more detailed list) - but instead the combination of all of these things. Seems like no other animal is capable of all of the other components of language in quite the same way we are, though I'm sure that some animal somewhere is capable and predisposed to each individual aspect of language in some way.

tim.

[TLTB]

Actually, HCF do not argue that recursion is unique to the language faculty. Quite the opposite. They are very clear about stating that they think recursion is behind the ability to count natural numbers to infinity and weakly suggest it underlies music composition and comprehension as well.

As for the experiments on starlings and tamarins, the experiments of Hauser do not prove once and for all that nonhuman primates don't have recursion, but then again they don't have to. In light of the claim that only humans have recursion, the onus is on experimenters to show that any other animal has it as well. As noted, the starling experiment doesn't hold up.

The definition of recursion as understood in CHF is that a recursive systems yields an output of discrete infinity. The fact that we can construct (in principle) infinitely long sentences and infinitely embedded structures, as well as the fact that we can count to infinity, is proof we have recursion.

One development that has occured since the article was published, is that the claim that nonhuman primates do not have an instinctive ability to immitate has been disproven, at least with regard to facial movements (though not yet with regard to vocalizations). You can find that article here.

http://biology.plosjournals.org/perlserv/?request=get-document&doi=10.1371/journal.pbio.0040302

----------
THE BLOGGER RESPONDS

“The fact that we can construct (in principle) infinitely long sentences and infinitely embedded structures, as well as the fact that we can count to infinity, is proof we have recursion.”

There is the numb of the argument and I have to confess I have never been persuaded of the truth of this orthodox position, at least as far as it stands about syntax. No editor would accept a sentence like, “I don’t believe that you don’t believe that he doesn’t believe that a middle-aged man who bit a mangy dog covered in strange fleas contracted from a visit to a passing three-ring circus where imported, Asian elephants bellowed at the unhappily married and sadly childless Mahout, also bumped his nose.” The limit to understanding this sentence is precisely, memory. We get lost and forget what was being discussed.

This is a good example of the syntax vs semantics conflict. If you think syntax is all, then the notion of an infinite sentence isn’t troubling. If meaning and understanding are part of your idea of language, the idea becomes nonsensical.

[TLTB]

As you say, the problem with your sentence is memory, but arguably that is unrelated to linguistic ability. To deny that possibility seems to put an unecessary limit on research goals, and I can't think of a good reason to suppose that our on-line memory capacity should say anything about our ability to produce infinitely long sentences.