The New York Times, May 2, 2006
The warbles and rattles of a starling seem innocuous enough. But starlings are now the object of a fierce debate about the nature of language.
In the current issue of Nature, scientists report that starlings recognize song patterns based on rules of the sort that make language possible. Their paper has drawn sharp reactions pro and con from linguists and animal communication experts.
The debate is over what, if anything, the results mean for human language. Some scientists believe that the findings offer new clues to how it evolved. Others dismiss the notion.
Human language is unique in the world of animal communication. Humans can convey an infinite range of ideas with a limited vocabulary, because they are not limited to strings of disconnected sounds. Humans can generate meaning by combining words in various ways, building them into clauses and inserting those clauses into sentences.
It is possible to come up with all sorts of rules for stringing symbols together. In the 1950’s, Noam Chomsky of the Massachusetts Institute of Teachnology ranked classes of rules by their power. A very simple rule might call for one word always to be followed by another word.
Dr. Chomsky argued that the rules that govern language must be more powerful. At the very least, they must let people embed smaller groups of words in larger ones again and again — a process sometimes called recursion.
In 2002, Dr. Chomsky, with Marc D. Hauser of Harvard and W. Tecumseh Fitch of the University of St. Andrews in Scotland, wrote a paper about what all this meant for the evolution of language. Language evolution is vexing, they argued, because it relies on so many things that are not specific to language.
“Memory is critical to language, because if you couldn’t keep in mind several pieces of a sentence, you couldn’t understand anything,” Dr. Hauser said in an interview. “But memory’s not specific to language.”
Whether any feature was unique to language and to humans remained an open question. In their paper, Dr. Hauser and his colleagues suggested that only one thing might fit this bill: the ability to link words and meaning with the help of recursion.
To test this claim, Dr. Hauser and Dr. Fitch created simple artificial languages. The words in the languages were made up of two categories of words. In one category were short sounds made by men, such as “mo” and “li.” The other category were the same sounds made by women. The difference in the pitch of their voices made it easy to tell words from the two categories apart.
Dr. Fitch and Dr. Hauser then combined these sounds according to two different rules. One was a simple rule that a female sound should always be followed by a male sound.
Then they built sentences with a more complex rule that embedded a female-male sound pair within another pair. Let A and B stand for the two kinds of words. The simple rule produces sentences such as ABAB. The complex rule produces AABB.
As the scientists reported in a 2004 paper, humans figured out both patterns. College students listen to 30 sentences and then were tested on new ones. More than 80 percent could correctly say whether a sentence matched the pattern.
The scientists then tested cotton-top tamarin monkeys. In the evening, they played recordings of one kind of the pattern to the tamarins. The next morning they tested the monkeys, noting when the tamarins looked toward a speaker that played sentences that violated the patterns. The tamarins noticed when the simple ABAB rule was violated. But they failed to recognize patterns with the complex AABB rule.
Dr. Fitch and Dr. Hauser concluded that there was a ceiling on the ability of tamarins to recognize patterns in sounds. But they hesitated to draw broad conclusions, calling instead for tests on other animals.
Enter the starlings.
The songs of male starlings are made up of warbles, rattles, whistles and other sounds, collectively known as motifs. “They learn new motifs and embed them in their songs,” said Timothy Gentner, a neurophysiologist at the University of California, San Diego. Starlings can also recognize other individuals by learning the unique motifs used by each bird.
These skills, Dr. Gentner decided, made starlings a perfect choice for an experiment. He teamed up with three psychologists from the University of Chicago to see which patterns the birds could recognize.
“We said, ‘Well, if there’s any species that’s capable of doing this, it’s starlings,’ ” Dr. Gentner said.
Dr. Gentner and his colleagues built an artificial language from warbles and rattles. They constructed some songs according to the simple ABAB rule and others according to the complex AABB rule.
The starlings trained by listening to songs. They had to peck at a hole if a song had the right pattern, and do nothing if it did not. If they chose correctly, the birds got food; if they chose wrong, the lights went out briefly.
It took as many as 40,000 trials, but 9 of 11 starlings learned to recognize the complex AABB pattern over 90 percent of the time. They could even recognize it when three pairs of warbles and rattles were inserted between an original pair of warbles and rattles.
“They can do this, and they do it with a high degree of proficiency,” Dr. Gentner said.
He is confident that he can rule out the possibility that they are using a very simple strategy, like paying attention to whether a warble was followed by another warble. If that were true, the birds would react the same way to an AABB pattern and to AAABB, which violates the rule.
“I was surprised by how well they did when we threw really hard challenges at them,” Dr. Gentner said.
Gary Marcus, a New York University psychologist who wrote a commentary on the research for Nature, praised it highly. “I thought it was a cool paper,” he said. “The interest here isn’t that it’s the only species to do this, but rather that it’s the first. It opens the door to see how widespread this ability is.”
It will be important to see if tamarins or other primates can pass the test with training, Dr. Marcus said. It might turn out that this ingredient for language already existed in our distant ancestors. The evolution of language might be “a reconfiguration of old parts,” Dr. Marcus said. “So studies like this show us what those old parts might be.”
Some linguists strongly disagree. “I’m not buying it,” said Geoffrey Pullum, a linguist at the University of California, Santa Cruz, and co-author of The Cambridge Grammar of the English Language.
For one thing, he said he doubted that the rules the scientists used to build songs had much to do with the important features of human language. Dr. Pullum also argued that the tamarin and starling experiments used “sentences” that were too short and simple to detect any thought process involved in grammar.
“It’s purely about bird abilities, I think, and not about the foundations of human abilities,” he said.
Dr. Chomsky also rejects Dr. Gentner’s conclusions. He suggests the starlings are merely counting rattles, storing the number in their memory, then counting warbles. “It has nothing remotely to do with language — probably just with short-term memory,” he said via e-mail. Dr. Gentner argues that even if the starlings are counting, they are still using a strategy more sophisticated than has been seen before in animals.
“Chomsky may find this trivial, but that is a bit like saying apes use tools, but only the trivial kind that lack the sophistication of a tri-square or a laser level,” he said.
Dr. Hauser added: “This shows a capacity that goes way beyond what we showed with tamarins. That’s what makes it an important paper.”
He is intrigued that starlings can recognize patterns even though the new patterns create no new meaning. “It’s still, ‘I’m Fred, I’m a male,'” he said. In humans, this pattern recognition is linked not just to speech, but to meaning.
Dr. Gentner said, “It’s that interface between meaning and pattern where we humans really excel.”
Copyright 2006 The New York Times Company. Reprinted with permission.