Starlings learned to differentiate between a regular birdsong "sentence" and one containing a clause or another sentence of warbling, according to a study in Thursday's journal Nature.
It took Tim Gentner, a psychology researcher at the University of California in San Diego, a month and about 15,000 training attempts, with food as a reward, to get the birds to recognise the most basic of grammar in their own langauge.
Yet what they learned may shake up the field of linguistics.
While many animals can roar, sing, grunt or otherwise make noise, linguists have contended for years that the key to distinguishing language skills goes back to our elementary school teachers and basic grammar.
Sentences that contain an explanatory clause are something that humans can recognise, but not animals, researchers figured.
Two years ago, a top research team tried to get tamarin monkeys to recognise such phrasing, but they failed.
The results were seen as upholding linguist Noam Chomsky's theory that "recursive grammar" is uniquely human and key to the facility to acquire language.
Noam Chomsky says recursive
grammar is uniquely human
But after training, nine out of Gentner's 11 songbirds picked out the bird song with inserted warbling or rattling phrases about 90% of the time.
Two continued to flunk grammar.
"We were dumbfounded that they could do as well as they did," Gentner said.
"It's clear that they can do it."
Gentner trained the birds using three buttons hanging from a wall.
When the bird pecked the button, it would play different versions of bird songs that Gentner generated, some with inserted clauses and some without.
If the song followed a certain pattern, birds were supposed to hit the button again with their beaks; if it followed a different pattern they were supposed to do nothing. If the birds recognised the correct pattern, they were rewarded with food.
"Some of the cognitive sources that we deploy may be shared with other animals"
Director of Harvard University's Cognitive Evolution Laboratory
Gentner said he was so unprepared for the starlings' successful learning that he had not bothered to record the songs the starlings sang in response.
"They might have been singing them back," Gentner said.
To put the trained starlings' grammar skills in perspective, Gentner said they don't match up to either of his sons, ages 2 and 9 months.
Jeffrey Elman, a professor of cognitive science at UCSD who was not part of the Gentner research team, said that what the experiment shows is that language and animal cognition is more complicated than scientists thought and that there is no "single magic bullet" that separates man from beast.
Marc Hauser, director of Harvard University's Cognitive Evolution Laboratory, who conducted the tamarin monkey experiment, said Gentner's study was important and exciting, showing that "some of the cognitive sources that we deploy may be shared with other animals".
But Hauser said it still doesn't quite disprove a key paper he wrote in 2002 with Chomsky.
The starlings are grasping a basic grammar, but not the necessary semantics to have the language ability that he and Chomsky wrote about.