The Birds and the B’s
Challenging Chomsky, Starlings Learn ‘Human-Only’ Syntax Patterns
By Inga Kiderra
The European starling – long known as a virtuoso songbird and as an expert mimic too – may also soon gain a reputation as something of a “grammar-marm.”
This three-ounce bird, new research shows, can learn syntactic patterns formerly thought to be the exclusive province of humans.
Led by Timothy Q. Gentner, assistant professor of psychology at the University of California, San Diego, a study published in the April 27 issue of Nature demonstrates that starlings have the capacity to classify acoustic sequences defined by recursive, center-embedded grammars.
Recursive center-embedding refers to the common characteristic of human grammars that allows for the creation of new (and grammatically correct) utterances by inserting words and clauses within sentences – theoretically, without limit. So, for example, “Oedipus ruled Thebes” can become “Oedipus, who killed his father, ruled Thebes” or “Oedipus, who killed his father, whom he met on the road from Delphi, ruled Thebes,” and so on.
Chomskian linguists have held that this recursive center-embedding is a universal feature of human language and, moreover, that the ability to process it forms the computational core of a uniquely human language facility.
“Our research is a refutation of the canonical position that what makes human language unique is a singular ability to comprehend these kinds of patterns,” Gentner said. “If birds can learn these patterning rules, then their use does not explain the uniqueness of human language.”
The research also contradicts the 2004 conclusions of W. Tecumseh Fitch and Marc D. Hauser based on the failure of cotton-top tamarin monkeys to learn similar grammar patterns after being exposed to strings of human speech.
“This result reinvigorates the search for the evolutionary substrates for language processing among the primates, and most excitingly gives us an animal model to probe deeper into these aspects of language,” said Daniel Margoliash, who is a coauthor along with Kimberly M. Fenn and Howard C. Nusbaum at the University of Chicago.
Gentner and his coauthors created artificial starling songs that followed two different patterning rules: a “context-free” rule, which allows for a sound to be inserted in the middle of an acoustic string and is the simplest form of recursive center-embedding; and a “finite-state” rule, of the sort thought to account for all non-human communication, that allows for sounds to be appended only at the beginning or end of a string.
Capitalizing on the diverse range of sounds in starling songs, the researchers used recordings of eight different “warbles” and eight different “rattles” from a single male starling to construct a total of 16 artificial songs. Eight of these songs followed the context-free sequence AnBn (i.e. AABB or rattle-rattle-warble-warble) and eight followed the finite-state (AB)n (i.e. ABAB or rattle-warble-rattle-warble).
Eleven adult birds were then taught to distinguish these two sets of songs using classic reinforcement techniques. The birds were rewarded with food for pecking at a button when they heard a song from the context-free set and for refraining when they heard one from the finite-state set.
Natural starling song
(ppt w/embedded audio)
Male starlings are prodigious singers, producing some of the longest and most complex bouts of songs among the 4,500 different species of song birds. As male starlings age, they continue to add new sounds, called motifs, to their repertoires – picking up nearly any auditory object they can find, including the sounds of car alarms, squeaky doors, human speech and the vocalization of other species. Over time, their motif repertoire grows and their songs get longer and longer. Females, who also sing, are more attracted to these longer songs and preferentially mate with the males that sing them.
Courtesy TQ Gentner
Nine of the starlings – after 10,000 to 50,000 trials over several months – eventually learned to distinguish the patterns.
When tested with different combinations of rattles and warbles that followed the same rules, the starlings performed well above chance, suggesting they had learned the abstract patterns and not just memorized the specific songs.
The researchers also checked to see how the birds responded to “ungrammatical” strings, ones that violated the established rules. The starlings treated these differently, as expected if they had learned the patterns.
The experimenters then asked if the birds were capable of a key feature of human grammars: Could the starlings extrapolate these patterning rules to distinguish among longer strings? Remarkably, Gentner said, after learning the patterns with songs made up of pairs of rattles and warbles, the birds were able to successfully recognize grammatical strings of three rattles-three warbles and four rattles-four warbles, where the new rattles and warbles had been inserted in the middle.
The finding that starlings can grasp even these simple grammatical rules, Gentner said, suggests that humans and other animals share basic levels of pattern recognition and also hints at the likelihood of other cognitive abilities we have in common.
“There might be no single property or processing capacity,” Gentner and coauthors write, “that marks the many ways in which the complexity and detail of human language differs from non-human communication systems.”
Tim Gentner and his coauthors created artificial starling songs that followed two different patterning rules or grammars. Capitalizing on the diverse range of sounds in starlings' natural songs, the researchers used recordings of different “warbles” and different “rattles” from a single male starling to construct the artificial songs. The starlings were taught to distinguish between sets of grammars and songs using classic reinforcement techniques. They were rewarded with food for pecking at a button when they heard a song from the context-free set and for refraining when they heard one from the finite-state set.
Experimental/Artificial Song (audio)
Courtesy TQ Gentner
More generally, Gentner says, “The more closely we understand what nonhuman animals are capable of, the richer our world becomes. Fifty years ago, it was taboo to even talk about animal cognition. Now, there are Nova specials on the subject and no one doubts that animals have complex and vibrant mental lives. This study is a powerful statement about what even birds can do: Look at what they’re learning.”
The experiments were performed at the University of Chicago when Gentner was a postdoctoral researcher there. The study is supported in part by a National Institutes of Health grant to Margoliash.
Gentner is now exploring the possibility of working with juvenile starlings, who (like human children) may be more adept than their full-grown counterparts at learning syntactical patterns.
A current study is also underway to see if starlings can apply the rules they’ve learned to novel stimuli – something humans are masters at doing. Early results suggest they don’t, leaving humans the reigning champs of generalizing from patterns.
Media Contacts: Inga Kiderra (858) 822-0661