What the science of baby-speak can tell us about whale songs
A new study reveals that whale song and human languages share features that make them easier to learn.
Just like popular songs on TikTok, new humpback whale songs can rapidly spread across regions and populations to replace all-too-familiar tunes. But what makes their complex vocalizations so catchy?
Humpbacks songs each consist of sound elements that are combined into phrases that are then repeated to create the themes that together form songs. This is quite similar to how human language assembles sounds into words, words into phrases, phrases into sentences, and so on. In human language, a few words are used very frequently; most are relatively rare. A study published in Science today suggests that whale songs and human language share this pattern, which probably makes them easier to learn.
The analysis is one of two out this week that point to surprising commonalities between whale songs and human language. “This fascinating works shows how general linguistic principles shape the vocalizations of evolutionary distant species,” says quantitative linguist Ramon Ferrer-i-Cancho of the Polytechnic University of Catalunya in Barcelona, who was not involved in the research.
(Explore the hidden world of whale culture.)
Infants inspire a new way to study whales
A first indication of this came when researchers recorded a few whales in the process of changing their song, seamlessly replacing specific themes by new ones with a similar arrangement of sound elements. “They were singing old and new themes smushed together, but it was really structured. This way, they’d transition nicely from one theme to the other, or replace the old theme with the new theme entirely. That was our first hint that segmentation was important,” says Ellen Garland, a marine biologist at the University of St. Andrews in Scotland and one of the authors of the study.
Garland, who is also a National Geographic Explorer, has listened to hundreds of hours of whale songs for hundreds of hours and use short pauses and characteristic sequences to identify the individual elements they’re made of. But when she saw Inbal Arnon, a psycholinguist at the Hebrew University in Israel, present her work on how infants learn to identify individual words in the stream of sound that is spoken language, they had an idea. “What if I we applied the same logic to whale song? Would that work?”
Unlock science-backed answers to your health questions
Arnon and language evolution researcher Simon Kirby of the University of Edinburgh in Scotland, had already developed an algorithm that applies similar principles to those children use to segment language. “It basically learns how likely a certain sound is to be followed by another,” Arnon explains. “Sounds that commonly co-occur are probably part of the same word. But if a sound is followed by another one that is rarely heard right after, that might be a new word.”
When Arnon and Kirby applied the algorithm to more than 30 hours of humpback songs Garland’s team had recorded near New Caledonia, the song segments it detected were reassuringly close to those Garland would have identified herself—but not quite the same. And when they compared the frequency at which these different elements occurred, a pattern emerged that is also typically found in human lanuage. The most common word—or in this case, song segment—occurred twice as often as the second most common one, thrice as often as the third most common one, and so on. Some segments are very frequent, many others quite rare.
Some earlier studies found that dolphin whistles may follow a similar pattern, but apart from those, nothing of the kind had ever been seen in whales.
Linguistics call this pattern a Zipfian distribution. While this somewhat unwieldy term may sound like an entirely made-up word, years of research using actual made-up words has revealed why it is important. Arnon’s research has shown that made-up languages with Zipfian word frequencies are easier for people to learn, while Kirby’s has found that when made-up languages are passed across multiple generations of speakers, they tend to become more Zipfian.
Initially, the researchers were so surprised to find these patterns that they wondered if they were the product of the infant-inspired algorithm itself. “I honestly had that worry all the time,” says Kirby. “I'd wake up in the middle of the night and go, ‘Oh my God, maybe it's just our method.’” To make sure it wasn’t, the team analyzed thousands of made-up datasets in which the sounds or segments had been moved around, and to their relief, none of these revealed the same pattern.
Finding these patterns in whales “suggests their communicative behavior has culturally evolved to become easier for novices to learn, echoing how the structure of our languages seems to have evolved to be more learnable by human infants,” says primatologist Andrew Whiten of the University of St. Andrews, Scotland, who was not involved in the research but coauthored an accompanying perspective article on the study.
Garland, Arnon and Kirby think that applying this algorithm to the sounds of other species that have to learn complex communication skills by listening to others, such as many songbirds, may reveal similar patterns. “As we expand our investigation of ‘linguistic laws’ with origins in human language,” says National Geographic Explorer and marine biologist Shane Gero of Project CETI, who was not involved in the study. “It begs asking if they are simply biological laws?”
(Not all whale sing. Read up on some common whale misconceptions.)
Why whales keep it short
Another study published this week in Science Advances shows humpback songs have a few other things in common with human language. One remarkable feature is that song elements the whales use more frequently tend to be shorter, says study author and cultural evolution researcher Mason Youngblood of Stony Brook University in New York, just like the words we most commonly use – “the”, “of” and “and” are all really short. “Another is that longer words or segments tend to consist of shorter elements.” These patterns are less likely to be due to ease of learning, he adds. “There’s another good reason for shortening sounds: It makes them less effortful to produce.”
That last pattern is also found in many other whales, says Youngblood, who analyzed the vocalizations of 15 other whales in addition to humpbacks. He found an intriguing exception to the rule in three elusive species of dolphin (Hector’s, Commerson’s, and Heaviside’s) that communicate using click sounds so high they are nearly inaudible to the killer whales that hunt them. These three are not using shorter sounds in longer segments, says Youngblood. “But most other whale species I’ve looked at do.” This indicates that there may be another reason to keep it short: avoiding the attention of predators.
Is whale song a language?
All of these similarities may create the impression that the complex vocalizations of humpbacks and other whales may qualify as language. Even at its most complex, however, whale song is not quite like language, says Youngblood, but perhaps more like human song, the elements of which also tend to occur at Zipfian frequencies.
Garland agrees that whale song is not language. “Only males sing, probably to attract females, repel other males, or both,” says Garland. “They're communicating something to other members of their species, yes,” she says. “But it's not like, ‘Hey, there's lots of krill over there, let’s go get it.’ They don’t refer to specific things the way humans do.”
