Sign up for CNN’s Wonder Theory science newsletter. Explore the universe with news about fascinating discoveries, scientific developments and more.
Scientists have accomplished a great feat. They have identified previously unknown complexities in whale communication by analyzing thousands of recorded sequences of sperm whale clicks with artificial intelligence.
Variations in tempo, rhythm and length of the whales’ click sequences, called codas, weave a rich acoustic tapestry. These variables indicate that whales can combine click patterns in multiple ways, mixing and matching sentences to convey a wide range of information to each other.
What sperm whales say with their clicks remains a mystery to human ears. Still, uncovering the range of whales’ vocal exchanges is an important step toward linking whale sounds to specific messages or social behaviors, the scientists reported May 7 in the journal Nature Communications.
“This work builds on much previous work aimed at understanding sperm whale calls. However, this is the first work that has begun to look at sperm whale calls in their broader communicative context and in the context of exchanges between whales, which has made some of the findings possible,” says study co-author Dr . Daniela Rus, director of MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), said in an email.
“Understanding which aspects of their codas they can control and vary can help us understand how they can encode information in their calls,” Rus said.
The researchers called their catalog of sound combinations a “phonetic alphabet” for sperm whales, comparing variations in the whales’ click sequences to the production of different phonetic sounds in human speech.
But while the team’s findings are interesting, that term offers a misleading perspective on whales’ vocal interactions, said Dr. Luke Rendell, a researcher at the University of St. Andrews in the United Kingdom, whose work focuses on communication in marine mammals. e-mail.
“The presentation of the ‘phonetic alphabet’ – it’s nothing like that,” says Rendell, who was not involved in the study.
“The way the tempo variation is used is completely different from how we use, for example, elements of an alphabet to construct linguistic expression,” he said. “There is no evidence for that, and it is not a very useful interpretation, because it forces everything into a narrow and somewhat exaggerated perspective of ‘is it like human language or not’, when there is a much wider range of interpretations available.”
Pattern recognition
Sperm whales produce their clicks by forcing air through an organ in their heads called the spermaceti, and these sounds can be as loud as 230 decibels – louder than a rocket launch and capable of rupturing human eardrums – another team of scientists reported previously in the journal Scientific Reports.
For the new study, the researchers used machine learning to detect patterns in audio data collected by The Dominica Sperm Whale Project, a repository for observations of sperm whales living in the Caribbean Sea. The recordings represented the voices of about 60 sperm whales – a subgroup of a group of about 400 whales known as the Eastern Caribbean clan – and the vocalizations were recorded between 2005 and 2018.
Previous research had identified 150 types of codas in sperm whales worldwide, but Caribbean whales used only 21 of those codas.
The scientists examined the timing and frequency of 8,719 coda sequences – in lone whale utterances, in choruses and in call-and-response exchanges between whales. When visualized with artificial intelligence, previously invisible coda patterns emerged.
The study authors defined four characteristics in codas: rhythm, tempo, rubato, and ornamentation. Rhythm describes the series of intervals between clicks. Tempo is the duration of the entire coda. Rubato refers to variations in duration across adjacent codas with the same rhythm and tempo. And embellishment is an “extra click” added to the end of a coda in a group of shorter codas, Rus explained.
These so-called ornament clicks “occur more frequently at the beginning and end of the turn” during vocal exchanges between whales, “and behave as discourse markers,” Rus said.
The discovery that whales could synchronize variations in coda tempo was “a very interesting observation,” Rendell said.
“I’m less convinced about the ‘decoration,’” he added. “It’s very rare, and I think we need more evidence that it’s not just production issues,” or filler sounds, “like when we say ‘eh’ or ‘err’.”
In total, the program detected 18 types of rhythm, five types of tempo, three types of rubato and two types of ornamentation. These coda features can all be mixed and matched to form a “vast repertoire” of sentences, the study authors reported. Moreover, the meaning could be further modified depending on the placement of a coda – which follows or overlaps other codas – within an exchange or refrain involving two or more whales.
Interactive experimentation
“In fact, many of us have been waiting for decades for advanced technology that would allow us to do something like this!” said Dr. Brenda McCowan, a professor at the University of California Davis School of Veterinary Medicine, in an email.
McCowan, who was not involved in the study, was part of another team that had an interactive “conversation” with a humpback whale in the waters off Alaska in 2021. For about 20 minutes, a curious whale responded repeatedly to a recording of a humpback whale song broadcast from the scientists’ boat.
“This particular display (with the humpback whale in 2021) was an opportunistic experiment with a curious whale that engaged us behaviorally and vocally, and entirely at our discretion,” McCowan said.
Such interactive experiments with whales, along with observations of whale behavior, could be an important part of unraveling the syntax of sperm whale click sequences, the authors wrote in the study.
Their machine learning method could also be useful for studying other types of animal vocalizations, McCowan added.
“Tempo, rhythm, rubato and ornamentation are probably found in other species of whales,” McCowan said. “We already know that this applies to the humpback whale song. But there is also evidence for these types of patterns in other aquatic, terrestrial and tree species to which this approach could be applied.”
But while this technique is useful for identifying certain aspects of communication, it’s not a Rosetta Stone, Rendell cautioned.
“Machine learning is great for finding patterns in large data sets,” he said, “but it doesn’t create meaning.”
Mindy Weisberger is a science writer and media producer whose work has appeared in the magazines LiveScience, Scientific American, and How It Works.
For more CNN news and newsletters, create an account at CNN.com