Whales use complex communication systems we still don’t understand, a trope exploited in sci-fi shows like Apple TV’s Extrapolations. That show featured a humpback whale (voiced by Meryl Streep) discussing Mahler’s symphonies with a human researcher via some AI-powered inter-species translation app developed in 2046.

We’re a long way from that future. But a team of MIT researchers has now analyzed a database of Caribbean sperm whales’ calls and has found there really is a contextual and combinatorial structure in there. But does it mean whales have a human-like language and we can just wait until Chat GPT 8.0 to figure out how to translate from English to Sperm-Whaleish? Not really.

One-page dictionary

“Sperm whales communicate using clicks. These clicks occur in short packets we call codas that typically last less than two seconds, containing three to 40 clicks,” said Pratyusha Sharma, a researcher at the MIT Computer Science and Artificial Intelligence Laboratory and the lead author of the study. Her team argues that codas are analogues of words in human language and are further organized in coda sequences that are analogues of sentences. “Sperm whales are not born with this communication system; it's acquired and changes over the course of time,” Sharma said.

Seemingly, sperm whales have a lot to communicate about. Earlier observational studies revealed that they live a fairly complex social life revolving around family units forming larger structures called clans. They also have advanced hunting strategies and do group decision-making, seeking consensus on where to go and what to do.

Despite this complexity in behavior and relationships, their vocabulary seemed surprisingly sparse.

Sharma’s team sourced a record of codas from the dataset of the Dominica Sperm Whale Project, a long-term study on sperm whales that recorded and annotated 8,719 individual codas made by EC-1, a sperm whale clan living in East Caribbean waters. Those 8,719 recorded codas, according to earlier research on this database, were really just 21 coda types that the whales were using over and over.

A set of 21 words didn’t look like much of a language. “But this [number] is exactly what we found was not true,” Sharma said.

Fine-grained changes

“People doing those earlier studies were looking at the calls in isolation... They were annotating these calls, taking them out of context, shuffling them up, and then tried to figure out what kind of patterns were recurring,” Sharma explained. Her team, by contrast, analyzed the same calls in their full context, basically looking at entire exchanges rather than at separate codas. “One of the things we saw was fine-grained changes in the codas that other whales participating in the exchange were noticing and reacting to. If you looked at all these calls out of context, all these fine-grained changes would be lost; they would be considered noise,” Sharma said.

The first of those newly recognized fine-grained changes was termed “rubato,” borrowed from music, where it means introducing slight variations in the tempo of a piece. Communicating sperm whales could stretch or shrink a coda while keeping the same rhythm (where rhythm describes the spacing between the clicks in a coda).

The second feature the researchers discovered was ornamentation. “An ornament is an extra click added at the end of the coda. And when you have this extra click, it marks a critical point, and the call changes. It either happens toward the beginning or at the end of the call,” said Sharma.

The whales could individually manipulate rubato and ornamentation, as well as previously identified rhythm and tempo features. By combining this variation, they can produce a very large variety of codas. “The whales produce way more combinations of these features than 21—the information-carrying capacity of this system is a lot more capable than that,” Sharma said.

Her team identified 18 types of rhythm, three variants of rubato, five types of tempo, and an ability to add an ornament or not in the sperm whale’s communication system. That adds up to 540 possible codas, of which there are roughly 150 these whales frequently used in real life. Not only were sperm whales’ calls built with distinctive units at a coda level (meaning they were combinatorial), but they were compositional in that a call contained multiple codas.

But does that get us any closer to decoding the whale’s language?

“The combinatoriality at the word level and compositionality at the sentence level in human languages is something that looks very similar to what we found,” Sharma said. But the team didn’t determine whether meaning was being conveyed, she added. And without evidence of meaning, we might be barking up the wrong tree entirely.