Artificial intelligence (AI) is shedding new light on how animals communicate, offering insights into the complex ways species across land, sea, and sky interact with one another. From dolphin whistles to elephant rumbles and birdsong, animal vocalizations contain patterns and structures that convey information—patterns that are often too subtle for humans to detect. AI’s ability to identify and analyze these patterns is opening new doors for biologists and computer scientists.

Recent studies using AI have uncovered surprising revelations. African savannah elephants (Loxodonta africana) and common marmoset monkeys (Callithrix jacchus) have been found to assign names to their companions. Researchers are also employing machine learning to decode the vocalizations of crows. As AI tools become more advanced, they could unlock new understanding of animal communication, self-awareness, and even inspire stronger conservation efforts.

Despite these advances, David Gruber, a marine microbiologist and founder of the Cetacean Translation Initiative (CETI), cautions that we are far from developing an “animal Google Translate.” Unlike human languages, which have vast datasets of known meanings, animal communication lacks this foundational understanding. Gruber explains, “It’s a big assumption to think we can simply apply human-language AI models to another species and expect translation.”

CETI focuses on decoding the communication of sperm whales. Before CETI’s involvement, marine biologist Shane Gero spent years studying sperm whales through the Dominica Sperm Whale Project, which documented over 30 whale families near the Caribbean island. Sperm whales use echolocation clicks to hunt in deep, dark ocean waters and communicate through unique sequences of clicks known as codas.

These codas, which range from 3 to 40 clicks, serve different purposes depending on context. Underwater, they help maintain contact among whales; at the surface, they facilitate social interaction. Gero’s research has revealed that whales form clans with distinct diets, social behaviors, and habitats. Each clan communicates in a unique dialect, marked by specific tempos and pauses in their codas, creating cultural boundaries between groups.

Initially, researchers manually analyzed whale sounds using spectrograms, visual representations of sound that display characteristics like volume and frequency. This method allowed them to identify individual clicks, but it was labor-intensive—each minute of audio required 10 minutes of analysis. Machine learning vastly accelerated this process, enabling researchers to assign clicks to specific whales and explore patterns at the sentence and conversation levels.

Using AI, the team discovered nuanced features in whale vocalizations. For example, slight changes in the timing of clicks—termed ‘rubato’—and the occasional addition of extra clicks, called ‘ornamentation,’ mirror musical techniques that add expressiveness. These discoveries hint at a more intricate communication system than previously understood. By analyzing over 8,700 codas, researchers identified what they describe as a “sperm whale phonetic alphabet” that might serve as the foundation for complex information sharing.

The significance of these features is still being studied. For instance, does rubato increase before a dive or decrease during maternal communication? Gero’s team is exploring these questions to uncover the deeper meanings behind whale vocalizations. “If you don’t know rubato exists, you can’t start asking when it’s important,” he says.

As AI continues to decode the calls of the wild, it holds the potential to transform our understanding of animal communication and behavior, while encouraging greater efforts to protect these remarkable creatures.

Source