AI reveals sperm whale sounds are similar to human language


Researchers used AI to analyze sperm whale sounds, suggesting that the complexity of their calls bears a striking resemblance to human language, the first ever noticed in the animal kingdom.

Sperm whales are renowned for their complex social behaviors, group decision-making, and remarkable intelligence in problem-solving. This comes as little surprise, considering that the sperm whale boasts the largest brain of any known modern or extinct animal, weighing an average of about 7.8 kilograms – more than five times heavier than a human brain.

Sperm whales use a unique communication system characterized by sequences of short, less than a two-second long clicks, known as codas. New research published on May 7th reveals that their method of communication is far more intricate than previously believed – whale codas have structure and similar features to human vocalizations.

Past studies have typically analyzed coda repertoires by studying them in isolation without considering the interactive context in which they were produced. However, recent research highlights that whale codas are a dynamic conversation involving two or more chorusing whales.

While the exact meaning of many codas is still unknown, the findings indicate that whale communication shares similarities with human language systems like speech and musical notation, suggesting a rich potential for conveying meaning.

The results suggest that sperm whale communication could present the first-ever case of human-like communication occurring in a different species.

sperm whale
Source: Shutterstock

Whale alphabet

MIT's Computer Science and Artificial Intelligence Lab (CSAIL) spearheaded the research in collaboration with Project CETI, a nonprofit organization dedicated to harnessing AI for the study of the communication of whales.

Researchers developed the "sperm whale phonetic alphabet" to explain the variability in coda structure. The scientist used a mix of algorithms to analyze recordings of 8,719 codas from around 60 whales in the Eastern Caribbean collected by the Dominica Sperm Whale Project (DSWP) between 2005 and 2018. This dataset is the current largest sperm whale data repository.

whale communication
Sperm whales communicate through sequences of clicks. In panel A, we see a two-minute interaction between two whales, with clicks represented in blue and orange. Panel B displays these clicks on a time–time plot, where the horizontal axis shows the time within the exchange when a click occurs, and the vertical axis shows the time of the click relative to the first click in the sequence. This vertical axis acts like a microscope, allowing us to examine the internal structure of each sequence of clicks, or "coda." Panel C provides a time–time visualization for the entire two-minute exchange, with lines connecting corresponding clicks between adjacent codas. This reveals intricate variations in coda structure that depend on the context of the communication. Source: Research paper

Analysis of whale communication showed that the clicks made by sea mammals are not random but have a structure and depend on the context. The intervals between clicks change based on previous interactions, and sometimes extra clicks are added, which is what scientists call "rubato" and "ornamentation."

Additionally, scientists discovered that codas have distinct rhythms and tempos that can be categorized into a few types. By combining rhythm, tempo, rubato, and ornamentation, whales can produce a wide range of unique codas, and whales are using the variations intentionally.

According to researchers, the next steps would be to explore whether rhythm, tempo, ornamentation, and rubato serve similar functions, directly encoding the communicative intentions of the whales.

Whale phonetic alphabet
The whale phonetic alphabet is similar to the visualization of human speech sounds. Scientists used a grid where different types of tempo are on the vertical axis and various rhythm types on the horizontal axis. Each grid cell's color shows how often that tempo/rhythm combination appears in the DSWP dataset. Additionally, pie charts in each cell indicate the occurrence of rubato and ornamentation: the left pie shows how often rubato occurs compared to not, and the right pie shows the proportion of ornaments within that combination. Source: Research paper

More from Cybernews:

LockBit gang leader exposed in FBI ransomware breakthrough

Six arrested for crypto scam, over €750K seized

FBI warns of fraudsters targeting gift card systems

The next chapter in air travel: facial recognition and privacy concerns

FOMO over Netflix water cooler chats driving new subscriber surge

Subscribe to our newsletter



Leave a Reply

Your email address will not be published. Required fields are markedmarked