What We Still Don’t Know About Animal Languages and the AI Revolution's Role in Cracking the Code for the First Time in Human History
In a groundbreaking initiative of project CETI, a monumental and ambitious project has taken its first critical step toward decoding the language of whales. The group comprises experts in AI, linguistics, biology, robotics, and underwater acoustics. The Underwater Acoustics Lab at the University of Haifa, led by Professor Roee Diamant, is responsible for the data collection mission. This involves deploying mobile sensing platforms like autonomous submarines called "gliders" and deploying buoys equipped with massive sensor arrays. Additionally, they are developing automated processing capabilities for detecting, classifying, and characterizing whale signals, as well as for localization, tracking, and automatic identification of groups and individuals.
The initial phase of this ambitious project has been successfully completed with the development of an automatic algorithm for detecting echolocation signals—used by whales for navigation in the deep, dark ocean. The recently published paper details a method that enables reliable automatic detection of sperm whales even in noisy sea conditions. This open-source algorithm has quickly gained popularity among marine biologists and researchers worldwide due to its excellent and consistent detection performance across varying equipment and sea conditions, coupled with its user-friendly operation that requires no prior programming knowledge.
The forthcoming paper, to be published in the coming months, will present a method for the automatic identification and classification of whale communication signals known as "codas." This critical advancement will assist researchers in effectively and easily filtering years of recordings—a task previously deemed impossible.
These tools, alongside additional algorithms for whale localization and tracking being developed at the Underwater Acoustics Lab at the University of Haifa, are integrated into mobile sensing platforms such as autonomous submarines called "gliders" and stationary platforms, including three buoys equipped with massive sensor arrays extending to depths of 800 meters below the sea surface. These technologies represent a turning point, eliminating the need for the physical presence of researchers on boats to detect and record whales and for experts to listen to and analyze extensive recordings manually. Most of these processes are now conducted autonomously, significantly increasing the pace of data collection and analysis.
Following the development of algorithms for the automatic detection and classification of whale signals in recordings, the next phase will involve developing capabilities for whale localization and tracking, alongside methods for signal characterization—such as automatic identification of individuals and groups, size estimation, and more. This will enable researchers to link whale communication with their behavior.
In the third and final phase, all collected data will be analyzed using AI tools aiming to scientifically prove that sperm whale communication indeed constitutes a language. Ultimately, the goal is to decode the language or, in other words, translate some of their "words" and "sentences."
Approximately a decade ago, a group of researchers from Facebook developed an AI-based method for "unsupervised" translation of languages. In other words, this method eliminates the need for prior knowledge of the meanings of words and sentences from different languages. This method has greatly aided in deciphering ancient journals and recently sparked an ambitious race among researchers to apply it to non-human languages. So, are we on the brink of an era where we can communicate with our animal neighbors? What was once perceived as a centuries-old fantasy is now closer than ever to becoming a reality.
However, two essential conditions are necessary for this method to work:
The animal's communication system must indeed constitute a language.
The collection of an extensive database of words and sentences from the language we wish to translate.
To date, no animal communication system has been proven to constitute a language. On the other hand, there are increasing indications for a unique marine mammal living in the ocean depths—the sperm whale! One indication is their ability to communicate information about events that occurred at different times and places. This capability is a byproduct of a communication system that constitutes a language. This capability is claimed to have helped these whales develop new tactics to evade whalers in the 19th century quickly. Unlike other animals that use different frequency sounds to communicate information, sperm whales use unique clicking patterns to convey messages. These patterns resemble the famous "Morse code," with the pattern repository varying between whale pods in different parts of the world, akin to human vocabulary or alphabets.
For more details, please refer to the blog link and the paper link.
By Guy Gubnitsky