Competition for Translating Animal Sounds into Human Language

In 2025 we will see AI and machine learning help advance the understanding of animal communication, answering the question that has plagued humanity since the dawn of time: “What are animals telling each other?” The latest Coller-Dolittle Awardawarding half a million dollars to scientists who “crack the code” is an expression of strong confidence that modern machine learning technologies and major linguistic models (LLMs) are putting this goal in our minds.

Many research groups have been working for years on algorithms to understand animal sounds. Project Ceti, for example, has been improving click trains of sperm whales and songs of humpbacks. Modern machine learning tools require a large amount of data, and until now, the amount of high-quality and well-known data has not been available.

Consider LLMs such as ChatGPT that have teaching data that includes all of the available online content. Such stories about communication with animals have not been found before. Not only are human data corpora much larger than the kind of data we get from wildlife: More than 500 GB of words were used to train GPT-3, compared to just over 8,000 “codas”. ” (or quotes) in Project Ceti’s latest analysis of sperm whale communication.

Also, when we are working with human language, we already to know what is being said. We also know what “voice” means, which is a great opportunity for interpreting animal communication, where scientists often do not know if a wolf howl, for example, means something different from the howl of another wolf, or even if wolves perceive a howl as a howl. in a way they are similar to “words” in human language.

Even so, 2025 will bring new advances, both in the amount of animal interaction data available to scientists, and the types and power of AI algorithms that can be applied to that information. Only the recording of animal noises has been put out of reach by any scientific research group, with inexpensive recording equipment such as the AudioMoth becoming increasingly popular.

Larger archives are now coming online, as recorders can be left in the field, listening to the chirping of gibbons in the forest or birds in the forest, 24/7, long hours. There were times when it was too big to handle manually. Now, new automatic algorithms based on convolutional neural networks can run through thousands of hours of recordings, extract animal sounds and classify them into different types, based on their natural characteristics.

When these large animals are found, new analysis methods are possible, such as using deep neural networks to find hidden patterns in animal vocal sequences, which may be similar to the structure in human language.

However, an important question that remains unclear is, what do we hope to do with the cries of these animals? Some organizations, such as Interspecies.io, put their goal clearly as, “to convert symbols from one species to compatible symbols of another species.” In other words, to translate animal communication in human language. However, most scientists agree that non-human animals do not have their own language—not in the way that we humans do.

The Coller Dolittle Prize is a little more advanced, seeking a way to “communicate or interpret the communication of an organism.” Interpreting is a slightly narrower goal than translating, considering that animals may not have a language that can be translated. Today we don’t know how much information there is, or how little the animals speak. In 2025, humans will have the ability to leapfrog our understanding of not only how many animals speak but also how they communicate with each other.