How scientists decoded the "sperm whale phonetic alphabet"

We're learning our A-B-Seas! Researchers at the Cetacean Translation Initiative are using machine learning to parse sperm whale communication. Maybe if we learn the structural components of sperm whale language, we can use it to find out what they're clicking about us behind our backs?
Are you passionate about science communication, STEAM, AI/robotics, and/or marine and environmental science? Want to be at the forefront of Whale Duolingo? The group responsible for these findings, Project CETI, are accepting applications through May 24, 2024 for their Marine Conservation Fellowship! Find out more at www.projectceti.org/dominica/...
Article - "Contextual and combinatorial structure in sperm whale vocalisations." doi.org/10.1038/s41467-024-47...

[@jdrobertson42]

This research would have made Star Trek IV a much shorter movie.

[@kittehboiDJ]

Yeah, in the future aliens are going to destroy earth because it doesn't have any whales.

[@Kram1032]

12:38 as somebody fluent in that particular sperm whale dialect: You called me WHAT??? :P

[@sgpch1983]

i love your smile and laugh.. you are so beautiful and cute sam :) can not wait to see you learning whale :D

[@post1084]

could birds that can make a wide and varied number of sounds be taught whale ?

[@meander112]

If you like the idea of animal languages, I heartily recommend Ursula K. Le Guin's story, "The Author of Acadia Seeds."

[@sirnukesalot24]

I'd always wondered if the reason nobody could decipher cetacean languages is because they experience linguistic drift at a rate we would find unsalvageable

[@eruiluvatar236]

I wonder what would happen if we ran a dataset of sperm whale conversations through a transformer (the architecture behind chatgpt). Surprisingly they can be trained without knowing the meaning of the words and even if would be inefficient, just phonemes should also work. The end result wouldn't be a translator but a chatgpt for whales, we wouldn't know what it is saying but it could be a step towards that. If it worked we could try another experiment that is a much longer shot, training another one but forcing it to shoehorn whale speak into the same embeddings as human speech and looking for similar clusters or even straight training one system to do both and hoping that it somehow learns to translate (for that to work, whale speech would need to be conceptually similar to human speech).

[@post1084]

CLICK.......click

[@thomasgoodwin2648]

Click click ... WHHHEEEEEeeeeeeeeeezzze CLICK click!