Studying whale language using machine learning reveals a complex vocal system and suggests that cetaceans may talk to each other in the same way that humans do.
The study, published this week in Nature and titled “Context and combinatorial structure in sperm whale vocalizations,” was conducted by researchers at the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and Project CETI. . The researchers used a relatively simple machine learning algorithm to analyze the communication sounds made by sperm whales, which were previously thought to be a fixed message system.
The algorithm analyzed whale communication records from a family of sperm whales in the Caribbean and discovered a complex acoustic system. The study found that codas, the patterns of clicks that form the basis of whale communication, are actually much more complex than previously thought.
We don't know exactly what the whales are saying, but thanks to software their chatter is more detailed than some might assume and is built around an alphabet of sorts. I found out that it seems like it. That may mean what you think is a relaxing CD of sperm whale songs is actually just hours of listening to cetaceans moaning about the weather. .
“Similar to the International Phonetic Alphabet of Human Languages, this 'Sperm Whale Phonetic Alphabet' has a small set of axes of variation that cover a diverse set of observed (human) phonemes and (sperm whale) codas. “We show how this can occur,” the paper states. Observe.
Human sounds are classified based on where in the mouth they are produced, how they are produced, and whether the vocal cords vibrate, but whales use a combination of rhythm and tempo. Other variables include rubato (fine changes in the spacing between clicks) and embellishments (addition of additional clicks to the coda).
The important parts of the paper are:
One thing to keep in mind is that human speech (also called phonemes) is not completely similar to whale vocalizations. MIT researcher Jacob Andreas explained: register “Their underlying physiological processes are quite different; codas are much longer than typical phonemes, much of this structure is temporal, and there are no clear similarities between any of these features. I don’t think you can draw a dot.”
“In some languages, the number of codas is almost the same as the number of phonemes,” says Andreas. “But we don't actually know whether a coda is like a phoneme, a word, a sentence, or something else. has no name.”
Machine learning could be used to further analyze languages, animals, and humans
As mentioned above, discovering the internal structure of whale language phonetics alone does not reveal much about what the coda actually means.
“There's a lot going on in these vocalizations (whales singing together) that is completely different from human language,” Andreas pointed out. “But to answer that question definitively, we need to characterize what information is carried by these vocalizations, and this is the next big direction we're moving in.”
Andreas attributed the study's conclusions to the “really good visualization” done by co-author Pratyusha Sharma, rather than the “very simple” algorithms used in the paper, but Future research on both language and human language may be facilitated by AI. . One such use case is a general toolkit for analyzing the structure of unknown animal communication systems, which is actively being investigated by researchers at MIT and Project CETI.
Although human communication systems are much better understood than animals (including whales), there are still unanswered questions that machine learning could help answer. For example, was there only one original human language, or does our brain have a universal grammar (the biological basis of language theorized by former linguist Noam Chomsky)? Please, etc. ®
