The Friday Roundup - 12/8/2023 Edition

Whale Language, Archeological Surprises, The Syntax of Reality, Free Speech & Antisemitism, The Cult of AI

Every Friday, I share some of the most interesting articles, videos, and books I’m looking at with our subscribers. It’s an eclectic mix fueled by my unique personal variant of ADHD and pattern recognition, so you won’t find compilations like these anywhere else.

If you’d like access to the The Friday Roundup and all subscribers-only features and posts, you can sign up for just $5/month or $50 per year, right here:

---

Happy Friday, TSFers!

It’s a long one today, so let’s get right to the links:

Gabriel Barathieu, Mother and baby sperm whale, CC BY-SA 2.0

Scientists Have Reported a Breakthrough In Understanding Whale Language - Jordan Pearson/Vice.com

I don’t know what’s cooler, the fact that Sperm whales — which, perhaps not coincidentally, have the largest brains of any animal on earth — have been found to have a kind of language, or that it was only discovered through the application of AI that could analyze audio of the whales’ underwater sounds.

From the article:

Sperm whales are giants of the deep, with healthy adults having no known predators. Scientists studying their vocalizations have already picked out key elements of their communication, namely clicks, sequences of which are called codas. Now, researchers led by Gašper Beuš from the University of California, Berkeley report the discovery that the acoustic properties of these clicks—for example, pitch—are “on many levels analogous to human vowels and diphthongs,” which is when one vowel sound morphs into another such as in the word “coin.” The researchers even identify two unique “coda vowels” that are “actively exchanged” in conversation between whales, which they term the a-vowel and i-vowel. 

The researchers explain in their paper, published as a preprint online this week, that the first clue that so-called spectral properties could be meaningful for whale speech was provided by AI. Beuš previously developed a deep learning model for human language called fiwGAN which “was trained to imitate sperm whale codas and embed information into these vocalizations.” Not only did the AI predict elements of whale vocalizations already thought to be meaningful, such as clicks, but it also singled out acoustic properties. 

To follow up on the AI’s tip, the researchers analyzed a dataset of 3948 sperm whale codas recorded with hydrophones placed directly on whales between 2014 and 2018. They only analyzed one channel from the hydrophones to control for underwater effects and whale movement, and removed click timing from their visualization to better isolate patterns in the acoustic properties themselves. 

These visualizations vindicated the AI’s prediction: The whales reliably exchanged codas with one or two formants—frequency peaks in the sound wave—below the 10kHz range. The researchers termed these codas “vowels,” with single-formant codas being a-vowels and two-formant codas being i-vowels. “This is by analogy to human vowels which differ in their formant frequencies,” the authors wrote. They also identified upward and downward frequency “trajectories” in these codas, which they considered analogous to diphthongs in human language.

The researchers say that the whales are purposefully controlling the frequencies they are using to communicate, in a way that is somewhat reminiscent of tonal languages like Mandarin Chinese.

From a different article, this one written back in 2011 for Smithsonian Magazine (before the help of AI), we get a bit more about the acoustic biology of these fascinating and mysterious beasts, and what some of the sounds might mean:

Biologists now believe that the sperm whale’s massive head functions like a powerful telegraph machine, emitting pulses of sound in distinct patterns. At the front of the head are the spermaceti organ, a cavity that contains the bulk of the whale’s spermaceti, and a mass of oil-saturated fatty tissue called the junk. Two long nasal passages branch away from the bony nares of the skull, twining around the spermaceti organ and the junk. The left nasal passage runs directly to the blowhole at the top of the whale’s head. But the other twists and turns, flattens and broadens, forming a number of air-filled sacs capable of reflecting sound. Near the front of the head sit a pair of clappers called “monkey lips.”

Sound generation is a complex process. To make its clicking sounds, a whale forces air through the right nasal passage to the monkey lips, which clap shut. The resulting click! bounces off one air-filled sac and travels back through the spermaceti organ to another sac nestled against the skull. From there, the click is sent forward, through the junk, and amplified out into the watery world. Sperm whales may be able to manipulate the shape of both the spermaceti organ and the junk, possibly allowing them to aim their clicks. The substance that made them so valuable to whalers is now understood to play an important role in communication.

Whitehead has identified four patterns of clicks. The most common are used for long-range sonar. So-called “creaks” sound like a squeaky door and are used at close range when prey capture is imminent. “Slow clicks” are made only by large males, but no one knows precisely what they signify. (“Probably something to do with mating,” Whitehead guesses.) Finally, “codas” are distinct patterns of clicks most often heard when whales are socializing.

Codas are of particular interest. Whitehead has found that different groups of sperm whales, called vocal clans, consistently use different sets; the repertoire of codas the clan uses is its dialect. Vocal clans can be huge—thousands of individuals spread out over thousands of miles of ocean. Clan members are not necessarily related. Rather, many smaller, durable matrilineal units make up clans, and different clans have their own specific ways of behaving.

A recent study in Animal Behaviour took the specialization of codas a step further. Not only do clans use different codas, the authors argued, but the codas differ slightly among individuals. They could be, in effect, unique identifiers: names.

Interesting side note: