The Ideal Intermediary

This content was translated using artificial intelligence and may therefore contain errors.

First, it should be pointed out that, despite their size, sperm whales exist on the same spatiotemporal scale as we do. We can look them in the eye and will probably experience the passage of time in a similar way. The problem of the forest and the cosmic ocean is ruled out.

Second, everything seems to indicate that the sperm whale's brain is up to the challenge. The human brain weighs around 1.4 kilograms; sperm whales' can reach 8. What's more, both they and we have a neocortex, an outer layer where the most complex thinking takes place.⁴⁴Note to neurosurgeons and tactical divers: connecting an MRI machine to the head of a live sperm whale remains a challenge humanity has yet to overcome. For now, what we know about these brains comes from studies such as high-resolution post-mortem magnetic resonance imaging (MRI) and computed tomography (CT), histology and cell staining to identify specialized neurons (such as von Economo neurons), computer modeling and 3D reconstruction from brain sections, digital acoustic tags (DTAG) to record depth, movement, and sounds in real time, artificial intelligence and natural language processing (machine learning) to decipher clicks and codas, comparative genomics to study genes related to brain development, and large-scale hydrophone arrays for passive acoustic monitoring of social interaction. Although the human brain is denser and has more layers, the difference would not be significant. Moreover, the enormous surface area and the many folds and wrinkles of sperm whale brains suggest a great capacity for processing information. Thus, the various analyses carried out by modern neuroanatomy allow us to infer that sperm whales have emotions and even abstract thoughts. Just one example: in the human brain, the cells specialized in these functions are known as von Enconomo neurons. The same cells, found in greater numbers and distributed across more areas of the brain, have been discovered in different cetacean species and in dolphins. Sperm whales' emotions could even be more complex than ours.

Third, Project CETI has managed to determine that whales call one another by proper names, which necessarily implies a symbolic representation of an individual associated with a particular sound. A sign, at least as Saussure thought of it: signifier and signified.

Then, on Levin's spectrum of persuadability, whales sit comfortably at the highly persuadable end of the spectrum. And potentially persuasive, too. They have a sense of “self,” and it's even possible to speculate that they have something like a theory of mind: they imagine that other individuals may have their own points of view. For example, they adapt their vocalizations and behaviors depending on the individual they're communicating with, and show empathy in specific contexts, such as assisting at the births of sperm whales with whom they have no social relationship.

Lastly, the differences in the “ways of speaking” among different groups of sperm whales in different geographic areas constitute what researchers consider true dialects, bringing them one step closer to the debate over whether or not they have culture.⁴⁵A note to self: if the saying attributed to Max Weinreich is true—that “a language is a dialect with an army and a navy”—then take note, because sperm whales have dialects and are, in themselves, an armada.

If that weren't enough, we could add anecdotal evidence, plop it on our heads like a salmon hat, and go strutting around. In fact, events like the salmon hat are indications that orcas have complex social behavior, that they are playful creatures, and that their hunting methods suggest they can use tactical deception like the kind we've seen in dogs (for example, coordinated actions to surround a seal floating on a piece of ice). But let's not go that far. By now there's no doubt that we've found our ideal intermediary. The animal species most likely to hold a spoken conversation with humans in the not-too-distant future. An unprecedented event, inscribed in humanity's great epic history, from Magellan to Neil Armstrong and all the names worthy of tribute here.

All that would be missing, then, is the translator. The AI capable of acting as an interface between our way of understanding the world and the sperm whale's. That translator is being developed by biologist Shane Gero and a team of computer scientists at MIT, as part of Project CETI, and it's called WhaleLM. It's a large language model (LLM), similar to the technology behind today's chatbots or computer code generators, but designed specifically to analyze sperm whale vocalizations.

How does it work? WhaleLM is fed codas, those sequences of clicks emitted by sperm whales. The model learns the underlying patterns in these calls and uses what it has learned to predict which coda will come next in the sequence, just as LLMs learn to predict the next word a human will produce in a sentence. In a recent study (still in its preliminary phase), researchers used the coda sequences generated by WhaleLM to predict what sperm whales would do next. They succeeded with 72 % accuracy overall and 86 % accuracy for their future actions (such as diving).

A valid question Moriarty is sure to ask me is whether all codas convey meaning. He's seen me whistle around the house without rhyme or reason and knows that not every sound hatches an idea. Probably sharing the same concern, these scientists trained new language models by altering one element at a time: shortening sequences, changing their order, or modifying the rhythm, tempo, rubato (the flexible treatment of a piece's rhythm and tempo), and ornamentation of the clicks. They found that making any of these alterations made the model's predictions less accurate. According to the authors, this experiment provides the first scientific evidence that sperm whale vocalizations contain real, structured information that these creatures can act on.

The truth is, sperm whale clicks seem like a kind of underwater Morse code. This makes them particularly useful for study with AI models, since the input the algorithm has to work with already comes segmented (unlike the blue whale's long song, which is harder to divide into parts). Codas are groups ranging from 3 to 40 clicks in length. But saying that clicks are words and codas are sentences is a hasty claim. For that to be true, sperm whales must have what Nim lacked: syntax. The ability to group and arrange those clicks in different ways to say different things, things that have not been said before and go beyond mere repetition. In other words, they need to compose a language, and that language must be an indivisible part of their world.

So far, what we know is that this hypertrophied head, packed with fat and spermaceti, so valuable to Nantucket sailors, evolved for echolocation: sperm whales emit bursts of short codas and then receive them again, amplified by this “acoustic lens”—the fat surrounding the head—which allows them to discern important variations in the signal. Yet humans and other species with nervous systems know very well that the fact that a mental structure evolved for one reason doesn't mean we can't use it for others. Our ability to appreciate music, for instance, inhabits deep brain architectures that predate the invention of any musical instrument by millennia. Some authors, such as Stanislas Dehaene, argue that our ability to read is a neural skill we were able to build on the visual training we developed to distinguish elements standing out in the landscape (such as angular shapes that used to be lions or snakes and that we now quickly identify as letters). The same could be true of whales. The very system that makes echolocation possible may have enabled them to develop a certain kind of sound communication. For years, science considered this a rudimentary form of communication, a way to recognize one another and signal where they were, like a system of beacons. But researchers' work has suggested that sperm whales' intraspecific communication has a combinatorial architecture, phonological properties equivalent to human vowels, and perhaps even syntactic mechanisms. Here we find ourselves once more with the problem we started with: animal syntax versus human syntax.

Extrapolating the concept of syntax from humans to animals is extremely problematic, as anyone who has made it this far in the essay will know. That's why we need to do what I announced in the first chapters: use new terms, bring in words from traditions outside linguistics, or create new, hybrid, metaphorical, perhaps irreverent ones. Hence, linguists will have to forgive me for adopting the terms used by researchers, even though they do not yet have a place in canonical linguistic traditions. At this point I use the concept of phonological syntax to refer to the ability to combine sounds into emerging structures, and lexical syntax to refer to the ability to combine sound clusters with other clusters in an equally emergent way—that is, producing new meanings according to how freely the speaker—the cetacean—can combine them within certain rules.

If we accept that conceptual framework, even if only momentarily, we can say that sperm whale syntax undoubtedly exists at the phonological level: it works without difficulty to recombine sounds using different devices such as lengthening or ornamenting codas. But calling that syntax is incorrect from our human point of view, since syntax proper belongs to another level of analysis of the phenomenon (phonemes combine into morphemes, and morphemes into words, which are only then governed by syntax). That's why researchers speak of phonological syntax to account for a hierarchical structure at the level of sounds that has the potential to structure meaning. To show that creativity doesn't stop there, researchers also speak of a kind of spectral syntax, which involves altering frequencies so as to turn simple clicks into vowels and diphthongs. As if that weren't enough, sperm whales can also vary the acoustic resonance of sounds so that they fit fluidly with the voices of other whales, forming social groups that express themselves in dialects spread throughout the oceans. This would demonstrate a much more effective conversational capacity for coordinating the pod (for example, giving the order to dive in search of food). Finally, what we simply call syntax, they call lexical syntax. Both refer to the same thing: can sperm whales make sentences with hierarchies and movable elements that change the overall meaning of an utterance depending on their position? Can they say more than they can repeat? Can they lie?

Under Chomsky's terrible shadow, there is only one answer: we don't know yet.