The Challenges of Animal Translation – The New Yorker

This much we know. But Magnasco doesn’t think that anyone has achieved a basic understanding of dolphish. “I’m not yet confident that I know what is the signal, what is the variation, what is the intention,” he said. “You need an extremely large body of data to do that, and it’s unclear that we have enough yet.” Still, there are hints that it might be possible. In 2013, Herzing and her team at the Wild Dolphin Project used a machine-learning algorithm called Cetacean Hearing and Telemetry (CHAT), designed to identify meaningful signals in dolphin whistles. The algorithm picked out a sound within a dolphin pod that the researchers had earlier trained the dolphins to associate with sargassum seaweed—a clumpy, floaty plant that dolphins sometimes play with. The dolphins may have assimilated the new “word,” and begun using it in the wild.

And yet, in an important sense, dolphish may be more than a language. Dolphins don’t just make whistles—they also employ body language and a variety of sounds, including clicks, which they use for sonar echolocation. From the acoustic reflections created by the clicks, a dolphin can form a mental picture of an object’s size, shape, and density. Dolphins can interpret one another’s sonar signals. “They are able to see shapes of things when they passively eavesdrop on someone else’s clicks,” Magnasco said. Using sound alone, they can see what another sees.

With cephalopods such as octopuses and squid, the gap widens further. Our common ancestor with them is thought to be a flatworm with only the most rudimentary of nervous systems; octopus brains are essentially a separate evolutionary experiment in developing intelligence. An octopus has around five hundred million neurons in its body—in the same range as a dog—but they are spread around, mostly in the arms, where they form clusters called ganglia, connected to one another. Even the brain in the center of the body is bizarre, because the creature’s esophagus, through which food is ingested, runs right through the middle of it. Some researchers hold that, with this distributed nervous system, cephalopods might host a “community of minds.” It isn’t clear, for instance, whether it’s the brain or the arms that “decide” what the arms do.

“An octopus mind is nothing like a primate mind, nor indeed like a dog’s, elephant’s or bat’s mind,” the evolutionary ethologist Phyllis Lee, of the University of Stirling, in Scotland, has written. According to the Australian philosopher of mind Peter Godfrey-Smith, cephalopods are “probably the closest we will come to meeting an intelligent alien.” Some researchers still hesitate to attribute “mind” to octopuses at all—and yet their behavior is often indicative of memory, problem-solving, cunning, personality, and even, some argue, sentience. They figure out how to unscrew jars, how to sabotage laboratory lights with jets of water (they may not like brightness), how to escape from their tanks just when their human wardens aren’t looking. They appear to gather items sometimes not for any obvious use but simply because they find them interesting. Some octopuses in captivity have been known to take what seems to be a dislike to individuals, squirting them with water at every opportunity. “They talk to you, reach out to you,” Michael Kuba, a marine biologist who has worked at the Okinawa Institute of Science and Technology, in Japan, told me. “But only to people they know.”

Octopuses seem to have designs of their own, which may subvert ours. Their agendas are often unfathomable. “When I first saw octopuses play,” Jennifer Mather, a professor at the University of Lethbridge, in Alberta, Canada, who specializes in cephalopod behavior, said, “I realized that we only saw it as play because it looked like our play.” She instead describes such behavior as motivated by exploration and led by the question “What can I do with this object?” (And yet an octopus might not even have an “I.”) Ultimately, Mather said, it’s hard to know for sure what the actions mean, because we don’t know where each one starts and finishes; we have no lexicon for translation.

Traditional efforts in animal cognition have attempted to build such a lexicon. Researchers have devised systems of symbols that animals can use by touching or pointing. In the nineteen-eighties, Reiss developed an underwater keyboard for dolphins; the animals quickly figured out, without instruction, how to request a body rub or a ball. Reiss also used mirrors to explore dolphin self-recognition: the animals not only appeared to recognize themselves (a sign, some researchers think, of a degree of consciousness) but also seemed to “play” with their reflections (by spinning, for example). Between 2016 and 2019, at the National Aquarium in Baltimore, Reiss and Magnasco collaborated on studies that used an eight-foot underwater touch screen fitted with dolphin-friendly interactive apps, including a version of Whac-A-Mole in which fish move across the display.

Using such systems, it’s possible to ask animals about their preferences among two or more alternatives—the same approach that child psychologists often take in trying to understand the reasoning of preverbal infants. Roger Payne, a whale-song expert—he co-discovered the songs of humpback whales, in the late nineteen-sixties—has explained how groups of alternatives might be used to pose ever-more-specific inquiries. “We might try asking dolphins direct questions,” he said, at a workshop of the Interspecies Internet project, at M.I.T., in 2019. “Do dolphins fear boats? Are sharks scary? Which of the following sharks is most scary? Is your mother afraid of sharks?” We might find out if dolphins lie regularly to each other as humans do, he said. “I would be surprised if they didn’t.”

The challenge, of course, is that it’s humans posing the questions and determining the choice of answers. But that’s changing. “The exciting thing about artificial intelligence and computer technology is that we are beginning to be able to decipher animal languages and animal cognition on terms that are meaningful to the animals, and not on our terms,” Slobodchikoff told me. Today’s machine-learning systems analyze data and look for correlations with startling efficiency; often, they find statistical connections that human analysts miss. They can, for example, deduce the “shape” of a language space, which depicts where words and concepts sit in relation to one another (“king” will typically be as far from “man” in this space as “queen” is from “woman”); these conceptual spaces turn out to be surprisingly similar for different languages—presumably because they are all representations of the same external world. Remarkably, the same sort of conceptual mapping will work not just for languages but for images. Researchers at Google have developed an A.I. system that can translate from an “image map” to a language map. After being trained to label a wide variety of images, it can be given an image that it has never seen before—of a dog, say—and make a good, sometimes even excellent guess at the word for what it has been shown. Given enough training data, these A.I. algorithms can extract semantic meaning from a range of non-linguistic inputs.

Britt Selvitelle, a computer scientist who worked on the team that created Twitter, is a founding member of the Earth Species Project, an organization founded in San Francisco that is developing A.I. approaches like this to animal communication. “We’re working on decoding the first nonhuman language,” he said, at the M.I.T. workshop—a goal that he thinks can be reached in five to ten years. In theory, a machine-learning system is particularly well suited to the problem of translating animalese. The loose correspondences between human and animal words and concepts may not matter to an A.I.; neither will the fact that animal ideas may be expressed not as vocalizations but as gestures, sequences of movements, or changes in skin texture. A neural network makes no assumptions about the nature of the input data; as long as there is some aspect of an animal’s behavioral repertoire that represents or expresses something that our languages can also express—a type of species, a warning, a spatial direction—then the algorithm has a chance of spotting it. “We’re really asking people to remove their human glasses, as much as possible,” Selvitelle said. One Earth Species Project collaboration, called Whale-X, aims to collect and analyze all communications among a pod of whales over an entire season.

It will be hard for the team to tag and track the individual whales. But Magnasco told me that he is also skeptical of the approach on a conceptual level. Even if the data can be gathered and analyzed, he said, it’s not obvious that we’ll have a word-for-word translation of whale to human terms, particularly without more understanding of their behavior. “If there is a vocabulary that has to do with their living environment, there is a massive amount of our vocabulary that just won’t make sense to them,” he said. In trying to import language-translation techniques to other species, the Earth Species Project might be “postulating an inherent similarity that we have no reason to assume.”

Many human languages seem to converge on a small list of omnipresent concepts formulated as individual words. Perhaps the most widely used lists of such words were derived in the mid-twentieth century by Morris Swadesh, an American linguist. The canonical Swadesh lists have between a hundred and two hundred and fifteen items. They contain personal pronouns, body parts, common animals such as “bird” and “dog,” verbs such as “eat,” “see,” and “hear,” and objects and substances such as “sun,” “water,” “stone,” and “smoke.” Magnasco points out that most of the items on the Swadesh list could have no “dolphish” equivalents, even in principle, because they have no relevance to the dolphin’s world. Among those excluded, he argues, would be “common words from our terrestrial environment, like ‘dog,’ ‘louse,’ ‘tree,’ ‘leaf,’ ‘root,’ ‘bark,’ ‘horn,’ and ‘mountain’ ”; words from terrestrial-animal anatomy—“nose,” “claw,” “foot,” “knee,” “hand,” “neck,” “feather,” “hair”; and words related directly to gravity, such as “walk,” “lie,” “stand,” “path,” and “swim”; and also the colors red and yellow, which dolphins can’t see. Finally, there are “words that do not exist or lose meaning in an aquatic environment”: “water,” “drink,” “rain,” “earth,” “fire,” “burn,” “ash,” “dry,” and “wet.”

If we could speak to them, dolphins wouldn’t understand the metaphor of a glass being half full or half empty. But how much does that matter? We can be discouraged by the fact that concepts that are universal among humans have no place in the conceptual landscape of the dolphin; alternatively, we can be encouraged by the possibility that there might be any overlap at all. It’s incredible to think that people and dolphins might communicate about anything, even seaweed; also, it’s striking to imagine dolphins shaking their heads, or the equivalent, over our inability to grasp concepts that seem obvious to them. It may be that the most interesting, revealing part of dolphish is precisely the part that lies outside our own lexicon—which is to say, outside our own minds. If, in fact, we find ourselves unable to fully reconstruct another creature’s mental world, it may be enough just to acknowledge the reality of what we can’t articulate.

In other ways, even basic communication may be of value. Some of our mistreatment of other species is obviously callous and selfish, as in factory farming, but some of it arises from a communications breakdown. Dogs are often surrendered to shelters, Slobodchikoff said, because people have trouble “reading and understanding the signals with which they are trying to communicate with us.” And, by changing what we believe about the minds of animals, even attempts at communication may affect how we think of them as legal entities. More than a hundred experts have signed a declaration urging the banning of octopus farming on the grounds that these “sentient and sophisticated” animals should not be kept in “sterile” and “monotonous” environments. Octopuses have long been denied the consideration and welfare that we give to vertebrates, but many marine biologists now agree they should be seen as possessing minds. Organizations such as the Great Ape Project and the Nonhuman Rights Project are seeking to extend minimal legal rights to certain animals such as great apes, elephants, dolphins, and whales.

In “King Solomon’s Ring,” Konrad Lorenz suggested that Solomon could communicate with animals not because he possessed a magical object but because he had the gift of observation. Lorenz “made the space to see and hear what other animals were doing,” Reiss told me. New technology may or may not help us to communicate with animals. But even the attempt at translation suggests a deepening of respect for them—and a willingness to free ourselves from our human preconceptions and prejudices.