The Whisper That Built a World

The War That Spoke in Tongues

In the first week of February 1942, a twenty-year-old Navajo man named Chester Nez sat at a wooden desk in Camp Elliott, near San Diego, and waited for an order he did not yet understand. Around him, twenty-eight other young men sat in the same silence. They had been chosen from more than two hundred candidates, all of them bilingual in English and Navajo, all of them tested for fluency so intense that even the cadence of their breath was examined. The Marine Corps wanted to build a code that could not be broken, and the men in that room were about to become the alphabet.

The order, when it came, was simple. Each man was handed a list of English words: *tank, grenade, fighter plane, dive bomber, torpedo, machine gun, fighter plane, ammunition*. Beside each word, he was asked to write a Navajo equivalent that did not yet exist. For some terms there was a natural substitution. *Hummingbird* became the Navajo word for fighter plane because the aircraft moved the way a hummingbird moves—sudden, hovering, capable of retreat. *Iron fish* became submarine. *Ear* became the abbreviation for the commanding officer. But for other words, the young men had to invent. They had to take a foreign concept and squeeze it into the bones of their own language, building new vessels for new thoughts, sometimes by describing a thing rather than naming it. The Navajo word for the United States, *Kéyah dah siání*, literally means "land that extends far out," because in Navajo grammar a country is not a label but a place with a kind of motion.

Within four weeks the code was complete. It contained 274 terms. It could encode any English message in less than thirty seconds. And within a year, it would be put to the test on Guadalcanal, where six Navajo code talkers took turns transmitting combat orders throughout a thirty-six-hour battle while enemy cryptographers worked frantically to break the cipher. They did not break it. Not because the code was mathematically sophisticated—the Navajo code, in cryptanalytic terms, was a relatively simple substitution cipher layered onto a transmission language whose structure was completely opaque to the Axis powers. The reason it could not be broken was far stranger and far more interesting. The reason was that the language itself, in its bones, did not fit inside the minds of those trying to listen.

The Navajo language belongs to the Na-Dené family, a vast grouping of tongues that once stretched from the Bering Strait to the American Southwest. Its sounds include a glottal stop, four distinct kinds of nasalization, tones marked by pitch contour, and a consonant inventory so unusual that the language contains no clusters—every syllable must end in a vowel, which gives spoken Navajo a soft, breathing rhythm utterly unlike the clipped efficiency of English. Its grammar is built on a system of classifiers that distinguish between moving and stationary objects, between objects that are long and slender, between objects that are flat and flexible, between things that are being handled and things that are simply being observed. To speak fluent Navajo is not merely to choose from a list of words; it is to enter a particular kind of attention, one in which the world is sorted by physical shape, motion, and posture before it is sorted by name. When the code talkers invented Navajo terms for American military equipment, they were not simply translating. They were translating between two entirely different ways of perceiving a battlefield.

And yet they did it. They transmitted six hundred messages in the first two days of Iwo Jima, all of them accurate, all of them without error, all of them under fire. Chester Nez went on to survive the war and lived until 2014, when he died at the age of ninety-three. Near the end of his life, when journalists asked him what it felt like to have invented a code based on his mother tongue, he said something that should hang on the wall of every linguist in the world. He said: "It wasn't the code. It was the language. The language was already a code."

That sentence is the door to this entire book.

---

The Mouth That Has Never Spoken

Three thousand miles east of Camp Elliott, in a small apartment in Seattle, a baby girl named Maya is seven months old. She is sitting in her high chair, banging a wooden spoon against the rim of a plastic bowl. She is not making language yet. She is making sound—a rich, varied, exploratory sound, full of clicks and squeaks and throaty purrs and vowel-like noises that rise and fall across almost every pitch the human larynx can produce. Her mother, a pediatrician who has read all the right books, asks me with a slightly worried smile whether Maya's babbling is on track. I tell her what linguists have known for forty years. Maya is not behind. Maya is, at this moment, ahead of every adult she will ever meet.

The research that revealed this fact is one of the most quietly astonishing in modern cognitive science. In a series of experiments that began in the late 1970s and intensified through the 1990s, the linguist Patricia Kuhl and her collaborators at the University of Washington placed infants from many different language backgrounds inside a high-tech sound booth and measured their ability to distinguish between phonetic contrasts. They used a head-turning procedure: when a baby heard a change in sound, she turned toward a speaker that lit up with a flashing bear. The bear, predictably, was irresistible. The findings were staggering. A six-month-old Japanese baby could distinguish the English /r/ from /l/ with the same accuracy as an American baby. A six-month-old American baby could distinguish the Hindi retroflex consonants that English does not even contain. A six-month-old English baby could hear the difference between consonants in Salish, a North American indigenous language whose sound system is so different from English's that adult English speakers can spend years failing to hear contrasts that Salish speakers produce without effort. At six months, every baby on earth is a perfect universal listener.

By ten months, this has changed. The Japanese baby has stopped hearing the /r/-/l/ distinction, because Japanese does not use it. The English baby has stopped hearing the Salish contrasts. The Salish baby has stopped hearing certain English contrasts that do not exist in Salish. By the first birthday, the infant's perceptual system has narrowed to match the language, or languages, of her environment. The universal listener is gone, replaced by a specific kind of listener whose nervous system has been tuned, like an instrument, to a particular set of frequencies.

What has happened is not that the baby has learned a vocabulary. The baby has not yet said a real word. What has happened is that the baby has begun the process of acquiring a particular cognitive architecture—a way of sorting the stream of sound into discrete units, of grouping vowels and consonants into syllables, of hearing some contrasts as the same and others as different. She is not yet thinking in her language. She is being installed into it, the way a person is installed into a particular operating system before they ever consciously run a program.

But the story goes deeper. Because what we now know is that this narrowing is not just a matter of sound. By the time Maya is two, and certainly by the time she is four, her categories for time, for space, for causation, for gender, for quantity, for evidence and inference, will also have been calibrated by the particular language or languages she hears. The famous Mandarin experiment by Twila Tardif and her colleagues showed that English-speaking toddlers who learn Mandarin as a second language after age seven have great difficulty with the Mandarin system of verbal aspect markers—tiny grammatical particles that indicate whether an action is ongoing, completed, or experienced. Children who learn Mandarin from birth acqui