neural language acquisition

In the 1980s, the DARPA speech recognition challenge settled the debate between statistical methods and hand-written rules, setting a common standard for the entire field.

In March 2020, France emptied out: people stopped going to work, school, and cafés, and the streets filled with silence. Epidemiologists stared at the graphs and understood that the virus was not spreading on its own, but through handshakes, conversations, and metro rides. But how could they predict how many contacts a person would keep if the government announced new restrictions?

In the Antarctic ice, a ghost particle leaves behind only a faint flash of blue light. The IceCube detector captures these flashes, and from them we need to infer where the particle came from. The PolarBERT model has learned to do this, but what exactly it sees inside has remained a mystery.

At a crowded reception, an older person asks their conversation partner to repeat themselves again and again: the words drown in the hum of voices. This is what the bridge looks like, the one that carries sound from the ear to meaning. While that bridge holds, noise merely gets in the way, but it doesn't cut the conversation short.

The classroom goes quiet after the lesson, but the echo still rings in the child's head: faces, objects, places. It is this echo, not the lesson itself, that decides what will stay in memory. For the first time, researchers have caught it in children aged 6 to 17. Children memorized scenes inside an MRI scanner: they were shown a room, then an object or a face in a particular spot.

Every heartbeat sends a wave through the cerebral arteries that not only pushes blood but also briefly inflates the vessel. The arterial wall, like a tightly stretched rubber tube, passes that motion to the neighboring brain tissue, and the brain pulses ever so slightly in time with the heartbeat.

The brain is like an orchestra, with the cortex as the conductor and the thalamus as the first violin. The thalamus relays signals from the eyes to the visual cortex. But the cortex can also send commands back, tuning the thalamus for the task at hand. This was once a hypothesis, but now it's been tested in humans.

A magnetoencephalograph, a helmet-like device with hundreds of sensors, picks up the weak magnetic fields generated by active neurons. When a person listens to speech, these signals contain imprints of the heard words. But extracting them is tricky: signals from thousands of sources merge into a tangled mess.

A stadium packed with fans hears the announcer’s voice for the first time, and instantly the whole place erupts. The message sounds again and again. Only a handful of people in nearby rows respond. That’s roughly how neural networks behave in brain organoids — tiny lab-grown replicas of the brain.

The human mind is like a construction set that builds itself. New ideas don't die with their creators: they are passed on, mixed, and selected. This process is called cultural evolution: useful inventions stick, useless ones are forgotten. That's how a ratchet works — a mechanism that only turns forward. But what exactly happens in the head when one thought displaces another?

Radio signals in a room act just like sound in a concert hall: bouncing off walls and objects, they create a unique pattern — a silent melody for every location. Your phone, like a keen ear, catches this melody amid noise and distortion. It remembers which chord matches the corner of the couch, and which one fits the doorway.

Ordinary AI looks at medical images like puzzles made of unfamiliar pieces: it considers any two unmatched pictures completely different. It's like trying to tell twins apart without knowing they're related. In medicine, however, patients with the same condition look nearly identical on MRI, and the algorithm mistakes 'relatives'.

An AI predicting room acoustics is like a student taking an exam with a cheat sheet. If the test points are the same as the training ones, the model simply memorizes the acoustic 'fingerprint' of each position — like a student recognizing a question by the dog-eared corner.

A typical tracking program resembles a navigator steering a ship by the stars. But as soon as the sky clouds over, coordinates drift, and entropy (a measure of uncertainty) builds up in the data. The new algorithm gets a ‘co‑pilot’ — a module that monitors how confident the main navigator is.

The modern standard approach to AI training is like a child locked in a library: he remembers millions of books but doesn't know that fire burns and falling hurts.

A large neural network like Whisper masters hundreds of languages and resembles the standard model of the universe: everything is balanced. But training on a rare language disrupts this equilibrium — it turns into a black hole, pulling in all training resources. The network starts forgetting the other languages.

Complex machinery, from drones to surgical robots, sometimes behaves like a wild mustang: full of disorder and whims, where the slightest touch triggers a storm. Instead of rote memorization of formulas, the new method observes the system like a horse breaker.

Machine learning has long ceased to be just a tool — it is becoming an investigator, independently fishing out the laws of nature from data. But how does a silicon mind come to understand?

In embryonic development, a cell resembles a musician in a vast orchestra: it must play its part, following a conductor — the gradient of signaling molecules. But the conductor is unreliable: morphogen concentrations constantly fluctuate. How then does a cellist not veer into a violin melody?

In the courtroom, a string is presented as input. Twelve jurors (computation paths) examine it symbol by symbol. Each can render a 'guilty' (accept) or 'not guilty' verdict. For a conventional nondeterministic automaton (NFA), one 'yes' vote suffices.

A closed loop: the machine proposes an idea, tests it on people, sees the failures, and spawns a new, more accurate guess. That's how AutoCog works — an automatic cognitive scientist. Previously, in the science of thinking (cognitive science), the hardest part, coming up with a theory, was done only by humans.

You're telling a friend a story, and their brain seems to tune into your wavelength — the same regions light up in sync with your words.

A person looks at a door lock and sees a keyhole. They know that a key is inserted and turned. But if instead of a keyhole there's a touch panel, the familiar frame breaks. The person doesn't just make a mistake: they realize that the usual approach won't work and can say, "There's no place for a key here; a different approach is needed." Such an honest stop with an explanation is rare for machines.

Molecules float in a test tube. Their abundance (say, the concentration of some protein) becomes the input signal. And the output isn't just a cloudy solution but a clear yes-or-no answer. The authors build a chemical cascade that does exactly that.

The hand traces the word "mom" in the air — familiar arcs and strokes. The brain, meanwhile, is not flipping through a photo of each letter. It launches a chain of simple movements.

An artist paints a picture, glancing at the audience in the hall. If the viewer smiles, the artist adds more red and rounded lines to the canvas; if the viewer frowns, the artist removes blue and sharpens the angles.

A hundred people are talking at once in a large hall. You try to hear one, but voices blend, some whisper, and you don't even suspect the existence of others. That's roughly how a conventional laser gas analyzer feels when dozens of substances mix in a combustion chamber or a factory pipe.

A summer meadow surfaces in memory not just as a picture — along with it come the scent of grass and the warmth of the sun. The brain glues sensations into a single image, and that image holds up better than any one sensation alone. But exactly how neurons stitch the visual to the olfactory has long been a mystery.

A doctor studies a brain scan: somewhere in the tangle of tissue, a tumor is hiding. One glance picks out small spots and edges, another takes in the overall shape and location. A person combines both, but across hundreds of scans that gets tedious. A new program does the same thing automatically.

52.4 nDCG@10—that's the result ModernBERT achieves after Vocabulary Transfer. Just yesterday it stumbled over letter case, and today it outranks the best sparse retrievers. A neural network flunks search because it can't agree on words. 'Token' and 'token' are two distinct universes to it.

Speech is like an orchestra where every note must sound at just the right moment. The brain turns a thought into a chain of sounds in a fraction of a second, and until now it was unclear how it manages this hierarchy — from individual sounds to syllables to whole words.

Neurons in the brain and nodes in artificial neural networks don't just fire with simple yes/no signals; they also handle smoothly varying quantities, like where an object is and how big it is. But neural responses are finicky: the same stimulus triggers slightly different bursts of activity each time.

A chef's knife and a peeler each handle different tasks, but when a cook learns a new recipe, they often grab the same knife for everything. The brain seems to do the same when mastering a new skill of distinguishing patterns. In the experiment, monkeys trained for weeks to tell apart barely noticeable differences in complex patterns.

Science is like a city where residents pave the streets themselves. Where paths are already worn, new houses go up, while untouched vacant lots go unnoticed — not because building there is impossible, but because no one ever looks.

A medicinal chemist is like a detective searching for the right compound among millions of suspects. Modern neural networks speed up the search but behave like a witness who says "it's him" and refuses to elaborate. In pharma, such secrecy is dangerous: a spurious correlation could send a toxic compound to the next stage of trials.

Proteins are chains of amino acids, and they often contain repeated chunks. Sometimes the repeats are exact, like the same phrase said twice. Sometimes they come with typos: a couple of letters swapped, but the meaning is the same. These repeats matter: they help the protein fold into the right shape and do its job.

Proteins in nature are as numerous as grains of sand on a beach, and each has its own job: some break down food, others build cells. But to understand what a specific protein does, it must be studied in the lab — a slow and expensive process.

At night, a person breathes, and the rhythm of that breathing can tell you a lot about their sleep. But usually, doctors look at the brain's electrical waves — recorded by sensors on the head. New work shows how to reconstruct those waves from breathing alone, without touching the head. The model acts as a translator from the language of the body to the language of the brain.

A detective is looking for a criminal in a file of mugshots, but the portraits are mixed up. Usually, they expect a man with a gray beard, but the culprit is young and clean-shaven. Similarly, doctors often fail to recognize Alzheimer's disease in those who show unusual symptoms — not memory loss, but problems with vision or speech.

When a tumor affects the entire brain, doctors irradiate the whole organ. It's a reliable method, but setting up the radiation is painstaking: you need to adjust parameters so the tumor gets the right dose while the eyes and lenses suffer as little as possible. A specialist typically spends about 15 minutes manually tweaking settings in the software.