You're reading a paragraph in a language you've been studying for a year, maybe two. Somewhere in the middle of a sentence, something is different. You understood it. Not understood it, then confirmed it in English. Just understood it. The English version never arrived.
The first time this happens, it's so quiet you almost miss it.
If you've felt it, you weren't imagining it. That moment has a name, a mechanism, and a paper from 1994 that explains exactly why you couldn't have forced it to happen any sooner.
The detour your brain is running right now
When you encounter a word in a second language, your brain does not go straight to its meaning. It takes a detour.
In 1994, two cognitive psychologists, Judith Kroll and Erica Stewart, mapped exactly what that detour looks like. They called it the Revised Hierarchical Model, but the picture is simpler than the name: a beginner sees chien and thinks dog, and dog gets them to the concept. The foreign word is a label stuck on top of a word they already know. Meaning arrives in English, via French.
Kroll and Stewart knew the detour was real because it leaves a measurable trace. In reaction-time experiments, beginning learners show a delay, observable in milliseconds, between seeing a foreign word and accessing its meaning. That delay is the brain making the extra hop. In proficient speakers, it's gone. The foreign word connects directly to the concept, the same way dog does for a native English speaker, without routing through anything else.
What you felt mid-sentence, that moment when understanding arrived before the English version did, was that delay collapsing.
Why deciding to think in the language doesn't work
There's advice that moves through every language-learning community like a rumour: Stop translating. Just think in the language.
It isn't wrong. It is just backwards.
In 1988, Gordon Logan published a paper on how any skill becomes automatic: not language specifically, but typing, driving, anything. His argument was precise. A skill becomes automatic when it has been practised enough times that the brain switches from slow, effortful computation to fast instance retrieval. The ten-thousandth time you change gears, your hand moves before you've thought about it. You did not get there by deciding to stop thinking about gears. You got there by changing gears ten thousand times.
The direct-access pathway for language works the same way. It does not form because you will it into existence. It forms through volume of second-language input, at a level where you understand rather than decode word by word. Once there is enough of that, routing through your first language becomes slower than going direct.
Telling yourself to stop translating is the cart before the horse. By the time the advice could work, you don't need it.
What the volume has to look like
Not all reading builds the direct path at the same rate.
Paul Nation's vocabulary research established that reading without outside help requires understanding roughly 95 to 98 percent of the words on the page. Below that threshold, comprehension fractures too often. You stop to look things up, you guess at meaning, you translate what you can't guess. That effort is the opposite of automaticity: high-load, deliberate decoding that reinforces the laboured route rather than building the fast one.
At around 5,000 word families, something shifts. You know enough that comprehension becomes sustainable at the right level of text. Not effortless, but possible without constant interruption. The brain begins encountering words so often, in so many different contexts, that the translation shortcut costs more time than it saves. The direct path starts forming.
Reading speed is the clearest proxy for which mode you're in. Fluent native readers average around 250 words per minute in their first language. A learner in translation mode typically reads L2 text at 80 to 120 words per minute; the bottleneck isn't vocabulary, it's the detour. When you find yourself reading at 180 to 200 words per minute without losing the thread, the shift is underway.
This takes longer than almost any account suggests. No technique substitutes for it. The only input that matters is comprehensible text, at volume, at a level you can actually follow.
How it starts
That moment mid-sentence is not the arrival. It is one of the first signs.
Before it, there are smaller ones. You catch the tone of a sentence, ironic or urgent or affectionate, before you've parsed all its words. You finish a paragraph and realise you never heard an English version of any of it. You laugh at a joke in your second language. That requires holding two meanings of the same phrase at once, and it falls apart the moment you stop to translate.
The shift doesn't announce itself. It starts with the words you've met most often: the highest-frequency words in the language, the verbs you've read a thousand times, which begin to fire directly while rarer words still make the old hop. The field of translation gets smaller and more uneven over months.
Most learners notice it first in reading, not in listening. Reading is slower; the direct-access path has time to form before the faster demands of speech arrive. Listening follows, with a lag. The gap between reading comprehension and listening comprehension, which can feel like stagnation, closes the same way the translation habit closes: through volume, at the right level.
The practice that gets you there
The reading that builds the direct path is reading where you follow almost everything: the words you don't know are the exception, not the texture of every sentence. At that level, comprehension is automatic enough that the brain can start wiring direct connections between foreign words and concepts rather than spending its resources on decoding.
ToTo picks a story in your target language matched to the level where you understand around 95 percent of the words: the level where the brain starts building the direct path rather than reinforcing the detour. Three minutes a day. Boring, at first. The shift doesn't care about interesting. It cares about volume.