Reading a word list again feels like studying and is not. Every line looks familiar, and that familiarity is the problem: recognising something and producing it are two different operations, and only one of them is what you need at the moment somebody asks you a question.
Nothing on this page changes when the language does, which is why it sits here once instead of inside each of the languages in every tongue this atlas speaks. What changes is how much of it you need, and when.
Three findings that decide almost everything else
- input has to be understood to count. Hours of audio you cannot follow build almost nothing. Material where you get most of it and miss a little is where acquisition happens, and finding that band is the practical skill.
- review just before you forget. Recall that takes effort strengthens a memory; recall that is easy barely touches it. This is why a schedule of expanding intervals beats any amount of rereading.
- output is what breaks a plateau. Comprehension keeps growing on input alone and production does not, which is why the intermediate wall is almost always a speaking and writing problem rather than a listening one.
What a study hour is actually resting on
The method has four parts and they are not four steps. They hold each other up, which means the top one cannot be practised alone however much time you give it, and the bottom one is invisible until it fails.
The part that does the work is the one that feels worst. When you reach for a word and it does not come, the search itself is what strengthens the route to it, so a card you answer instantly gave you almost nothing and a card you nearly missed gave you a lot. This has a consequence that people resist for years: easy review is close to worthless, and a session that felt smooth was probably a session where nothing moved. Why you forget, and the schedule that beats forgetting is the mechanism in full, and it is worth an hour because everything else on this page is downstream of it.
Underneath that sits the material. A page where you meet one unknown word every couple of lines is study. A page where you meet one every couple of words is decoding, and decoding teaches patience rather than language, because there is nothing left over to notice the pattern with. That threshold is higher than almost anyone expects, and it is the entire argument of filling your day with the language without moving country: the point of the volume is not exposure, it is finding enough material you can already nearly follow.
The loop runs per word, not per session
Any single word is somewhere on the same short circuit, and it can be at a different point of it in each direction at once. You can recover a word reliably while reading and still miss it every time you open your mouth, because those are separate routes to the same place and only the one you practise gets built.
The practical form of that is a rule about order. Reach for it before you look it up, out loud, even when you are certain you have forgotten, because the half second of failing is the thing you came for. Looking first turns the whole exercise back into recognition, which is the mode that feels productive and is not.
One word at a time
A word is somewhere on this loop, and somewhere different in each direction
The circuit runs per word rather than per session, which is why a word you read fluently can still be missing when you try to say it.
Any one word
- 01Meet itIn something you followed without stopping
- 02Reach for itOut loud, before you look
- 03Miss itThe half second of failing is what you came for
- 04Recover itLook, then say the whole sentence again
- 05Use itSomewhere with a person in it
Skipping the third stop is what re-reading a list does. The loop still closes and nothing gets stronger.
The debt nobody warns you about
Spaced repetition works by putting each item as far into the future as it can without losing it. That is the same property that punishes a break. Stop for eleven days and a deck that was fifteen minutes a day is 800 cards deep on the morning you come back, and the app presents all of them at once, in the tone of something you have failed.
Almost everybody quits there rather than at any point in the actual learning, and they describe it afterwards as having lost motivation. What happened is arithmetic. Two things prevent it, and both are settings rather than virtues. Cap the new cards low from the first week, low enough that the number looks insulting, because every new card is a promise of reviews for a year. And when the backlog does arrive, do not grind it: reschedule it forward and take the loss, since a card you meet a week late is a card, and a thousand cards you meet resentfully is the end of the deck.
That is one stall of several, and the later ones have different causes and different exits. The stalls every learner hits, and the way through each takes them one at a time.
The scheduler changed in 2023 and most people never noticed
Anki is the tool, and the interesting thing about it in 2026 is not the program. Since version 23.10, released in late 2023, it ships FSRS, which replaced SM-2, the algorithm SuperMemo published in 1987 and that nearly every other flashcard app still copies. SM-2 moves each card through a fixed ladder of intervals; FSRS fits a model of your own forgetting to your own review history and schedules against that, so the same card sits at different intervals for two different people. Check which one your collection is using before you conclude that spaced repetition does not suit you. A deck built years ago may still be on the old one.
For the half of this that needs another person, italki is the marketplace: it replaced the local language school and the card in a shop window, and it is paid, with community tutors as the cheap tier rather than a free one. HelloTalk replaced the pen pal, is free, and its paid tier mostly sells translation help that defeats the point of using it. Finding someone to speak to, and what to do in the hour is where the difference between those two matters, because an unstructured hour with a native speaker is a pleasant hour and not a lesson.
A model will also sit there and talk, and unlike a person it will not get bored, but you have to ask it for the correction explicitly or it will match your level and hand the mistake back to you politely. Getting that out of it is a useful answer instead of a plausible one, which is a skill of its own.
Where the hours go
An honest week has three kinds of hour in it and they do not substitute for each other: time under review, time in material, and time with a person. Turning a vague intention into hours on a calendar is mostly the work of deciding which of the three you are short of, and the answer is nearly always the third, because it is the only one you cannot do alone at midnight.
What happens inside the third one is the part people improvise, and improvising it is what turns a paid hour into a pleasant chat. Arrive with something: a page you half understood, or five sentences you tried to say during the week and could not finish. An hour of open conversation only exercises what you already have, which is why it feels comfortable and why three months of it leaves people at the level they started. The hour is worth its price at the moment you are working at the edge of what you can produce, and somebody who knows the language is watching you do it.
The question this page does not answer is which language, and the reasons people give for their choice are usually not the reasons that keep them going in month four. Difficulty, usefulness and the honest reasons people pick argues that out. If it is already decided, the shape of the first year differs enormously between, say, the gentlest big language to start and the world's second language, and the difference is not difficulty. It is what each one asks for first.