The Science of Learning a Language: How Memory, Input and Practice Actually Work

Learning a language feels like it should be about talent, or a magic method, or sheer hours. It is mostly about none of those. It is about working with the way memory actually forms, instead of against it. Decades of research converge on a surprisingly clear picture of what works, and once you see it, most of the advice out there sorts itself into signal and noise. Here is the whole picture, with the deep dives linked at each step.

There is a version of language learning that runs on willpower: study harder, do more hours, want it more. It mostly fails, and not because the people trying lack discipline. It fails because it ignores how memory works. The brain did not evolve to hold onto a list of foreign words you saw once; it evolved to forget almost everything and keep only what proves useful and recurring. Every effective method is really just a way of convincing your memory that a word is worth keeping.

The good news is that what convinces memory is well understood. This is a map of it: the one problem everything is fighting, the two engines that drive real progress, the techniques that make each engine run faster, and how to assemble them into something you can actually keep doing. Each section links to a full guide if you want to go deeper.

The one problem: you forget almost everything

Everything in language learning is downstream of a single stubborn fact, first measured by Hermann Ebbinghaus in the 1880s and replicated in the modern era: we forget new information fast, losing a large share of it within a day or two unless something brings it back. That is the forgetting curve, and it is why the honest answer to "I studied those words but I can't remember them" is: of course you can't, that is the default. Forgetting is not a failure of your method. It is the thing your method exists to fight.

Once you accept that, the whole field reorganises around a simpler question: not "how do I learn this word" but "how do I stop myself forgetting it." And the answers split into two engines that do different jobs.

Engine one: input, or you acquire a language by understanding it

The first engine is exposure to language you can follow. The applied linguist Stephen Krashen argued that we acquire a language mainly by understanding messages slightly above our current level, what he called comprehensible input. The strong form of his theory is debated, but the practical core is about as well supported as anything in language teaching: you need a large volume of language you mostly understand, because that is how your brain infers grammar, meaning and the feel of what sounds right.

In practice this means input at the edge of your ability, in quantity: shows, podcasts, and especially reading, where you can meet far more language at your level than speech allows. The most efficient version is extensive reading - reading a lot of easy, enjoyable material for meaning rather than study. Input is the engine that gives you raw material and range. What it does not do, on its own, is guarantee you keep any specific word.

Engine two: memory, or turning exposure into retention

This is where most self-teaching quietly wins or loses, and where the two most robust findings in the science of learning live.

The first is retrieval practice, better known as active recall. In a landmark study, Henry Roediger and Jeffrey Karpicke showed that students who tested themselves on material remembered far more a week later than students who simply re-read it the same number of times - even though the testers had spent less time with it. The act of pulling a word out of memory, not putting it back in, is what strengthens it. Recognising a word when you see it is easy and misleading; producing it from nothing is the skill you actually need, and the only way to train it is to practise producing it.

The second is the spacing effect: reviews spread out over time beat the same reviews crammed together. A large meta-analysis by Nicholas Cepeda and colleagues confirmed the pattern across hundreds of experiments, and added the crucial refinement that the ideal gap grows as the memory matures. Put retrieval and spacing together and you get spaced repetition: reviewing each word by recalling it, at expanding intervals timed for just before you would forget. It is the single most efficient memory tool we have for vocabulary, precisely because it targets the forgetting curve at its steepest point.

The scheduling can be handled by an algorithm so you never have to think about it. If you are curious how those work, the piece on the SM-2 and FSRS algorithms explains how an app decides when each card should come back, but the principle matters far more than the algorithm.

Making memory faster: cues and images

Retrieval and spacing decide whether a word sticks; how you first encode it decides how easily. A word learned as a bare translation is a thin thread; a word tied to a vivid image or a story has many more handles to grab it by. That is the logic behind the keyword method, the mnemonic technique Richard Atkinson studied in the 1970s: link the new word to a similar-sounding word you know, and picture the two together in one absurd scene. It feels childish and it measurably works, because the mind holds concrete images far better than abstract labels. A good memory cue does not replace review; it makes each review land faster.

What to learn: chunks, not just words

There is a second, quieter question underneath all of this: which units to learn. Beginners collect single words, and it works for a while, then stalls around the intermediate level. What is missing is not more words but the way words combine. Fluent speech runs on chunks and collocations - ready-made blocks like "make a decision" or "look forward to" that native speakers store and retrieve whole. Learning the pairing, not the bare noun, is what lets you produce natural language instead of assembling it stiffly from parts.

It also helps to know the difference between the words you recognise and the words you can use. Everyone understands far more than they can produce - the gap between passive and active vocabulary - and closing that gap, by retrieving and producing words rather than just meeting them, is much of what "getting fluent" actually means. As for how many words you need, the honest answer by level is in the guide on how many words it takes to be fluent.

The multipliers: sleep and consistency

Two things quietly amplify everything above. The first is sleep. Memory is consolidated while you sleep, and research on vocabulary suggests that spacing study across a night, rather than cramming it into one sitting, both speeds up learning and slows down forgetting - which is why a few minutes in the evening and a few in the morning beats one long block. The full story is in the piece on sleep and vocabulary memory.

The second is consistency, which sounds like a motivational cliche but is really a mathematical one. Twenty minutes a day for a year is over a hundred hours; a heroic weekend you never repeat is a few. Because memory is built by repeated, spaced contact, the person who does a little most days beats the person who does a lot occasionally, almost every time. The method that works is the one you will actually keep.

Putting it together

Assemble the pieces and the routine almost writes itself: get a steady diet of input you can mostly follow, produce the language rather than only consuming it, and review the words worth keeping by retrieving them on a spaced schedule, ideally with a cue and a little help from sleep. That is the entire science, and the practical version - how to fit it into a real week without a class or a tutor - is laid out in the guide on learning a language on your own.

MindDory exists to run the memory engine for you. You get your input the fun way, and whenever a word is worth keeping - typed in, scanned from a book, or captured from your AI chats - MindDory wraps it in a memory cue and brings it back on a spaced schedule right before you would forget, on web, iOS and Android. The input is yours; the fighting-the-forgetting-curve part is handled. If you want to start with the method most people get wrong, read what active recall is and why it beats re-reading, and the rest of this map will click into place around it.