How Memory Builds a Language: Retrieval, Spacing & the Words That Matter

Two people tell you their story. One learned Italian in a few months using an app for twenty minutes a day. The other had six years of school Spanish and still cannot order lunch. Both stories are true, and neither is mostly about talent. What differs is what each person did with their hours, and above all how they handled memory.

A language is mostly memory: thousands of words, sounds and patterns that have to be there when you need them. Research on how adults learn a second language is old, large and, on the ingredients, fairly consistent. This piece covers the memory half: why recalling beats rereading, how spacing the recalls out fights forgetting, and which words are worth remembering first. The next piece covers the rest, from input and speaking to how many hours it takes.

Retrieval: testing yourself beats rereading

In 2008 Jeffrey Karpicke and Henry Roediger had students learn 40 Swahili-English word pairs. Some kept testing themselves on the full list, round after round, and others dropped each word from testing once they had got it right once, though they kept studying. A week later the groups that kept testing recalled about 80% of the pairs; the groups that stopped testing recalled far less, roughly a third. Further study of words already recalled added little. Later work has added caveats about the design of the study, but the main point has held up in many experiments: pulling a word out of memory strengthens it more than looking at it again.

The practical rule is simple. A flashcard you flip without trying to answer is rereading. A card you commit to an answer on, before you look, is retrieval. The effort and the occasional failure are the point, which the psychologist Robert Bjork calls a desirable difficulty. It also means that the “Again” button is not a punishment. It is information that tells the schedule to bring the word back sooner.

Spacing: how fast we forget, and the schedule that fights it

Hermann Ebbinghaus published the first forgetting curve in 1885, learning lists of nonsense syllables and testing himself. Memory fell fast at first and then slowly. In 2015 Jaap Murre and Joeri Dros repeated the experiment with one participant, Murre, and reproduced the same general shape, with a small bump at 24 hours. Both are single-person studies of nonsense syllables, so the exact numbers do not carry over to your vocabulary. The shape does: forgetting is steep early and flattens later, and each successful recall tends to push the next forgetting further away.

That is why spacing works. A review of more than 300 experiments (Cepeda and colleagues, 2006) found that spreading practice over time beats cramming it, and a larger study in 2008 found that the best gap depends on how long you need to remember: a longer horizon wants a longer gap, roughly a tenth to a fifth of the time until you need it, by rule of thumb. Flashcard schedulers put this into code. The Leitner box system (1972) moves cards between boxes reviewed at longer intervals. SM-2, from Piotr Wozniak in 1987, multiplies the gap each time you recall. Newer systems fit a memory model to your history: Duolingo published a “half-life regression” method in 2016, and the open-source FSRS scheduler is an option in recent versions of Anki.

The simulator below uses a deliberately simple model so you can feel the effect. Tap days to add reviews and watch the curve. Every number is invented for illustration; real curves vary by person and by word.

Vocabulary: how many words, and which first

To follow a language you need to know most of the words in it. Researchers measure this as coverage: the share of running words in a text that you know. Paul Nation (2006) estimated that for English, reading novels and newspapers with 98% coverage takes about 8,000 to 9,000 word families (a base word plus its regular forms), and following conversation takes about 6,000 to 7,000. Batia Laufer and Geke Ravenhorst-Kalovski (2010) found that around 4,000 to 5,000 families (about 95% coverage) was a minimum for adequate reading comprehension, with about 8,000 as the comfortable level. Comprehension rises steadily with coverage and has no sharp cliff, which is why researchers argue about the exact threshold.

The shape is the useful part. The first thousand words do a huge amount: about 80% of an English text. After that each thousand adds less. That makes frequency order the best way to choose what to learn first, and it is why this library's paths teach the first 150 words, then the first 1,000 in frequency bands. The studies are about English; for other languages the shape is likely similar but the counts differ, so treat the numbers as orders of magnitude.

Words in context or in lists? Both have a job. Paired cards (a word and its meaning) are an efficient way to make a first connection, and a lone list gives you only shallow knowledge of a word. An example sentence adds how the word behaves, and meeting it in reading and listening adds the rest. That is why every card in these paths has a sentence on it.

Putting memory to work

Three research findings turn into three habits, and none of them takes more than ten minutes.

• Answer before you look. Say or type the word, then check. If you only flip the card, you are rereading. A miss is fine: it tells the schedule to bring the word back sooner.
• Let a schedule pick the day. Do not decide by feel which words to review. Your sense of what you know is poor, and a scheduler (a paper box, an app, or the deck in the Spanish pieces) spends your minutes where forgetting is closest.
• Start with the words that carry the most. Choose by frequency, put each word on a card with a short sentence, and add a few new ones a day rather than a hundred on Sunday.

If you only change one thing, change the first. A session of ten honest recalls does more than twenty minutes of looking at a list.

What sticks

• Retrieving a word strengthens it more than re-reading it, and the failed attempts are useful.
• Forgetting is steep at first, then flat; each successful recall at a longer gap buys a longer one.
• Spacing beats cramming, and the best gap grows with how long you need to remember.
• A few thousand frequent words give most of the coverage; learn in frequency order, with a sentence on each card.