AI Vocabulary Practice: How Spaced Repetition Builds French Faster

Vocabulary is where French ability is ultimately limited. Grammar you can work around; a word you do not have stops the sentence. The problem is scale, because reviewing fifty words works fine and reviewing four thousand does not, at least not by hand. This covers how spaced repetition solves that, and how to use it without generating a backlog you abandon.

How many words you actually need

Commonly cited estimates put active vocabulary at roughly these levels. They come from vocabulary research rather than from the CEFR itself, which describes what you can do rather than how many words you hold, so treat them as orientation.

The raw count understates the job, because each French noun also carries a gender you have to store with it, words travel in collocations that decide which verb goes with which noun, and most words carry several meanings that shift with context. Above all there is the gap between recognising a word on a page and producing it in the half-second you have while speaking.

How the system works

Spaced repetition is a scheduling answer to a memory problem, and the parts that make it work are the interval, the context around each word, and what it replaces.

How spaced repetition works

The principle is one sentence long: review a word just before you would have forgotten it. Reviewing earlier wastes the repetition, reviewing later means relearning from scratch, and the interval that sits between those two moves memory furthest per minute spent.

In practice each word runs its own schedule. You meet it, it comes back tomorrow, and if you recall it the gap widens to three days, then a week, then a fortnight. Miss it and the interval collapses back to something short. Multiply that across thousands of words and you have a scheduling problem no one solves on paper, which is the entire reason the algorithm exists.

The second thing it does is stop treating all words as equal. A traditional list reviews everything at the same frequency regardless of whether you find it easy, so your time gets spread evenly across words that need very different amounts of attention. Difficulty measured from your actual performance concentrates the time where it changes something.

SavoirX runs this over more than 18,000 words organised by DELF level, and structures each session the same way: words that have come due are cleared first, then new words are introduced at a pace that keeps tomorrow's review manageable. That ordering is what stops the backlog forming, and the backlog is what kills vocabulary habits.

Why context does the heavy lifting

A word paired with a single translation is fragile, because you have stored one association and recall needs several. Systems that show the word in an example sentence, next to its word family, with the collocations it actually appears in and audio for how it sounds, are building a network rather than a pair.

Varying how you meet the word matters just as much. Recognition (see the French, choose the meaning) is the easiest and least useful on its own. Production (see the meaning, produce the French) is what speaking requires. Spelling forces the accents, which change meaning in French. Gender drills attach le or la permanently, and context completion puts the word back in a sentence. SavoirX tests across translation in both directions, spelling against audio, gender, and word forms, with audio on every entry, because a word that survives only one question type has not been learned.

How it compares with paper flashcards

Paper cards are not wrong, they simply do not scale. The moment your deck passes a few hundred, the scheduling decision becomes the hard part of the work, and that is the part software does perfectly and humans do badly.

How to use it well

Consistency beats intensity, and not by a small margin. Fifteen to twenty minutes daily outperforms two hours on a Sunday, because the algorithm is scheduling against a daily rhythm and a weekly one leaves every interval wrong.

Skipping compounds. Words were queued for today because today was the day they pay, and a skipped day rolls that cohort forward into tomorrow's session, which is longer, which makes it likelier you skip again. Most abandoned vocabulary habits die exactly this way.

Engage rather than click through. Say the word aloud, build a sentence with it, notice its gender and whether it follows a pattern you have seen. Passive recognition produces the pleasant feeling of knowing a word and none of the ability to use it.

Then use the words somewhere real. Put them into writing practice where the usage gets corrected, into speaking, into narrating your day to yourself. A word you have produced in a sentence of your own is a different object from a word you have merely reviewed.

One caveat on trusting the schedule: if a word genuinely is known and keeps returning, most systems give you a way to fast-track it rather than making you sit through the early drills. SavoirX has an explicit "I already know this" for that. Use it for words you truly own, not for words that merely feel familiar, because feeling familiar is what recognition does right before recall fails.

What the exams ask for

DELF levels have characteristic domains, and building against them is more efficient than building generally.

TCF and TEF Canada pull from a different pool, since they are built for immigration: administrative language, professional situations, daily life in Canada, and the vocabulary of stating and defending an opinion. If your target is one of those, general French vocabulary lists will leave gaps exactly where the exam concentrates.

Which mistakes waste the effort

Learning words in isolation, which produces the fragile single-association problem above.

Ignoring gender, which means learning table rather than la table and then guessing every time you use it. Gender has to be stored with the word or it is not stored at all.

Practising recognition only. Multiple choice feels productive and trains the wrong direction, since exams and conversations both require production.

Making lists and never reviewing them. The list is satisfying to write and does nothing on its own.

And going too fast, which is the most common. A hundred new words a day generates a review load that arrives about ten days later and buries you. Ten to twenty daily, sustained, beats an ambitious month followed by abandonment.

Review time grows as your collection does, so start at the conservative end and increase only once the daily load has settled.

Whether it is working

The signs that it is: retention improving over time, sessions getting faster as more words become automatic, recognising your studied words out in reading and listening, catching yourself using them unprompted in writing or speech, and exam practice scores moving.

The signs that it is not: a review backlog that keeps growing, retention flat despite consistent work, words that feel familiar but will not come when you need them, and the particular fatigue that precedes quitting. Each of those has the same set of fixes: fewer new words per day, more active production, and meeting the vocabulary somewhere other than the app.

Frequently Asked Questions

How many French words do I need for B2?

Around 4,000 active words is the figure usually cited for B2, against roughly 2,000 for B1 and 8,000 for C1. Treat these as orientation rather than targets, since how well you use words matters more than how many you hold.

How many new words should I learn per day?

Ten to fifteen is sustainable for most people at 15 to 25 minutes a day. The limit is not how many you can absorb but how much review those words generate later, and an ambitious start usually collapses into a backlog within three weeks.

Is spaced repetition better than flashcards?

It is flashcards with the scheduling fixed. Paper cards get reviewed on a fixed cycle or whenever you feel like it, which wastes time on words you know and lets the hard ones slip. Spacing each word on its own schedule targets the moment just before you would forget.

Why do I forget words I studied last month?

Usually because you recognised them rather than produced them, or you never met them outside the app. Recognition and recall are different abilities, and a word you have never used in a sentence of your own is fragile.

Should I skip vocabulary review days?

Try not to. Words were scheduled for today because today was when reviewing them pays most, and a skipped day pushes that whole cohort into a backlog that makes the next session longer and less effective.