Spaced repetition is one of the best-researched techniques for language learning, but most learners use it badly — grinding through decks of isolated words that never make it into active use. Here's a practical guide to using SRS for language learning in a way that actually produces speaking and reading fluency, not just trivia recall.
The Core Principle
Vocabulary is not a list. It's a network of associations between sounds, meanings, grammatical contexts, and lived experiences. Your flashcards should mirror that network as closely as the card format allows. Isolated word → translation cards don't.
What Good Language Cards Look Like
Sentence cards over word cards. Instead of 犬 → dog, use うちの犬はとても大きいです → Our dog is very big. The sentence teaches usage, grammar, and meaning simultaneously. Clozing the target word (犬) turns it into a retrieval event while preserving context.
Audio on every card. Language is sound first, text second. Every card should have a native-speaker audio clip — generated with high-quality TTS or recorded from a real sentence source. If your app doesn't support audio, it's only half a language-learning tool.
Image cards for concrete nouns. An image of a dog bonded to the word 犬 creates a more durable memory than a translation ever will. Avoid the translation layer entirely when you can.
Production cards for high-frequency words. Most cards are recognition (see the word → recall the meaning). For the top few thousand words, add production cards (see the meaning → recall the word). Production is harder and more useful for speaking.
Desired Retention for Language
Language learners typically run lower retention targets than medical students — 0.85 to 0.88 is a good default. Language vocabulary is long-tail: you'll encounter the same words in real content over and over, so occasional forgetting in SRS is usually re-corrected by reading and listening. Higher retention just increases workload without adding real-world gain.
Frequency Lists Are Your Friend
The 1,000 most common words in a language cover roughly 75–80% of everyday text. The top 2,000 cover around 85%. The top 5,000 cover 95%+. Start with a frequency-ordered deck and work top-down. Build your own only after you've exhausted the shared decks available for your target language.
Don't Just Do Flashcards
SRS is a supplement to immersion, not a replacement. The research on vocabulary acquisition (see Nation 2013, Learning Vocabulary in Another Language) consistently shows that learners who combine flashcards with reading and listening acquire vocabulary much faster than those who rely on either alone.
A reasonable split: 30% flashcards, 70% real input (books, articles, video, podcasts, conversation). The flashcards prime the pump; the input makes the words stick.
Mining Sentences from Real Content
The most effective language deck is one you build from material you actually want to read or watch. When you encounter a sentence in a book or a show that contains a word you don't know, capture the sentence, add audio, and cloze the target word. This technique is sometimes called "sentence mining" and it's how many of the most successful self-directed language learners reach fluency.
Neurako's photo-to-card and audio-capture features are specifically useful here — you can snap a page of a foreign-language book and generate sentence cards in seconds, then edit them before committing.
Common Mistakes
Learning words in isolation. Produces cards you can ace but can't use in a sentence.
Adding every unknown word. You'll drown. Add only words that are at or just above your current level, ideally ones that appear more than once in your input.
Skipping audio. The written word is not the language; it's a representation of it. Audio is essential.
Never producing. Recognition without production is half a skill.
Treating the deck as the goal. The deck is a tool. Fluency is the goal. Measure yourself by what you can read and say, not by review counts.
Sources
- Nation, I.S.P. (2013). Learning Vocabulary in Another Language. Cambridge University Press.
- Nakata, T. (2015). Effects of feedback timing on second language vocabulary learning. System, 48.
- Refold language learning method for a modern sentence-mining workflow.