SRS for Japanese: Building Decks That Actually Work

What spaced repetition does and does not do

Spaced repetition schedules a question at widening intervals, so an item you would have forgotten in three days is asked again on day three, then day eight, then a month later. The value is arithmetic rather than magical: a large stock of half-known words stays reachable for a few minutes of daily effort. That is genuinely useful, because vocabulary is the part of Japanese that leaks fastest when you stop meeting it.

What the software cannot do is give a word meaning. A card can only test a link you already have, and a link you have never formed cannot be strengthened. It also cannot make a word available in speech, because recalling a meaning from a prompt is a different act from producing the word under time pressure while planning the rest of the sentence. Decks stall when learners quietly expect them to cover both jobs.

One fact per card, always

A card that asks for a reading, a meaning and an example at once cannot be graded honestly, because you will half-know it and press a button anyway. Split it. One prompt, one answer, one judgement. For Japanese this usually means a card that shows a sentence with the target word and asks for the reading and meaning of that word alone, with everything else on the card already known to you.

Keep the rest of the sentence easy on purpose. A card whose supporting words are also new is really three cards pretending to be one, and it will fail repeatedly for reasons that have nothing to do with the word you were trying to learn. This is the single biggest quality difference between a deck that survives and one that becomes a wall of red.

Card types and what each one trains

Those six lines are good production candidates because you can name the moment you would need each one. Recognition and production are separate cards because they are separate abilities, and mixing them into one card lets you pass on the easier half. Keep production cards to a deliberate minority of the deck, chosen because the word belongs in your own speech, not because a list told you it was common.

Well-built cards look like this

Each of these rows is a card front. The target word sits in an ordinary sentence, the grammar is simpler than the word being learned, and the sentence is short enough to read aloud in one breath. Say every card out loud when it appears; a silent review trains only your eyes.

Compare those with a bare word and a two-word gloss. The bare word gives you nothing to hang the memory on, no particle, no register and no sense of what usually appears beside it, so it is exactly the kind of card that becomes a leech three months later.

Why premade decks stall

A downloaded deck of several thousand words looks like a shortcut and behaves like a debt. The sentences were written for someone else's life, the words arrive in an order set by frequency rather than by your experience, and most items are first encounters rather than reviews. You end up using a maintenance tool to do teaching work, which is the job it is worst at.

There is a reasonable middle path. Use a small premade deck for the first few hundred core words, where almost any beginner needs the same material, then switch to cards you make yourself from things you have read, heard or said. A card born from a real encounter already has context attached, which is why it needs fewer repetitions than an imported one.

• Add a word only after you have met it somewhere outside the deck.
• Keep the source on the card: the chapter, the article, or the conversation it came from.
• Write the example sentence yourself where you can, even if you copy the structure from the source.
• Suspend rather than study anything from an imported deck that you cannot picture using.
• Cap the imported portion so your own cards always outnumber it within a few months.

The daily number that survives a year

New cards are a promise about future workload. Every card you add today returns several times over the following months, so a mature deck settles at roughly ten times your daily new-card count in reviews. Ten new cards a day is a comfortable habit; forty new cards a day is a part-time job you will resign from in about six weeks, usually during a busy fortnight when the backlog becomes visible.

Pick the number you could still do on your worst realistic day, then keep it constant rather than surging on free weekends. Consistency matters more than volume, because a deck's whole value is the intervals, and intervals only work if the reviews happen when they are due. If you must adjust, change the number for a month at a time, not day by day.

Leeches, and the mercy of deleting

A leech is a card you keep failing. It usually means one of three things: the card asks two questions at once, the word is too abstract to picture, or you have never met it outside the deck. Grinding it harder does not fix any of those, so the cure is to change the card or remove it. Words genuinely worth knowing come back to you in reading and conversation, and the second encounter is a far better place to start again.

Some words are simply hard to card. Vague adverbs and discourse words carry a feel rather than a definition, and they are learned by hearing them used, not by matching them to an English gloss. Put them in a sentence you have heard, or leave them out of the deck entirely and pick them up in conversation.

Each of those words is nearly impossible to define on a card and quite easy to absorb from a sentence you have said aloud a few times. When a word resists a definition, give it a situation instead.

Common deck problems and their fixes

Keep the session under a cap

Set a time limit before you open the deck, somewhere near fifteen or twenty minutes, and treat any session that regularly exceeds it as a signal to lower new cards rather than a reason to feel guilty. The purpose of the cap is not comfort. It is that reviewing while tired produces careless grading, careless grading corrupts the scheduling, and a corrupted schedule turns the deck into noise.

Split the time if it helps: reviews on the train in the morning, new cards in the evening when you can write proper example sentences and record audio. Whatever the split, finish the reviews before adding anything new, because the due cards are the ones the whole system exists to protect.

• Do reviews first, new cards last, every single day.
• Grade honestly; a hesitant answer is a fail, not a pass.
• Say the sentence aloud before you reveal the answer.
• Stop at the cap even with cards remaining, and lower the new-card number tomorrow.
• Once a month, scan the deck and delete anything you no longer want to be able to say.

Getting the words out of the deck and into your mouth

A word you can recognise on a card is not yet a word you own. Ownership happens when you choose it under pressure, in a sentence you are inventing while someone waits for you to finish. So build a bridge: at the end of every review session, pick five words and use each one in a sentence about today. Not the example sentence on the card, a new one about your own morning, your own work, your own weekend.

Then say those sentences to someone, or to an AI teacher, and let the corrections come back into the deck. The loop closes properly this way: input gives you the word, the deck keeps it available, speaking makes it yours, and the mistakes you make while speaking become the next batch of cards. A deck with no output step keeps producing learners who can read a menu and cannot order from it.

A teacher's likely replies are worth expecting: はい、自然ですよ (hai, shizen desu yo, yes, that sounds natural), 少し硬いですね (sukoshi katai desu ne, it is a little stiff), or もう一度お願いします (mō ichido onegai shimasu, once more please). Any of the three tells you something a card never could.

A deck you can keep for years

• Add cards only from material you have met: a chapter, an article, a conversation.
• One question per card, a short sentence around it, audio wherever you can.
• Keep new cards small and constant, and protect the reviews above everything else.
• Delete leeches without ceremony and let real encounters bring the word back.
• Finish every session by speaking five of the words in sentences about your own day.