Japanese Vocabulary: How to Remember Words That Stick

Most learners who say they cannot remember Japanese words are not wrong about the symptom. They are wrong about the cause. They picture memory as a container — some people have a big one, some a small one — and conclude that theirs leaks. What actually happens is simpler and more fixable: they study words in a way that guarantees forgetting, then read the forgetting as evidence about themselves.

Vocabulary is not stored by effort; it is stored by retrieval, spacing, and context. Get those three right and an average memory will carry you to several thousand words. Get them wrong and no amount of discipline saves you.

Why word lists fail

A word list asks you to do the hardest possible version of the task: bind an arbitrary sound to an arbitrary meaning with nothing to hang it on.

Take 高い (takai, "expensive, tall"). As an isolated pair, it is two syllables glued to two English words. There is no reason for the connection, so your brain — which is a meaning-making machine, not a filing cabinet — files it nowhere.

Now put it in a sentence: この本は高いです (kono hon wa takai desu, "this book is expensive"). The word now has a subject, a register (polite), a typical usage, and a small scene. When you recall the word later, you are not retrieving a bare pair; you are re-running a fragment of experience. That is a far easier retrieval, and it comes with free extras: you learn that 高い (takai) is followed by です (desu) rather than だ (da) in polite speech, and that it describes books and prices, not just "tall things."

A word learned in a sentence arrives with its grammar attached. A word learned from a list arrives with nothing.

There is one place a list genuinely helps: as an index of what to study next. Use it to decide what to learn, then go learn it inside a sentence.

The forgetting curve, without the mythology

The shape of forgetting is well established and not controversial: newly learned material drops off fast at first, then the loss flattens. Something you met once an hour ago is already shaky; something you have recalled five times over three weeks is close to permanent.

Two things follow, and they are the whole game:

First, forgetting is normal and expected, not a personal defect. If you meet a word once and cannot produce it tomorrow, that is the system working as designed. You have not failed; you have simply not finished. The correct response is not disappointment, it is scheduling.

Second, the review has to happen while the memory is fading, not while it is fresh. Reviewing a word you learned ten minutes ago feels great and does almost nothing, because the retrieval is too easy. Reviewing it two days later, when you have to reach for it, is what strengthens it. This is the single most counterintuitive point in vocabulary learning: easy recall is wasted effort, and the slight struggle is the point.

Be careful with the numbers you see online. You will find precise forgetting percentages attributed to nineteenth-century experiments, quoted to the decimal in blog posts. The general shape — steep early loss, flattening later — is solid. Exact percentages for you, on this word, on this day are not something anyone can hand you. Treat the curve as a design principle, not a measurement.

Spaced repetition without drowning

Spaced repetition software (Anki, and the SRS built into most vocabulary apps) automates the scheduling question: it shows you a card, you grade how hard it was, and it pushes the next showing further out — a day, three days, a week, a month. Done well, this is the most efficient way to hold a large vocabulary. Done badly, it is a treadmill that eats an hour a day and teaches you to hate the language.

The failure mode is almost always the same: too many new cards at once. Add thirty new words a day and within two weeks your daily review queue is three hundred cards. You spend your session clearing a backlog, you learn nothing deeply, and you quit. The mathematics of compounding, which is what makes SRS powerful, is the same mathematics that makes an overstuffed deck unpayable.

Practical guardrails:

If you would rather not run software at all, a paper schedule works. Learn a small batch, then revisit it the same day, the next day, three days later, a week later, and a month later. What matters is that the intervals grow and that you actually attempt recall rather than re-reading.

One word, several channels

Every time you meet a word, you have a choice about which channel you use: you can see it, hear it, say it, write it, or type it. Using two channels beats using one, and using it in production beats using it in recognition.

This is why reading a word list feels hollow: it uses one channel (sight) and one direction (recognize). Typing the romaji of a sentence uses sight, sound, and motor production at once — you have to know the reading, not just the meaning, and you have to produce it, not just recognize it. That extra effort is exactly the kind of "desirable difficulty" that makes the memory durable.

The same principle explains why producing a word out loud while you look at a street sign, or writing three sentences with a new word, locks it in faster than another pass through the list.

Context does more than you think

Grouping words by topic helps, but not for the reason people assume. The benefit is not the topic label; it is that related words start to define each other.

Learn these together and each one has neighbours:

Now here is the same idea for a different domain:

When you retrieve 切符 (kippu, "ticket"), you drag 駅 (eki, "station") and 電車 (densha, "train") along with it. You have turned a single-point lookup into a small network, and networks survive forgetting far better than points.

Watch for the trap in the second table, though: 切符 (kippu, "ticket") and 切る (kiru, "to cut") look related and are not. False friends inside Japanese are as common as false friends between Japanese and English, and grouping by appearance rather than meaning builds exactly the wrong network.

"I have a bad memory" — what that usually means

If you have read this far and still think your memory is the outlier, test the claim against these five patterns. In practice, almost every learner who says "I have a bad memory" is describing one of them.

1. Too many items at once. Thirty words a day is not ambitious, it is a queue you cannot service. Cut to ten and watch your retention roughly triple. The bottleneck was never capacity; it was scheduling.

2. No retrieval practice. Re-reading a list feels productive because the material becomes familiar. Familiarity is not memory. If you never close the book and try to produce the word, you have been studying recognition of the page, not the word.

3. No context. A word with no sentence has no retrieval path back. This is the most common single cause, and the easiest to fix.

4. Recognition only. You can pick 単語 (tango, "word") out of a list but cannot produce it from "word." Those are two different skills and only the second one works in conversation or typing. Train production.

5. Studying when exhausted. Vocabulary work is retrieval work, and retrieval degrades badly with fatigue. Twenty focused minutes beats an hour of glazed scrolling.

There is a sixth, less comfortable possibility: you are comparing yourself to a number. People who "know 10,000 words" are usually counting words they recognise in reading, not words they can produce on demand. Those are different inventories, and the second one is always much smaller. Comparing your productive vocabulary to someone else's recognition vocabulary is a good way to feel bad about a normal amount of progress.

A routine that fits in fifteen minutes

• Five minutes of review. Clear yesterday's and last week's items first, always before anything new.
• Five minutes of new material. Ten new words, each met inside a real sentence, never as bare pairs.
• Five minutes of production. Type or write two or three sentences using today's words. Not exercises — sentences about your actual day.

Fifteen minutes daily will outrun two hours every Saturday, because the second schedule lets you forget everything in between.

And when you do forget something you had learned — which you will, repeatedly, for years — the correct interpretation is not "my memory is bad." It is 忘れた (wasureta, "I forgot") followed immediately by scheduling a review. 忘れる (wasureru, "to forget") is a verb, not a verdict.

Common mistakes for English speakers

1. Studying kanji, reading and meaning as three separate facts. 台所 is not three cards (the shape, "daidokoro", "kitchen"). It is one card with the reading and meaning attached, learned inside a sentence.

2. Trusting katakana loanwords to be free. テレビ (terebi, "television") is genuinely easy. マンション (manshon) is not — it means a condominium, not a mansion. Loanwords are fast to recognise and frequently false.

3. Avoiding words because they look long. 美味しい (oishii, "delicious") is no harder to learn than すし (sushi, "sushi"). Length is not difficulty; frequency is difficulty.

4. Learning only nouns. Beginners collect nouns because nouns are picturable. But conversation runs on verbs and adjectives, and a learner with 500 nouns and 40 verbs cannot say anything. Deliberately over-weight verbs early.

5. Adding cards from a JLPT list you have not studied. A test vocabulary list is a checklist for your course, not a learning sequence. The words arrive with no context, and those are the cards you will fail forever.

6. Reading hiragana as if it were optional scaffolding. 高い (takai, "expensive, tall") is not "taka-i, whatever." If the reading is not automatic, the retrieval is not happening — which is why kana memory methods come before vocabulary strategy.

Summary

Vocabulary sticks when you retrieve it, space it out, and meet it in context. Word lists tell you what to learn; sentences teach it. Spaced repetition handles the schedule, but only if you cap the daily intake at something you can actually clear. Two channels beat one, production beats recognition, and a forgotten word is a scheduling signal rather than a personal failure.

If your current method is a list and a highlighter, the upgrade is smaller than you expect: take ten words, put each in a sentence, and come back to them tomorrow and next week.

Romaji.live was built around the fact that a word is easier to keep when it arrives inside a sentence. Every practice item is a real Japanese sentence you type word by word in romaji, so you read it, hear it, and produce it in the same pass — three channels instead of one. Start with the kana table if the readings are not yet automatic, then take a session at romaji.live practice. For the schedule side of the problem, see how long it takes to learn Japanese; for the input that feeds vocabulary, see reading practice.