Why the number is smaller than expected
Word frequency is extremely uneven in every language. A small number of words appear constantly, and the rest appear rarely, which means coverage rises very quickly at first and then very slowly. The first thousand words of Japanese do an enormous amount of work; the fifth thousand does comparatively little for conversation.
Everyday speech is also narrower than writing. It repeats, it relies on context, and it uses general words where writing would use precise ones. A conversation about a broken washing machine mostly uses ordinary verbs and a handful of nouns, not specialist vocabulary.
This is why learners are often surprised by how much can be said with very little, and equally surprised that adding vocabulary does not make speaking flow.
Realistic numbers for different goals
Any number here is approximate and depends on what counts as a word, but the relative sizes are informative. What matters is the distinction between the vocabulary you produce and the much larger one you need to understand.
Note how the two columns diverge. Understanding a news broadcast needs several times the vocabulary of holding a conversation about your day.
| Goal | Words to produce | Words to understand |
|---|---|---|
| Travel basics | 300-500 | 800-1000 |
| Everyday conversation | 2000-3000 | 5000-6000 |
| Comfortable social life | 3000-4000 | 8000-10000 |
| Work in Japanese | 4000-6000 | 10000-15000 |
| News and non-fiction | Not required | 15000+ |
The production column is the one that governs speaking, and it is much smaller than most learners assume they need before they are allowed to start.
The gap between knowing and saying
Most learners have a recognition vocabulary several times larger than their production vocabulary. That is normal and it is also the whole explanation for the common experience of understanding a conversation perfectly and being unable to contribute to it.
A word moves from recognition to production through use, not through review. Seeing it on a card again strengthens recognition; saying it in a sentence you built yourself is what makes it available when you need it.
- Recognition: you understand it when you see or hear it.
- Cued production: you can produce it when prompted with the meaning.
- Free production: it arrives on its own when you need it in speech.
- Only the third one helps you talk, and only speaking builds it.
When the problem is not vocabulary
Before adding words, it is worth checking whether the words were missing at all. Write down the sentences you could not say, look at them an hour later, and see how many you could build with what you already know. For most learners the answer is most of them.
That result points at retrieval speed and sentence planning rather than coverage, and those have different fixes: more speaking, more automation of frequent patterns, and simplifying the thought rather than searching for the perfect word.
| Japanese | Romaji | Meaning |
|---|---|---|
| 言葉は知っているのに、出てこないんです。 | Kotoba wa shitte iru noni, dete konai n desu. | I know the word, but it does not come out. |
| 簡単な言い方に変えてみます。 | Kantan na iikata ni kaete mimasu. | I will try changing it to a simpler phrasing. |
| 似た意味の言葉でもいいですか。 | Nita imi no kotoba demo ii desu ka. | Is a word with a similar meaning all right? |
| あとで調べます。 | Ato de shirabemasu. | I will look it up later. |
| その言葉、なんでしたっけ。 | Sono kotoba, nan deshita kke. | What was that word again? |
| 説明で言ってみます。 | Setsumei de itte mimasu. | I will try describing it instead. |
Which words are worth adding
When vocabulary genuinely is the gap, the words worth adding are not the next thousand on a frequency list. They are the words your own conversations proved you needed, because those are the ones your life will ask for again.
A second high-value category is the general-purpose words that let you talk about anything: words for degree, tendency, comparison and cause. Thirty of these do more for conversation than three hundred specific nouns.
- Words that failed you in an actual conversation this month.
- Words for your own work, hobbies and daily life, however uncommon.
- General-purpose words: degree, frequency, comparison, cause.
- Verbs before nouns; a verb carries a sentence, a noun does not.
- Words you have now heard three times without learning them.
Building production vocabulary rather than a bigger list
The practical version of everything above is a loop: speak, notice what failed, decide whether it was coverage or retrieval, and act accordingly. Without the middle step, every gap looks like a vocabulary gap and the response is always another list.
This is where a spoken session with a record afterwards earns its place. Conversations with an AI Sensei in Unihongo's immersive 3D classroom produce a session report showing what you said and what you were reaching for, and words from it can be saved straight into flashcards, so the deck is built from your own failures rather than from a generic frequency list.
Because the cards come from sentences you actually tried to say, they arrive already attached to a situation, which is what moves a word from recognition into free production instead of just strengthening a card you can already read.
Track words you have used in speech, not words you have reviewed. It is a smaller number and a much better predictor of how a conversation will go.