← All posts

How many words do you need to be fluent in a language?

Ask ten teachers how many words fluency takes and you will get ten numbers between 500 and 50,000. The spread is not because nobody knows. It is because "fluent" is doing too much work in the question.

The numbers, and where they come from

Word frequency in every language studied follows the same lopsided shape: a small set of words does an enormous amount of the work, and the rest show up rarely. Counts from large collections of everyday speech land in roughly the same place each time:

The gap between the second and third number is the whole reason the estimates disagree. Holding a conversation and reading a novel are different skills with different vocabulary budgets, and "fluent" gets used for both.

The 95% ceiling is not as good as it sounds

Ninety-five percent coverage means one word in twenty is unknown. In a twenty-word sentence, that is one gap — usually survivable from context. But the missing word is rarely "the". It is the noun the sentence was about. Coverage statistics count tokens, not importance, and the words you are missing are disproportionately the ones carrying the meaning.

This is why learners who have "finished" a 3,000-word course still stall. The number was met; the specific words were wrong.

Which 3,000 matters more than the 3,000

A frequency list is built from an average of everybody. You are not everybody. If you talk about cooking, your personal high-frequency list contains words a general list ranks at 6,000. If you are learning a language to talk to a partner's family, the vocabulary you need is whatever that family talks about.

Two learners with the same 3,000-word count can have wildly different experiences. The one whose words came from their own life will feel fluent. The one whose words came from a textbook will feel like they are reciting.

How to find your own list

You do not need a corpus. Look at the messages you sent last week in your own language. The nouns and verbs that repeat are your actual high-frequency words. That is the list worth learning first — and it is almost never the list a course starts with.

Recognition is not the same as retrieval

There is a second reason the counts mislead. Most people measure vocabulary by recognition: shown a word, do you know it? But conversation requires the reverse — given a meaning, can you produce the word, in under a second, while also constructing a sentence around it?

Recognition vocabulary is typically several times larger than production vocabulary. A learner who "knows" 3,000 words by recognition may be able to produce 800 of them at conversational speed. That gap is the difference between understanding a reply and being able to give one.

Flashcards train recognition efficiently and production barely at all. Anything that makes you construct a sentence under mild time pressure trains production.

A better target than a number

Instead of chasing a word count, aim for a behaviour: have a real conversation where you needed a word, did not have it, and found it. Words acquired that way arrive with a retrieval cue already attached — you remember who you were talking to and what you were trying to say. Words acquired from a list arrive with nothing attached, which is exactly why they evaporate.

If you want a number anyway, use 800 as your first milestone. It is achievable in a few months at a relaxed pace, it is where conversation starts feeling possible rather than theoretical, and it is small enough that you can afford to pick the words yourself.

The short version