How many Chinese words do you need to speak?

By Carson 木林森 — Middlebury College, ICLP at National Taiwan University, Yenching Academy of Peking University · July 2026

85% of what a learner actually says is 200 words. Not the 200 most frequent words — the 200 with the most leverage, which is a different list and a much more useful one. All 200 are on this page, with pinyin, meanings and a real phrase for each.

The bands

Each row is the top N words by leverage, what share of a learner's spoken output they cover, and how many of the 4,991 words in an HSK 1–6 course they reach once you count everything they help build.

What "leverage" means, and why not frequency

A frequency list tells you which words are common. It cannot tell you which ones get you more language, because that is a property of what you can do with a word, not how often it shows up. The move worth optimising for is the naming loop:

Point at something you cannot name → ask what it is called → ask again, slower → keep the person talking long enough to be told.

Run that loop and every conversation becomes a vocabulary source. So the ranking scores a word by how much it contributes to running it. That is why 吗 outranks 心 here: 吗 appears in dozens of conversational lines and builds no other words, while 心 builds 42 words and appears in almost none. A frequency list would reverse them, and the learner would be able to read more and say less.

Words, not characters

This model ranked characters until July 2026, and the character version put 么 at #1, glossed "suffix, used to form the interrogative 什么". True about a morpheme, and useless to hand a learner — nobody says 么 on its own, and the word a person actually says, 什么, appeared nowhere in the ranking. Ranking word nodes fixed it without a single hand-written filter: bound morphemes simply are not words, so they cannot rank. Single-character words — 这, 个, 吗, 是 — still dominate the top, as they should.

How it was measured, and what that does not prove

"Share of what you say" is the proportion of word tokens across 225 lines of shipped learner conversational moves that the band covers. That basis is worth stating plainly, because it bounds the claim in two ways:

• The corpus is this course's own conversational material, not a recording of strangers talking. It is a measure of how efficiently the course's own speaking curriculum is sequenced, and only an estimate of anyone else's conversations.
• Coverage saturates at 100% by the 500-word band — not because 500 words is all of Chinese, but because only a limited number of distinct words appear in the corpus at all. Read the top of the table, not the bottom.

What it does establish is the shape of the curve, and the shape is the actionable part: the first two hundred words are worth wildly more than the second two hundred, and choosing them by leverage rather than frequency changes which two hundred you get. It is also audited against the shipped course — all 200 are taught, and most are met in the first quarter of it.

The 200 words

Ranked by leverage. Each links to its full dictionary entry.

Frequently asked questions

How many words do you need for HSK 4?

About 1,200 cumulatively — 150 at HSK 1, 300 by HSK 2, 600 by HSK 3, 1,200 by HSK 4, 2,500 by HSK 5, 5,000 by HSK 6. Those are exam targets and a different question from what you need to speak. The full lists are free: HSK 1, 2, 3, 4, 5, 6.

So can I stop at 200 words?

No — you can start at 200. The band covers the share of what you say, not the share of what you understand, and listening is the harder half. What the number buys you is the ability to keep a conversation running early, which is what turns every future conversation into a lesson.

What order should the rest come in?

The whole dependency-ordered curriculum — all 829 topics from tones through HSK 6 — is published here.