Chinese word statistics

How to read this. Coverage counts every occurrence of a word: in 我的妈妈的哥哥 (我 / 的 / 妈妈 / 的 / 哥哥), knowing 的 means knowing 2 of 5 words, 40%. It is the share of word tokens in a corpus that you would recognise knowing the selected set, so it measures how much of a film or text you actually follow. A dictionary word counts only if it is in the set itself; a plain combination such as 一个 or 很多 counts once all of its parts are, and regular forms count when their base word is known: 他们 from 他, 看看 and 看一看 from 看, 研究研究 from 研究, 这儿 from 这. Fillers and interjections (嗯 呃 唔 啊 哦 噢 喔 哇 咦 诶 欸 唉 …, 58 in all) count as known for every word set: they are not really vocabulary, and a listener gets them without study. Names and numbers written as digits count as unknown, so the figures are a conservative estimate. Character coverage is the share of character tokens whose character appears in the set. A common rule of thumb is that around 95% coverage is needed to follow a text comfortably and 98% to read it with ease.

Sources: SUBTLEX-CH: subtitles of 6,243 films and TV episodes; Leipzig Corpora Collection: Chinese news (2020) and web text (2015), 200,000 sentences. HSK word lists from the complete-hsk-vocabulary project (HSK 2.0 levels 1–6, HSK 3.0 levels 1–6 and band 7–9). The HSK 2.0 list also leaves some words implicit that no rule can recover (说话 but not 说, 没 for 没有), which lowers its coverage somewhat. Generated 2026-10-27.