Readability Score

Paste a paragraph or a whole article. For Latin-script text you get the three standard readability formulas. For Chinese you get a separate index built from things Chinese actually has, and an explicit statement that it cannot be compared with a Flesch score.

What the standard formulas actually measure

Flesch Reading Ease combines two ratios: words per sentence, and syllables per word. Flesch-Kincaid rearranges the same two inputs to produce a US school grade level, and Gunning Fog swaps syllables per word for the proportion of words with three or more syllables. All three are proxies for the same intuition — long sentences and long words are harder — and all three were validated on English prose in the mid twentieth century.

They are useful precisely because they are crude. Nothing in the formula understands your argument, your structure or your vocabulary choices, so a high score is not a claim that your writing is good; it is a claim that your sentences are short and your words are phonetically simple. Used as a check against your own earlier draft, that is informative. Used as a target to optimise, it produces text that scores well and says nothing.

Why they cannot work for Chinese

Both inputs are undefined. Chinese has no syllable count in the sense the formula needs — a character maps to one syllable when spoken, so syllables per word collapses into a constant that carries no difficulty information at all. And there is no whitespace-delimited word to divide by, so word count itself has to be invented before the formula can even run.

What most tools do with Chinese input is not refuse it. They run the same arithmetic on whatever their tokeniser produced, and return NaN, a large negative number, or an absurdly high score for a genuinely difficult text. That is worse than no answer, because a number on a screen gets acted on. On this page the Flesch, Flesch-Kincaid and Gunning Fog fields say not applicable when the text is mostly Chinese, and we would rather show you that than a fabricated figure.

What we measure for Chinese instead

Three things Chinese does have. Average sentence length in characters, because sentence length is the one difficulty signal that transfers across writing systems — and Chinese prose that runs three clauses deep without a full stop is hard for the same reason it is hard in English. Common-character coverage, the share of characters drawn from a high-frequency list, because rare characters are the Chinese analogue of rare words. And punctuation density, because a passage with almost no punctuation is a passage the reader has to segment unaided.

The index weights those at 45, 40 and 15 per cent respectively, and the exact figures are shown so you can see what moved. Two honest limitations: there is no industry-standard readability scale for Chinese to calibrate against, and our common-character list is our own compilation rather than an official curriculum table. Treat the index as a way of comparing two drafts of your own writing, not as a grade level, and never as a number comparable to Flesch.

Reading the sentence-level output

The longest-sentences table is usually more actionable than any single score. Readability figures are averages, and an average hides the specific sentence that made a reader give up. Sorting by length puts that sentence in front of you, and splitting one forty-character clause into two is a bigger improvement than shaving a point off an index.

One thing to check when the numbers look strange: if your text has no terminal punctuation, everything is measured as a single sentence and every per-sentence figure is meaningless. We flag that in the notes rather than quietly reporting the resulting nonsense.

Nothing is uploaded

The formulas run in this tab. That matters here because readability is usually checked on material that is not public yet: submissions before review, patient information leaflets, policy documents, exam papers, and marketing copy under embargo.

Open DevTools and watch the Network tab while you type, or disconnect entirely and reload. The page keeps scoring, because scoring never needed a server.