Chinese text analyzer

How to read the result

The colours are not a score. They are a map of where the difficulty sits, and the useful question is not how many words are above me but what fraction, and are they load-bearing.

Research on second-language reading keeps landing on the same threshold: to read something comfortably on your own — following the argument, guessing the odd unknown from context — you need to know roughly 98% of the words on the page. That is one unknown word in fifty, about one every two or three lines. At 95% it is one in twenty, and that is where most people stop following the meaning and start assembling it.

So: count the words above your level, not the characters. More than one in twenty and the passage is work rather than reading. Then look at where they sit — ten unknowns spread evenly is a harder read than ten concentrated in one technical paragraph you can skip, and an unfamiliar word in the main verb stops you where one in a list of examples does not.

Why this splits words, not characters

Chinese hides its difficulty twice over. There are no spaces, so an unknown word does not announce itself as an unfamiliar shape — it looks exactly like the text around it. And you can know every character in a word and still not know the word: 东西 is east plus west and means “thing”; 马上 is horse plus on and means “immediately”. Neither meaning is recoverable from the parts.

That is why skimming a passage for “characters I don’t recognise” will tell you it is easy when it is not. There is a measurement behind this: across the 9,005 example sentences in our own library — one per word, written to fit each level — the characters barely get harder as the levels climb. In the HSK 7–9 sentences, 47% of all characters are HSK 1–2 characters and 83% are HSK 1–4; only about 5% belong to the advanced band at all. An advanced sentence is not made of advanced characters. It is made of ordinary characters in combinations you have not met — which is exactly what a word-level split shows you and a character count hides.

Using it to build graded reading

• Pre-learn the gap instead of grinding. A passage at 90% coverage teaches very little per hour, because too little of the context is solid enough to hang a guess on. Pull the twenty flagged words, learn those, then read it — you fixed the coverage rather than trying harder.
• Keep a text you gave up on. Run it again in a month. Watching the colours go pale is a more honest progress report than any level number.
• Click through on anything flagged. Every word links to its entry with pinyin, stroke order, audio and an example sentence.

What it cannot tell you

Levels come from the official HSK 3.0 word list, so anything outside those 11,000 words has no level to show — proper nouns, technical vocabulary and most classical usage fall through. Segmentation is not perfect either: names and rare compounds can be split wrongly, and a wrongly split word gets a wrongly optimistic level. And a level is not a difficulty score — grammar, sentence length and how much background knowledge a text assumes are all invisible here. Two passages with identical colour maps can read very differently.