CEFR level checker for English texts

A textbook unit is labeled by the course, and a single text inside it is often one step above, on purpose — it is there to stretch. Look at the three sub-scores: if vocabulary says B1 and only grammar says B2, the text is B1 with one or two structures from the next level, and the Grammar tab shows exactly which and where. The scale under the badge also says whether the score sits low or high inside its level.

It is the share of running words in the text that a reader at that level knows — words at or below the level, plus any you marked as known. Research on reading puts comfortable, independent reading at 98% and reading with a dictionary at 95%. Below 95% the reader keeps losing the thread. The line under the figure says how many words, learned in order of how often they occur in this text, would get the reader to 95% and 98%.

They are not vocabulary to learn: a story with Mr Peterson, London and 1999 in it is not harder because of them. Names are recognized by their capital letter and a list of 12,000 names; acronyms like BBC and numbers are also left out of the count.

57 structures — from the present perfect and the passive to third and mixed conditionals, participle clauses, cleft sentences, inversion after "never" and the subjunctive. Their levels come from the CEFR-J Grammar Profile, and from the English Grammar Profile for the C1 structures. Click "Show" next to a structure to see it highlighted in the text.

A language model rewrites the text for the chosen level and is given the exact words the analysis found above it. The result is then measured by the same analysis, so you see whether it really reached the level and which words are still above it. Your original stays in the field until you press "Use this version", and you can undo it. Rewriting spends tokens; the price is shown on the button before you press it.

Choose the level of the class, open the "Words" tab and press "Handout (PDF)". You get one file with the text — words above the level in bold — a glossary with translations, a gap-fill exercise built from sentences of the text, and an answer key.

Yes. Word, PDF, EPUB, FB2 (also .fb2.zip), RTF, ODT, plain text and subtitle files open directly, and the analysis covers the whole text; very long texts are drawn in color up to the first 20,000 words. A scanned PDF has no text layer — take a photo of the page instead and the text will be recognized.

In this browser's storage — the same list that "Movie vocabulary from subtitles" uses, so a word marked here counts there too. They raise the "% known" figure and drop out of the glossary.

Yes, within limits we measured. A part-of-speech tagger trained on an open English treebank decides whether "address" is a noun (A1) or a verb (B1) — on held-out text this picks the right CEFR-J level 95% of the time, against 81% for simply taking the lowest level. For a few dozen words with a much harder second meaning ("succeed" someone, "jog" someone's memory, the "bank" of a river) the meaning is recognized from the words around it, and the card says which meaning it took. Set expressions ("according to") are one unit with their CEFR-J level. Idioms ("break the ice") are recognized as one unit with an English definition and listed apart in the "Words above the level" tab; their level is only an estimate, so it does not change the level of the text. An idiom used literally ("he broke a leg") can still be taken for the idiom, so the card says "part of the idiom" rather than hiding it.