Méthodologie
How the vocabulary test works
This page describes exactly what the test measures, how the score is calculated, and the published research behind our CEFR mapping. Written so you can judge how much to trust your result.
The Yes/No format
You're shown 60 items — real words and phonotactically plausible non-words mixed together. For each one you answer yes (I know this) or no. That's the whole instrument. It's fast (~4 minutes), language-agnostic, and doesn't punish typos or slow typing.
The Yes/No format is well established in vocabulary assessment research — Meara & Buxton, 1987 introduced it, and it's been refined by many researchers since.
Why pseudowords?
The obvious failure mode of Yes/No testing is overclaiming — some people say "yes" to almost everything, either strategically or from wishful thinking. To correct for this, we mix in pseudowords: made-up words that sound plausible in the target language but don't actually exist. Every "yes" on a pseudoword tells us something about how noisy that user's answers are.
Your score is Δ = (h − f) / (1 − f), where h is the hit rate on real words and f is the false-alarm rate on pseudowords. Δ is clamped to [0, 1]. If you claim more than 30% of pseudowords, we mark your result low confidence rather than issuing a firm CEFR band.
How we map frequency ranks to CEFR
The test words are sampled from OpenSubtitles frequency lists — a widely-used, freely licensed corpus of word frequencies. Each CEFR level corresponds to a rank band:
| CEFR | Rank band | Published word-family range |
|---|---|---|
| A1 | 1–500 | 500–1,000 |
| A2 | 501–1,500 | 1,000–2,000 |
| B1 | 1,501–3,000 | 2,000–3,250 |
| B2 | 3,001–5,000 | 3,250–5,000 |
| C1 | 5,001–8,000 | 5,000–8,000 |
| C2 | 8,001–12,000 | 8,000–16,000 |
Ranges above are for English; the other four languages use the same shape at v1. See the source data for details.
Where the CEFR ranges come from
The word-family ranges per CEFR level are drawn from a small number of published studies that genuinely disagree with each other. We commit to a single range per level based on the mainstream literature — Nation (2006), Milton (2010), Schmitt (2010) — and we render the whole range on your results page rather than picking a single number. That uncertainty is real, and pretending otherwise would manufacture false precision.
What this test is not
- Not a certification. Nothing here counts for a visa, university admission, or job application.
- Not an assessment of grammar, listening, speaking, or writing. You can have B2 vocabulary breadth and A2 speaking fluency, or vice versa.
- Not affiliated with any exam board. Where we mention TOEFL, Cambridge, IELTS, DELF, DELE, CILS, PLIDA, or the Polish state exam, it's to give you a rough comparability point — those exams are properties of their respective bodies.