How many words do you know?

Sixteen yes-or-no questions in any of 49 languages. Each test draws from an offline dictionary ordered by frequency. No account, nothing stored, about a minute.

Start with one

Spanish · Ochre French · Genêt Japanese · Murasaki German · Graphit Italian · Siena Korean · Jaju Chinese · Yanzhi Portuguese · Telha Russian · Suprematist

Or take the whole map. Every line is listed further down, including Georgian, Armenian, Welsh, Basque, Esperanto, Tagalog, Ancient Greek and Latin.

Why sixteen questions

After every answer the test fits a curve over that language's frequency-ordered word list. It looks for the rank where your chance of knowing a word is about half, then asks the next word near that point. The estimate is the area under that curve. That figure differs from the rank you reached, and it is larger.

Sixteen is a deliberate number. More questions narrow the reported range, and the return falls off fast:

Questions	Width of the range
8	6.0×
12	4.9×
16	3.6×	what this test asks
18	3.3×
26	2.3×

Twenty-six questions would buy 2.3×. That is a better number for a test nobody finishes. At sixteen the range comes out about 3.6 times wide from edge to edge, and the result shows that width.

What the test will not do

It will not offer you four options. Multiple choice hands you a one-in-four chance of being right about a word you have never seen, and that noise lands straight in the estimate. Every question here is yes or no.

It will not take your word for it, up to a point. Say yes to all sixteen, including the rarest word it can reach, and you get no number at all rather than a flattering one. A partial exaggeration it cannot catch: the figure comes out wrong but plausible, and sixteen answers hold nothing that could tell the difference.

It will not keep anything. The test runs in your browser, there is no account, and no answer is sent anywhere. The word list is a static file your browser downloads once.

It will not measure your grammar. The result carries a CEFR band, read from one number and nothing else. The thresholds are 750, 1,500, 3,000, 5,000 and 8,000 words. The Common European Framework describes reading, writing, listening and speaking; this measures none of them. When your range crosses a threshold the band is shown as a range too.

Where the words come from

Every language ships as an offline dictionary compiled from open sources. Wiktionary via kaikki.org for the entries, Tatoeba for example sentences, FrequencyWords for ordering, Lingua Libre for audio, JMdict for Japanese and the Diorisis corpus for Ancient Greek. Across all of them that is 1,416,110 dictionary entries, 1,130,186 example sentences and 24,903 human recordings. Everything is CC BY-SA; the dictionaries are derived works.

Each language runs as its own line on Cognate's map, and each test draws from that line's dictionary. The per-language pages below carry the entry counts, the frequency bands and a sample word from each.