Vocabulary Size Test: method, data and credits
This page explains how the test estimates your vocabulary size, what data it is built on and how to reuse the item bank.
Start the testMethod
- We ranked 40,000 English lemmas (word families such as go/goes/went/gone) by how often they appear in general English, and split them into 12 frequency bands: 1,000-word bands up to 6,000, then wider bands up to 40,000.
- You answer Yes/No for words drawn from the bands, mixed with made-up words (pseudowords). Pseudowords measure how often someone says yes without knowing, and the score is corrected for it.
- A two-stage adaptive design: 24 words from all bands first, then 32 words around your level (70 items with pseudowords). The extended test adds 160 items around your level.
- Scoring fits a knowledge curve (probability of knowing a word by its frequency rank) with a Bayesian model; the size is the sum of that curve over the 40,000 lemmas. We show the middle 80% of the result as a range.
- The CEFR level is an estimate from the words you know among the 5,000 most frequent, after Milton & Alexiou (2009). It is not an official level. We do not estimate TOEIC or IELTS scores.
How accurate is it?
There is no real-world data yet: these figures come from 1,000 simulated test-takers. The quick test has a median error of about 19%, the precise test about 16% (11% above 12,000 words). Answers are stored anonymously (word, yes/no, time) so the item difficulties can be recalibrated.
Bands and items
| Band | Ranks | Words shown |
|---|---|---|
| 1 | 1–1 000 | 80 |
| 2 | 1 001–2 000 | 80 |
| 3 | 2 001–3 000 | 80 |
| 4 | 3 001–4 000 | 80 |
| 5 | 4 001–5 000 | 80 |
| 6 | 5 001–6 000 | 80 |
| 7 | 6 001–8 000 | 80 |
| 8 | 8 001–10 000 | 80 |
| 9 | 10 001–14 000 | 80 |
| 10 | 14 001–20 000 | 80 |
| 11 | 20 001–28 000 | 80 |
| 12 | 28 001–40 000 | 80 |
Pseudowords: 160
Word filter
Words that should not appear in a test were taken out of what is shown (they still count in the 40,000-word list): profanity and slurs (LDNOOBW list plus a hand review), names and brands, abbreviations and slang, and words that are not in a standard dictionary. Words removed: 3 255.
- too short: 5
- profanity / slurs: 127
- abbreviations: 63
- names and brands: 344
- slang: 51
- sensitive topics: 57
- not in a standard dictionary: 2 608
Data and credits
- Word frequencies: wordfreq by Robyn Speer (CC BY-SA 4.0), which includes data from Google Books Ngrams, Wikipedia, OpenSubtitles, SUBTLEX (Brysbaert et al.), the Leeds Internet Corpus, ParaCrawl and Twitter.
- Word lists: New General Service List 1.2 and the NGSL-SFI 31K list by Browne, C., Culligan, B. & Phillips, J. (CC BY-SA 4.0), newgeneralservicelist.com.
- Dictionary and name checks: Webster's Second International Dictionary (1934, public domain). Profanity list: LDNOOBW (CC BY 4.0).
- CEFR mapping: Milton, J. & Alexiou, T. (2009), EUROSLA Monographs Series 1; Milton, J. (2010).
Reuse the item bank
The item bank (frequency bands, ranks and pseudowords) is derived from wordfreq and NGSL, so it is released under the Creative Commons Attribution-ShareAlike 4.0 licence. You can download it as JSON:
https://toeicexamstore.com/api/v2/vocab-test/bankCC BY-SA 4.0 licence
Vocabulary size estimates are statistical estimates, not a certified score.