Vocabulary size test

The test samples its word list instead of asking about every word. The list is cut into frequency bands of 2,500 base forms each, from the most common to the rarest. For each band, the share of its real words you tick, minus the share of made-up words you tick, is scaled up to the whole band. The bands are then added up.

Round 1 is the same for every length: 40 real words, two from each band, mixed with 14 made-up words and shown at once as tick buttons. Round 2 adds 60 real and 20 made-up words in the standard test, 30 and 10 in the quick one, 120 and 40 in the long one. No word appears twice in one test.

This is a yes/no test: each word asks a single question, known or not. That lets a short run of ticks estimate a vocabulary size in the tens of thousands, because every band stands for thousands of words like the ones shown.

Round 2 does not share its words out evenly. Every band gets at least one more word. In a standard test, a band whose two words got one tick and one blank gets about twice as many as a band answered all yes or all no. Its share is the least certain. Bands that count for more base forms get more as well. Statisticians call this Neyman allocation; applying it to word bands is a choice made for this test.

Take band 10, the base forms ranked 22,501 to 25,000. Suppose 7 of the 10 words shown from it were ticked, and 1 of the 34 made-up words. Its known share is 70% minus 2.9%, which leaves 67.1%. The band counts for 1,806 real base forms, its 2,500 ranks times the 72% the review kept, so it adds 1,211 words.

Subtracting the made-up share is a correction for guessing. Keuleers and colleagues used it in 2015 with Dutch speakers, in a study that also measured word prevalence, the share of people who know each word. Brysbaert and colleagues used it in 2016 for American English. Two further steps were added for this test. The shares are smoothed so that a rarer band never counts as better known than a more common one, and the total is rounded.

Any answer to "how many words do I know?" depends on what counts as one word. What the test counts are base forms, also called lemmas: the form a dictionary lists, standing for its inflected forms. The Cambridge Dictionary's example is build, the lemma behind builds, building and built.

Derived forms, such as builder from build, are lemmas of their own. Word families go further and fold a base word, its inflections and its derivations into a single count, so the same person scores far lower in them. Brysbaert and colleagues show the gap: their median American 20-year-old knew 42,000 lemmas but 11,100 base words.

Headwords are the entries a dictionary prints in bold, and dictionaries differ in what earns an entry. Proper nouns, such as the names of people and places, are not counted here, and neither are abbreviations. Multiword expressions, such as look after or in spite of, are not on the list either, so a phrase known as a whole adds nothing.

The English list ranks 50,000 base forms: the nouns, verbs, adjectives and adverbs that the English Wiktionary lists, as extracted by kaikki.org. They are ordered by word frequency in film and TV subtitles, counted in OpenSubtitles 2018 through the FrequencyWords lists, and cut into 20 bands.

A review made for this site went through a fixed random sample of every band, 720 candidates in all, against general English dictionaries. It dropped names, abbreviations, foreign words, inflected forms, misspellings, offensive words, sound words and fragments of phrases. It kept 83% overall: 94% in the most common band and 69% in the rarest. Each band counts at its kept share, so the whole list counts for 41,700 base forms, the most any result can reach.

Subtitles track the words people hear. Brysbaert and New (2009) found that subtitle frequencies predicted American word recognition times better than counts from written texts. For British English, van Heuven and colleagues built SUBTLEX-UK from BBC subtitles in 2014, and its counts explained British readers' recognition times better than the British National Corpus did.

Among native speakers of American English, the median is 42,000 lemmas at age 20 and 48,200 at age 60, according to a study of 221,268 people. Smaller studies, other units and this test's own example give the other rows.

Marc Brysbaert's group at Ghent University published the American figures in Frontiers in Psychology in 2016. Each session showed a random set of 67 words from a list of 61,800 lemmas, mixed with 33 non-words and scored with the same subtraction as this test. The median at 60 was about 6,000 lemmas higher than at 20.

Studies of other groups use other units. Milton and Treffers-Daller (2013) found a typical figure of about 10,000 word families among students at British universities. Goulden, Nation and Read (1990), as Brysbaert and colleagues report, found that 20 graduates averaged 17,200 base words.

None of these rows is on the same scale as a result from this test, which counts base forms on a 50,000-word list. The American study drew on 61,800 lemmas, while a result here can never pass 41,700.

Because they show how readily someone says yes. Nobody can know a word that does not exist, so a ticked made-up word is a false positive. False positives reveal overclaiming, the habit of ticking words that feel familiar without really being known. Researchers call such items pseudowords or nonwords.

Every tick on a made-up word lowers the result in every band at once. In a standard test, one tick among the 34 made-up words takes 2.9% off the known share of each band. Above 20% of made-up words ticked, the result carries a warning, since a share that large makes the estimate less certain.

The made-up words start as real base forms with one or two letter groups swapped for others that English uses in the same place. That is the idea behind Wuggy, the pseudoword generator of Keuleers and Brysbaert (2010), in a simpler version written for this test. The site's review then chose strings that follow English spelling but are no English word, 127 of them in all.

The instruction asks for words whose meaning you know, not words you have merely seen. Under that rule, a word that looks familiar but brings no meaning to mind counts as not known.

The figure the test gives is how many base forms of its English list you know, rounded to the nearest hundred. It comes from your ticks on the 100 words drawn from 20 frequency bands over the two rounds of a standard test, less a correction for every made-up word ticked. Take an illustrative visitor who knows every word of the 15,000 most common base forms, then 6% less in each rarer band, down to 16% of the last. This visitor also ticks 1 of the 34 made-up words. The test then gives about 28,800 words, likely between 24,700 and 32,800, or 69% of the list. That is a count of words on this list, not of a whole vocabulary.

The likely range comes from statistics computed for this test, not from a study. It is a 95% interval, the estimate plus or minus 1.96 standard errors, worked out from the spread of the answers in each band and among the made-up words. No study publishes a standard error for this kind of estimate, so the margin of error is the test's own, and it widens when few words came from a band.

The bars show each band's share counted as known, most common words first, and after smoothing they never rise from one band to the next. Below them come the real words you did not tick and every made-up word shown, with the ticked ones marked.

The "Take it again (new draw)" button draws the words again at random, so the result moves a little from one attempt to the next. The previous result in that language appears under the new one. The last 10 results stay in this browser's local storage, with language, length, estimate, range and made-up words ticked, but no word of the test.

Nothing is sent anywhere, and the page link carries only the test language and length. Results from other online tests rarely match this one: lists differ in size and unit, and a test without made-up words cannot correct for guessing.

The test measures receptive vocabulary, the words you understand when you meet them, also called passive vocabulary. Productive vocabulary is the smaller set you use yourself in speech and writing. A tick shows that a meaning is recognised, not that the word comes to mind when needed.

Brysbaert and colleagues judge that productive vocabulary is less than half the size of receptive vocabulary. They also describe yes/no figures as upper estimates, since knowing a word can be as shallow as knowing that it exists.

Productive vocabulary is what a talk or an essay draws on. For a talk, the speech time calculator turns a word count into minutes at a chosen speaking pace.

The result does not convert into CEFR levels. The Common European Framework of Reference for Languages describes what learners can do in a language, from A1 to C2, while this test counts words of a single list.

For second-language learners, Paul Nation (2006) estimated how many words following a text takes, at the point where nearly every running word is known. He put it at 8,000 to 9,000 word families for written text and 6,000 to 7,000 for spoken text.

LexTALE, published by Lemhöfer and Broersma in 2012, is a short yes/no test for advanced learners of English, with 40 words and 20 nonwords. They cite work by Mochida and Harrington (2006) in which yes/no scores correlated above 0.8 with a vocabulary test.

The language menu offers the same test in German, Spanish, French, Italian, Dutch and Polish. Each list is built the same way, from Wiktionary entries ranked by that language's subtitles, so a result compares only with other results in that language. For nouns in those languages, the noun gender checker gives the gender and the article.

Teachers see a limited vocabulary as a barrier to learning. Oxford University Press asked over 1,300 teachers for its 2018 report on the word gap. They judged that 49% of Year 1 pupils and 43% of Year 7 pupils had a vocabulary limited enough to affect their learning.

Those figures are teachers' views, not measurements, and 69% of the primary teachers thought the problem was increasing.

The national curriculum in England sets no number of words. It says that "reading widely and often increases pupils' vocabulary because they encounter words they would rarely hear or use in everyday speech".

This test has no norms for any age, so a child's result says only how much of this list is known today. The review dropped offensive words from the ones the test shows, but the rarer bands hold words that few children will have met.