Vocabulary Quiz

The quiz above asks whether you know particular words. This is what happens when researchers try to count all of them.

A 2016 crowdsourced experiment with 221,268 participants puts the average at around 42,000 lemmas for a twenty-year-old native speaker of American English, rising to about 48,200 by age sixty. That is one answer among many, because published estimates of vocabulary size run "from less than 10 thousand to over 200 thousand", and almost all of that spread is an argument about what counts as a word.

Key Takeaways

• Brysbaert and colleagues put an average twenty-year-old native speaker of American English at about 42,000 lemmas, and a sixty-year-old at about 48,200. That is roughly one new lemma every two days across those forty years.
• The same paper notes the published range runs "from less than 10 thousand to over 200 thousand", driven by the definition of a known word, the age of the person, and how much language they take in.
• Count word families instead of lemmas and the number collapses. The same study's 42,000 lemmas come from 11,100 word families, and Paul Nation reports well-educated native speakers knowing around 20,000 word families.
• Nation's finding is conditional and worth quoting as such: if 98 percent coverage of a text is what unassisted comprehension needs, then reading takes 8,000 to 9,000 word families and listening takes 6,000 to 7,000.
• The most frequent 2,000 word families of the British National Corpus cover only about 83 percent of the running words in a newspaper. Getting from there to 98 percent is most of the work.

How many words does the average person know?

Around 42,000, if you count lemmas and you are twenty years old.

That figure comes from Marc Brysbaert and colleagues, who ran a vocabulary test online and collected results at a scale no laboratory could reach: "All in all, we tested 221,268 individuals who returned 265,346 sessions." An average twenty-year-old native speaker of American English knew about 42,000 lemmas. A sixty-year-old knew about 48,200, "a difference of 6,200 lemmas in 40 years' time (or about one new lemma every 2 days)".

The average hides a wide spread. The same paper reports the range running "from 27,000 lemmas for the lowest 5% to 52,000 for the highest 5%", so two people the same age can differ by roughly a factor of two.

A lemma is the uninflected form from which the inflected ones are derived. Count walk once and you have also accounted for walks, walked and walking. It is a deliberately conservative unit, and choosing it is why the number lands where it does.

How is vocabulary size measured in the first place?

By asking people whether they know words, then subtracting the ones they claim to know that do not exist.

The Brysbaert study used the standard shape of this research. Each participant saw "a random list of 67 words and 33 non-words and asked to indicate which words they knew". The non-words were machine-generated to look plausible, participants were warned they were there, and they were told a yes to one would be punished. Then the correction: "the proportion of yes-responses to the non-words was subtracted from the proportion of yes-responses to the words."

The authors give a worked example, quoted here as printed: "someone who selected 50/67 words and said yes to 1/33 non-words, obtained a score of 0.746 - 0.033 = 0.713, indicating that they knew 71.3% of the words." One in thirty-three is nearer 0.030 than 0.033, so the subtracted figure in that illustration is rounded up, but the method is the point. That percentage is then applied across the full list the sample was drawn from, in this case 61,800 lemmas and more than 100,000 non-words.

Two things follow. First, a vocabulary estimate is always an extrapolation from a sample, never a count, which is the same limitation the IQ test explainer describes for short cognitive tests. Second, the fake words are not a trick. They are the standard defence against the fact that people are confidently wrong about which words they know, and a test without them produces an inflated figure by construction.

Why do vocabulary size estimates vary so wildly?

Because "word" is not one unit, and the studies are not all counting the same thing.

Brysbaert and colleagues state the dependencies plainly: the estimates depend on "(a) the definition of word known, (b) the age of person, and (c) the amount of language input sought by the person". Their own summary table of earlier work runs from 14,400 lemmas at one end to 215,000 words at the other.

The largest single lever is the unit:

Unit	What it groups together	Effect on the count
Token	Every word as it appears, repeats included	Enormous, and not a measure of knowledge
Lemma	An uninflected form plus its inflections	The unit behind the 42,000 figure
Word family	A base word plus its inflections and regular derivations	Groups the most, so gives the smallest count

On word families, the numbers land somewhere else entirely. Brysbaert and colleagues derive their own 42,000 lemmas from 11,100 word families, and Paul Nation reports that reasonably conservative estimates from methodologically sound studies "indicate that well-educated native speakers know around 20,000 word-families", against attempts to size English itself that produced 114,000 word families in one study and 88,500 in another. Those two family figures are not in conflict: Nation's 20,000 describes well-educated speakers, while the 11,100 above is a population average for twenty-year-olds.

Nation is blunt about the older literature too, noting that a majority of it "is methodologically faulty", which is part of why figures above 200,000 are still in circulation. So a headline that says the average person knows twenty thousand words and one that says forty-two thousand can both be right, and neither is much use without its unit attached.

For comparison, the Oxford English Dictionary describes itself as covering "over 500,000 words and phrases across the English-speaking world". No estimate in this literature comes close to that, on any unit.

How many words do you need to read a newspaper?

About 8,000 to 9,000 word families, and the reason is a threshold rather than a preference.

Nation's 2006 paper set out to answer exactly this, using fourteen separate lists of 1,000 word families each, built from the British National Corpus. Its conclusion keeps its condition attached: "If 98% coverage of a text is needed for unassisted comprehension, then a 8,000 to 9,000 word-family vocabulary is needed for comprehension of written text and a vocabulary of 6,000 to 7,000 for spoken text."

The 98 percent is itself an empirical threshold rather than a rule of nature. Nation reports that when a regression model was fitted to comprehension data, "it was calculated that 98% text coverage (1 unknown word in 50) would be needed for most learners to gain adequate comprehension".

One unknown word in fifty sounds generous until you see how slowly coverage climbs. For newspapers, Nation reports that the most frequent 2,000 words of the corpus "account for about 83% of the running words", that 4,000 words plus proper nouns reach around 95 percent, and that "to get to the 98% coverage level a vocabulary of at least 8,000 words plus proper nouns is needed". Words there means word families, the same unit as the headline figure.

Hold that against the section above and the two literatures line up in a way neither paper spells out. The 42,000 lemmas of an average twenty-year-old are derived from 11,100 word families, and the reading threshold is 8,000 to 9,000. An average adult native speaker is already a couple of thousand families clear of it. Nation's target is aimed at learners of English deciding what to study next, not at people wondering whether they can manage a newspaper. Reading the two figures as rivals is the commonest way this research gets garbled.

Novels behave similarly. Nation found The Turn of the Screw reaches 98 percent coverage at 7,000 word families plus proper nouns, and The Great Gatsby at 8,000. Across a combined novel corpus, 2,000 word families gave 87.83 percent coverage and 9,000 plus proper nouns gave 98.24 percent.

And 98 percent is not comfort. Nation quotes Carver's finding that when reading material is difficult, "around 2% or more of the words will be unknown", and adds that "even 98% coverage does not make comprehension easy".

Why is the last two percent of a text the expensive part?

Word frequencies are extraordinarily lopsided, and the shape of that lopsidedness has a name.

Zipf's law says the frequency of a word is roughly inversely proportional to its rank. Steven Piantadosi states it as f(r) proportional to 1/r raised to a power close to one, and describes the consequence: there are "few very high-frequency words that account for most of the tokens in text (e.g., 'a,' 'the,' 'I,' etc.) and many low-frequency words".

That is why the first thousand words you learn buy so much and the ten-thousandth buys so little. You are climbing a curve that flattens almost immediately, and the remaining two percent of any real text is spread across a very long tail of words, many of which turn up only once or twice in the whole piece.

The law is also better described than explained. Piantadosi's review is a critical one, and he concludes that "no prior account straightforwardly explains all the basic facts", warning that a theory's ability to derive the law "provides very weak evidence for that account's cognitive validity". The pattern is solid. The story about why it happens is not settled.

That lopsidedness is also what makes a substitution cipher solvable: crack the handful of words carrying most of the tokens and the rest falls out. If you like the statistics of English more than the vocabulary of it, the daily cryptogram, the cryptogram solver and the wider codes and ciphers set all run on exactly this.

Frequently asked questions

Does vocabulary keep growing after twenty?

Yes, steadily and slowly, about one new lemma every two days on the Brysbaert figures. Worth knowing what "knowing" means in that count, though: the authors state that "the knowledge of the words can be as shallow as knowing that the word exists". Recognising a word as English is a much lower bar than being able to define it or use it. Linguists call the first a receptive vocabulary and the second a productive one, and every large estimate you will see quoted is receptive.

What is the difference between a lemma and a word family?

A lemma is a base form plus its grammatical inflections, so walk, walks, walked and walking are one lemma. A word family goes further and takes in regular derivations, so develop, development, developer and undeveloped collapse into one entry. Word families group more aggressively, which is why counts using them are much lower. There is a third bucket most summaries drop entirely: alongside the 42,000 lemmas, Brysbaert and colleagues credit the same average twenty-year-old with about 4,200 non-transparent multiword expressions, the phrases whose meaning you cannot work out from the parts.

How many words do you need to understand English?

For reading without external help, roughly 8,000 to 9,000 word families, which is what it takes to know about 98 percent of the words on a page. Spoken English is a little kinder at 6,000 to 7,000, because conversation reuses a smaller set. For newspapers specifically, Nation found around 4,000 word families plus proper nouns gets you to about 95 percent, so the final three points of coverage cost roughly as much again as everything before them.

Why do online vocabulary tests give such different results?

Because they choose different units, different sampling frames and different corrections for guessing, the same three levers the quizzes hub describes for short scored tests generally. A test that never shows you an invented word cannot tell whether you are recognising vocabulary or being agreeable, and the standard method exists precisely to subtract that. Sample size matters too: a short test extrapolates a small number of items across tens of thousands of words, so its error bars are wide even when its method is sound.

Does a bigger vocabulary make you a better reader?

It removes an obstacle, which is not the same thing. Coverage research shows comprehension improving sharply as the proportion of unknown words falls, and Nation quotes evidence that even at 98 percent coverage comprehension is not easy. The graded reader is the practical proof: Nation measured a simplified edition of The Picture of Dorian Gray at 10,578 running words drawn from a total vocabulary of just 682 word families. Cut the vocabulary far enough and almost anyone clears the threshold, which tells you the threshold is about the text as much as the reader.

Sources

• Marc Brysbaert, Michael Stevens, Pawel Mandera and Emmanuel Keuleers, "How Many Words Do We Know? Practical Estimates of Vocabulary Size Dependent on Word Definition, the Degree of Language Input and the Participant's Age", Frontiers in Psychology, 2016, open access at PubMed Central: pmc.ncbi.nlm.nih.gov/articles/PMC4965448. Source of the 42,000 and 48,200 lemma figures, the one-lemma-every-two-days rate, the 221,268 participants and 265,346 sessions, the 27,000 to 52,000 spread, the 11,100 word families behind the lemma figure, the 67-word and 33-non-word test format, the Wuggy-generated non-words and the warning given about them, the false-alarm correction and its worked example as printed, the 4,200 non-transparent multiword expressions, the shallow-knowledge caveat, the 61,800-lemma list, the "(a) definition ... (b) age ... (c) amount of language input" dependencies, and the "less than 10 thousand to over 200 thousand" range.
• I. S. P. Nation, "How Large a Vocabulary Is Needed For Reading and Listening?", The Canadian Modern Language Review, volume 63 number 1, September 2006, pages 59 to 82. The journal is paywalled; the full text is hosted at the Compleat Lexical Tutor: lextutor.ca/cover/papers/nation_2006.pdf. Source of the 98 percent coverage threshold and the regression finding behind it, the 8,000 to 9,000 and 6,000 to 7,000 word-family figures, the newspaper coverage percentages, the Turn of the Screw and Great Gatsby results, the graded-reader measurement of The Picture of Dorian Gray at 682 word families across 10,578 running words, the combined-novel figures of 87.83 and 98.24 percent, the 20,000 word families for well-educated native speakers, the 114,000 and 88,500 estimates for English itself, the "methodologically faulty" assessment of the older literature, and the quoted Carver finding.
• Steven T. Piantadosi, "Zipf's word frequency law in natural language: A critical review and future directions", Psychonomic Bulletin and Review, 2014, open access at PubMed Central: pmc.ncbi.nlm.nih.gov/articles/PMC4176592. Source of the statement of the law, the high-frequency versus low-frequency description, and both quoted cautions about the strength of existing explanations.
• Oxford English Dictionary, "About the OED": oed.com/information/about-the-oed. Source of the "over 500,000 words and phrases across the English-speaking world" description.