humanflow

Lexical diversity calculator

Measures how varied your vocabulary is using MATTR, a moving-window ratio that does not fall as the text gets longer. The plain type-token ratio is shown alongside it, with the decay demonstrated on your own words.

Runs entirely in your browser — nothing is uploaded, and there is no account or limit.

Last reviewed 15 August 2026 · The HumanFlow team

The number most calculators give you cannot be compared

Type-token ratio is distinct words over total words, and it is the figure almost every lexical diversity tool reports. It has one serious flaw: it falls as text gets longer, in every text, regardless of who wrote it or how well. You keep needing the and of and which, while genuinely new words arrive more and more slowly, so the ratio slides downward by arithmetic alone.

The practical consequence is that a 300-word abstract will nearly always out-score a 6,000-word chapter by the same author, and a comparison between them says nothing at all. Add a paragraph to your own draft and the score drops, which makes it useless even for tracking yourself. The chart above measures your text at increasing lengths so you can watch it happen.

What MATTR does differently

It holds the length constant. Take the first fifty words, compute the ratio, move the window forward by one word, compute it again, and keep going to the end; the answer is the average of all those windows. Every measurement is made on exactly fifty words, so document length drops out of the result and the figure means the same thing in a paragraph and in a thesis.

It is not the only length-corrected measure — MTLD and vocd-D are the two other names you will meet in the literature, and each makes different assumptions. MATTR is here because it is the easiest to state honestly in one sentence, which matters for a number people will quote.

What it does not tell you

Whether the writing is good, and whether any detector will flag it. Repetition is often correct: a methods section that keeps saying participants is being precise, and a thesaurus pass over it would damage the writing while improving the score. Treat the figure as a description, never a target.

The term itself has a fuller treatment in the glossary entry, including where the measure shows up in authorship arguments and why a raw ratio is not evidence unless both texts are the same length.

If the underlying worry is detection rather than style, the honest ground is how detectors actually work and perplexity, which is the quantity they are built on and is not the same as vocabulary variety. For the writing itself, word frequency shows you which words you are leaning on, and the sentence rhythm analyzer covers variation in length rather than in vocabulary.

Common questions

What is lexical diversity?
How much of your vocabulary is distinct rather than repeated. The simplest measure divides the number of different words by the total number of words, which gives you the type-token ratio. The idea is straightforward; measuring it in a way that can be compared between two texts is the hard part.
Why does my type-token ratio fall when I add more writing?
Because it always does, for arithmetic rather than stylistic reasons. Function words like “the” and “of” keep recurring however long you write, while genuinely new words arrive more and more slowly, so the ratio of distinct to total declines with length in every text ever measured. The decay chart on this page shows it happening in your own words.
What is MATTR and why use it instead?
The moving-average type-token ratio. It computes the ordinary ratio over a fixed 50-word window, slides that window one word at a time across the whole text, and averages the results. Because the window is always the same size, the figure does not drift as the document grows, so two texts of different lengths can be compared. It was introduced by Covington and McFall in the Journal of Quantitative Linguistics in 2010.
What is a good MATTR score?
There isn't one, and be sceptical of any source that gives you a threshold. Lexical diversity varies with genre, discipline, audience and purpose — technical writing repeats its key terms on purpose, and forcing synonyms in would make it worse rather than more diverse. The figure is useful for comparing drafts of the same piece or one writer over time, not for hitting a target.
Why 50 words for the window?
It is the conventional choice and the one most published figures use, so results here are comparable with results elsewhere. A window has to be small enough that short texts still yield one and large enough not to be dominated by a single sentence. Below 50 words this tool declines to report MATTR rather than shrink the window and hand you a number that means something different from everyone else's.
Do AI detectors measure lexical diversity?
Not as such, and do not treat this as a proxy for one. Detectors work from the probability a language model assigns to your word choices, which is a different quantity that happens to be correlated with vocabulary variety in some texts. Raising a diversity number tells you nothing reliable about what any detector will output.

Need more than a utility?

HumanFlow rewrites AI drafts to read naturally and scores them for AI, with the signals behind the number — free up to 1,500 words per scan.

Try the humanizer free

Other free tools

Text similarity checkerReference checkerNominalization finderQuote checkerAI disclosure generatorSentence opener checkerInvisible character detectorWord & character counterCase converterText cleanerSentence rhythm analyzerText diff checkerWord frequency counterCharacter counter with platform limitsAI word finderEssay structure checkerReadability checkerPrompt library

← All free tools