How to Use This Tool
Everyone has a word they lean on without hearing themselves do it. This finds them, shows you every place they appear, and — the part that matters — gives you a variety score that still means something when you compare two pieces of different lengths.
The measurement problem nobody mentions
The standard measure of lexical variety is the type-token ratio: distinct words divided by total words. It is easy to compute and, used naively, close to useless.
The reason is arithmetic. As a text grows, the token count keeps rising but the supply of genuinely new words runs out — you have already used the, and, and most of your topic vocabulary. So TTR falls as length increases even when the writing has not changed at all. Compare a 250-word essay with a 400-word one and the shorter piece wins automatically.
What MATTR does instead
MATTR — moving-average type-token ratio — slides a fixed window across the text, measures TTR inside each window, and averages the results. Because every window is the same size, the length bias disappears: a 250-word text and a 4,000-word one are both scored on 100-word windows.
That is the number to compare across drafts, and it is the one shown first here. Raw TTR is shown too, greyed out, so you can see how far apart they are — but do not use it to compare two pieces.
Two toggles worth understanding
Ignore function words removes the, of, is, and and the rest of the closed grammatical class. Leave it on. Function words are not a style choice — you cannot write English without repeating them, and counting them buries the words you actually control.
Group word forms treats important and importantly, and technology and technologies, as one item each. This is suffix stripping, not a real lemmatiser, and it misses two specific classes:
- Irregular forms. go / went, be / was, child / children stay separate, because nothing about their spelling connects them.
- Noun/adjective pairs. important and importance stay separate, because the shared part is import and stripping that far also turns parent into par.
- Short verbs. use / used are not grouped. The rule refuses to strip a suffix that would leave a stem of two or three letters, because loosening it turns need into ne — and a wrong grouping is worse than a missed one, since you would never think to check it.
So the grouped counts are a floor, not an exact figure. Read the list rather than trusting the number.
Reading the position strip
A word used five times spread evenly across 600 words is usually fine — it is your topic. The same word five times inside one paragraph is what a reader notices. The strip shows where each repeat falls, so clusters are visible at a glance, and words that cluster get flagged separately from words that are merely frequent.
What to do about a flagged word
Not necessarily replace it. Reaching for a thesaurus produces the other failure mode — elegant variation, where a paper calls the same thing a study, an investigation, an inquiry and an analysis and the reader starts wondering whether these are four different things. In technical and academic writing, repeating the precise term is correct.
Usually the better fix is structural: two sentences that both start with the same subject can often become one, and a paragraph that repeats a word four times is often making the same point four times.
Nothing leaves your browser
The analysis runs in JavaScript on your device and no request is made. That matters when the text is an unpublished paper or a graded assignment.
