Stop word remover

Remove stop words from text

Stop words are the high-frequency filler of a language — the, and, of, is — that carry grammar but almost no meaning. Search engines, keyword extractors, word clouds and text classifiers usually drop them before they do anything else. Paste your text, pick a language, and this tool returns the same text with the filler taken out. It all runs locally in your browser with WebAssembly; the text never leaves your machine.

Worked example

Paste:

This is a test of the emergency broadcast system.

With the English list selected, the result is:

test emergency broadcast system.

Five of the nine words (this, is, a, of, the) were stop words. Switch Output to Summary statistics and the same text reports the counts instead:

Total words: 9
Removed: 5 (55.56%)
Kept: 4
Distinct stop words removed: 5

Switch it to Removed words + counts and you get an audit list, most frequent first — for the cat and the dog and the bird:

3	the
2	and

Options

• Built-in stop-word list — English, Spanish, French, German, Italian, Portuguese, Dutch, or Russian. Pick None to skip the built-in list entirely and filter with only your own words.
• Extra words to remove — your own terms, on top of the built-in list. Separate them with commas, semicolons, spaces, or newlines. Handy for boilerplate that is filler in your domain: a brand name, lorem, a recurring column header.
• Words to keep — a protection list. Anything here survives even if the built-in list contains it. not, no and without are the classic case: they are stop words by frequency but they invert the meaning of a sentence, so sentiment work usually wants them kept.
• Case sensitive — off by default, so The, the and THE all match a list entry. Turn it on to remove only exact-case matches, which lets you strip a lowercase it while leaving the acronym IT alone.
• Also strip punctuation — off by default, so sentences stay readable. Turn it on to get a bare token stream for an NLP pipeline. Line breaks are kept either way.
• Output — the cleaned text, the list of removed words with counts, or a summary of how much was dropped.

How matching works

Matching is whole-word and tokenizer-based, never a substring replace. the never touches the the inside theatre, and a never eats the a in apple. Contractions stay a single token, so don't matches the list entry don't rather than being split into don and t. Accented and non-Latin words are handled the same way, which is why the Russian list works on Cyrillic text.

After a word is removed, the space it left behind is repaired: the output reads the cat sat. → cat sat., not cat sat . with a stranded space before the full stop. Line breaks and paragraph structure are preserved.

Good for

• SEO — stripping filler before you look at keyword density, or cleaning a list of search queries down to their content words.
• NLP preprocessing — producing the token stream a bag-of-words model, TF-IDF index, or naive-Bayes classifier expects.
• Word clouds and tag lists — so the biggest word isn't the.
• Shortening text — squeezing a long note toward a telegram-style summary.

Limits and edge cases

• Input is capped at 200,000 characters (roughly 30,000 words). Longer text is rejected with a clear error rather than being silently truncated.
• An unknown language code is an error that lists the accepted values; an unknown output view behaves the same way.
• Empty text is not an error — you get an empty result and zero counts.
• Stop-word removal is lossy and not reversible. It destroys grammar on purpose, so don't run it on text you still need to read as prose, on code, or on anything where not and no matter unless you add them to the keep list.
• The lists are the standard closed-class vocabulary of each language (articles, prepositions, pronouns, auxiliaries) — they are deliberately conservative. Add domain filler through Extra words to remove rather than expecting the built-in list to cover it.
• There is no stemming or lemmatisation here: running is not reduced to run. Stop-word removal and stemming are separate steps in a pipeline.

FAQ