How Synonymizer works
What it does
You paste some text, plain or Markdown, and Synonymizer swaps a share of its words for synonyms. Two sliders set the terms. Swap chance is the odds that any given word is swapped. Closeness is the lowest score a synonym needs before it can be used at all. Everything happens in your browser, and the text you type is never sent anywhere.
Is this AI?
No. No AI model runs in your browser or on any server, and there is no chatbot or language model involved. Every swap comes from a fixed dictionary, built from Open English WordNet and the Moby Thesaurus, and a seeded dice roll.
One ingredient is machine-learned. When the dictionary is built, its closeness scores are computed partly with GloVe word vectors, a statistical word-embedding model published in 2014. Those scores are frozen into the dictionary file; nothing is learned or generated while you use the tool.
How a word gets swapped
- Protect. In Markdown mode the text is split into prose and protected parts: code blocks and inline code, link targets, URLs, email addresses, HTML tags and entities, attribute blocks and YAML front matter. Only prose is touched. Contractions, hyphenated words and capitalized words in the middle of a sentence (probably names) are skipped.
-
Find the dictionary form.
bugsis read asbugplus a plural,stoppedasstopin the past tense, andspentasspendfrom a table of irregular verbs. Each reading is checked by inflecting the base form again and confirming it gives back the original word. -
Choose a part of speech. When a word can be a noun or a verb, the word before it decides.
After
the,foror an adjective it is a noun; afterto, a pronoun or a modal likewill, or at the start of a sentence, it is a verb. If nothing decides it, the word is left alone. -
Filter. A synonym survives if its score clears the closeness bar, it can take the same form
as the original, and the next word doesn’t rule it out (
ask forcan’t becomerequest for). - Roll. Each word gets a dice roll from the seed and the word’s position. Because the roll is tied to position, moving the swap chance slider adds or removes swaps without reshuffling the ones already there. Reroll picks a new seed.
-
Pick and repair. One synonym is drawn from the survivors, weighted toward closer ones, in a
way that keeps the same pick when the closeness bar moves (unless the pick itself drops out). The
winner is inflected to match, gets the original’s capitalization, and a preceding
aoranis corrected:an hour,a unit.
In the result, highlighted words were swapped. Select any word the dictionary knows to see its roll and every candidate with its score.
Choosing a word yourself
In that list you can choose any candidate, even one below the closeness bar, or the original word,
which is listed first. Your choice replaces the dice for that one spot and gets the same inflection,
capitalization and a/an repair as any swap. It stays when you move the
sliders or reroll, follows the word when you edit text around it, and is dropped if you change the
word itself. With consistent swaps on, a choice applies to every use of that word. Clear picks,
under Options, removes them all.
The dictionary
Synonym pairs come from Open English WordNet and the Moby Thesaurus. Each pair is scored from 0 to 100 by how close it is to the word’s most common meaning, blending word-vector similarity (GloVe), WordNet’s sense order and how common the synonym is. Each word keeps at most twelve synonyms, and a hand-edited overrides file pins, rescores or bans pairs and adds rules such as “not before for”. The file downloads once and your browser caches it.
Known limits
- Scores describe a word’s usual meaning, so a word used in an unusual sense can get an odd synonym: a garden “bug” may become a “glitch”. Raising the closeness bar is the way to rein this in.
- Context is limited to the words on either side. The grammar repair is heuristic and will occasionally produce something clumsy.
- English only.
Privacy
There is no server-side processing, no account, no cookies and no analytics. Your last text and settings are saved in this browser’s local storage so they are still there next time; the Clear button empties the text.
Possible next steps
- A rendered Markdown preview of the result
- Locking or unlocking individual words by tapping them
- Languages other than English
- A small on-device language model to judge which sense a word is used in
- Shareable links that carry the settings (your text would stay on your device)
Sources and credits
The synonym dictionary is generated from open data and a few hundred hand-written corrections. The dictionary file is offered under CC BY-SA 4.0 as a precaution, because wordfreq’s frequency data (share-alike) guided which words it includes. It contains no frequency data itself.
Synonym data
- Open English WordNet (2025 edition) by the Open English Wordnet team (https://en-word.net/), licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/); derived from Princeton WordNet. WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved.
- WordNet 3.0, Princeton University: sense frequencies, and the basis of Open English WordNet. Used under the WordNet 3.0 license; the full notice is reproduced below.
- Moby Thesaurus II by Grady Ward, public domain.
- GloVe word vectors (Jeffrey Pennington, Richard Socher and Christopher D. Manning, 2014), under the Open Data Commons Public Domain Dedication and License. Only similarity scores derived from the vectors are in the dictionary.
- wordfreq by Robyn Speer: word frequencies under CC BY-SA 4.0, built in part from Google Books Ngrams, OpenSubtitles, Wikipedia, the Leeds Internet Corpus, ParaCrawl and the freely available SUBTLEX word lists (Brysbaert, New, Keuleers, van Heuven, Mandera, Cai and colleagues; SUBTLEX).
- Offensive words are filtered out with the List of Dirty, Naughty, Obscene and Otherwise Bad Words (CC BY 4.0). It is used only as a filter and is not redistributed.
Typefaces
- Newsreader by Production Type, SIL Open Font License 1.1.
- IBM Plex Sans and IBM Plex Mono by IBM, SIL Open Font License 1.1.
The Synonymizer application code is MIT licensed.
WordNet license
WordNet Release 3.0 This software and database is being provided to you, the LICENSEE, by Princeton University under the following license. By obtaining, using and/or copying this software and database, you agree that you have read, understood, and will comply with these terms and conditions.: Permission to use, copy, modify and distribute this software and database and its documentation for any purpose and without fee or royalty is hereby granted, provided that you agree to comply with the following copyright notice and statements, including the disclaimer, and that the same appear on ALL copies of the software, database and documentation, including modifications that you make for internal use or for distribution. WordNet 3.0 Copyright 2006 by Princeton University. All rights reserved. THIS SOFTWARE AND DATABASE IS PROVIDED "AS IS" AND PRINCETON UNIVERSITY MAKES NO REPRESENTATIONS OR WARRANTIES, EXPRESS OR IMPLIED. BY WAY OF EXAMPLE, BUT NOT LIMITATION, PRINCETON UNIVERSITY MAKES NO REPRESENTATIONS OR WARRANTIES OF MERCHANT- ABILITY OR FITNESS FOR ANY PARTICULAR PURPOSE OR THAT THE USE OF THE LICENSED SOFTWARE, DATABASE OR DOCUMENTATION WILL NOT INFRINGE ANY THIRD PARTY PATENTS, COPYRIGHTS, TRADEMARKS OR OTHER RIGHTS. The name of Princeton University or Princeton may not be used in advertising or publicity pertaining to distribution of the software and/or database. Title to copyright in this software, database and any associated documentation shall at all times remain with Princeton University and LICENSEE agrees to preserve same.