Euspell
About

Frequently asked

Does Euspell change every word?
No — roughly four fifths of the lexicon is left exactly as it is. Most of those carry encoding 000, an ordinary unchanged word, but there are three smaller groups that are also left alone: 500 for rare or archaic words, 700 for words kept for a French pronunciation, and 800 for a Scottish one. The leading digits say which class a word is in; it is the last digit that says whether its spelling changes at all. Reform is spent only on words whose traditional spelling actively misleads about their pronunciation.
How does it handle words like “read” that are pronounced two ways?
It reads the surrounding words. A five-token context window feeds grammar predicates, a trained model for the common noun-vs-verb case, and per-word rules for genuine sense splits — so read is resolved to the right reformed spelling from its sentence. See Disambiguation.
Can I convert Euspell back to traditional English?
Yes. Conversion round-trips — revertText(convertText(x)) === x — and the office tools expose an explicit Revert to traditional command.
Is it a new alphabet?
No. Euspell uses the ordinary Latin alphabet and keeps each word's shape as close to the original as possible (night → niht), so fluent readers adapt quickly. It is a respelling, not a new script.
Which tools can I use it in?
A Chrome extension (web pages and PDFs), the Eupub EPUB reader, and converters for Microsoft Word, LibreOffice, and Google Docs. All share one engine. See Tools.
Will my spell checker flag Euspell words?
By default yes, since reformed words aren't in the built-in dictionary. Each office tool offers a fix — a custom dictionary, or turning off proofing on the converted text. A derived POS lexicon can even teach open grammar checkers to read it.
How big is the lexicon?

There are three, and only the first one does the reforming.

The spelling lexicon is the big one: about 205000 entries, each carrying a word, its parts of speech, an encoding, and its reformed spelling. It is compiled from source CSVs into a JavaScript map at build time and bundled with every tool, so conversion happens offline on your own machine. Three smaller companion files handle abbreviations, contractions, and phrases.

The other two are derived from it, for software that needs more than spelling. A pronunciation lexicon of about 35000 IPA entries tells a speech synthesizer how a reformed word sounds, so niht is not read aloud as a typo; there is an Arpabet twin for engines that prefer it. A part-of-speech lexicon covering roughly 34000 words in 46684 word-and-tag rows gives grammar checkers the tags they would otherwise be missing for words they have never seen. Both are described under SSML & POS lexicons.

Where is the source code?
On GitHub, all of it, under two licences. The programs are GPL-3.0-or-later. The lexicon data — the CSVs the reform is actually written in — is CC BY-SA 4.0 instead, so the reform itself can be reused by anyone building on it, including in projects that are not GPL. Three repositories: ossiak/euspell for the engine, the lexicon, the browser extension and the word-processor add-ins; ossiak/eupub for the Eupub reader; and ossiak/euspell-game for the conversion game. Each carries its own install and build instructions in docs/, kept current with the code.
Was Euspell generated by AI?
No. The reform architecture and the main lexicon are the author's own work. Anthropic Claude Opus 4.8/5 and Fable 5 assisted with software development, revisions to sections of the white paper, installation instructions, the design of this site, and video production — a disclosure the paper carries in full before its references. Inside the products, machine learning does one narrow job: choosing which of two or more existing spellings applies in context. Nothing is generated while you read; every tool ships with a fixed lexicon. The full statement is on the press page.
Keep reading
Rationale

Why reform, and why this way

Principles

The rules of the reform

Get the extension

Try it in your browser