Previous spelling reforms have failed for two reasons: they asked people to apply rules by hand or remember a large set of new spellings, and they couldn't handle homographs: "read" is reed or red, "records" is a noun or a verb, "wound" is an injury or a past tense - writing in pure phonetics will actually create hundreds of homographs. Euspell approaches spelling reform differently, with the goal of aiding reading and comprehension.
- A lexicon with 205505 entries, each with CLAWS7 part-of-speech tags. Of these, 164067 (80%) are left unchanged. No diacritics are used. Proper nouns are untouched. American spelling is the base, in the line of Webster's reforms. - 5920 entries are ambiguous and need context. The noun/verb -s split (records/recordz) is a linear SVM: 93.2% on held-out text. The hand-written rule it replaced scored 75.9%. The full evaluation is at https://github.com/ossiak/euspell/blob/main/docs/paper-evalu... - Semantic splits (e.g. bow, tear, lead) are hand-written rules and less accurate: "read" gets about four in five right on its test set. - Runs locally in a browser extension (Chrome/Firefox/Safari, including PDFs), MS Word, LibreOffice, Google Docs, and a standalone EPUB/PDF reader for Windows, MacOS, Android and iOS. - Code is GPL-3.0, lexicon is CC BY-SA 4.0: https://github.com/ossiak/euspell. White paper: https://euspell.org/paper
The changes can be toggled on and off: Euspell changes what you read, not the source text. AI tools (Claude) helped with the software, the site and the videos; the reform and the lexicon are mine. There is also a 14-part video series: https://www.youtube.com/playlist?list=PLfCd5hTV7IWg
Most useful feedback: sentences where it picks the wrong spelling.