Show HN: Euspell – Spelling reform: an enginnering approach | Hacker News Reader

Show HN: Euspell – Spelling reform: an enginnering approach

euspell.org

1 pointkamranossia4 comments

Hi HN. Euspell is a conservastive respelling of English and all the software needed to implement it. The link above converts the text you paste, in your browser. Other ways to try it out are the browser extensions and EPUB/PDF readers.

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.

Open articleView on HN

ok; I'll be that person:

enginnering <~ engineering conservastive <~ conservative

Normally, i just don't care about a person's spelling. In fact, in these days of LLM tokenized prose, a little random keyboard 'mahsup' can be endearing. However, proposing a spelling help product that was made with an LLM, and then either aren't using it or it doesn't work. Not a strong Show HN outing.

I like the semantic approach though. Using context to identify the words which are likely and checking those for correct spelling is novel, at least to my awareness. Perhaps look for options to 'de-Webster' the spelling as an user choice. The American version of English is not automatically the best in all cases, much like the Oxford comma rule is not always the right way to compose a long list of similar ideas within the same sentence.

Colourful and creative authors prefer to have freedom, to choose how they organise the spelling, prose, structure and composition of their verse.

Thank you naishoya, I agree with you that authors should have freedom to organize the spelling - it is the first step to spelling reform. Euspell is not a spelling help product, it is a reading help product. My target audience are the billion-plus non-English speakers and schoolchildren learning how to read. By making spelling more consistent, Euspell gives learners more confidence. In Italian, Spanish, Korean, Turkish, Russian, and other languages whith highly consistent spelling, there is no such thing as 'reading at 4th grade level'. By the end of first grade, the kids can read any text.

Ok, that makes some sense. It was the reform aspect of 'Spelling reform' that i missed. Having taught ESL to students where grammar and many awkward facets of English, i.e. read being pronounced 'reed' or 'red' dependent on the condition of non-past or past tense, i can see the utility of some disambiguation help. Most of my students came from the Japanese mother tongue, where spelling is both less and more involved, and indeed 'literacy' is a much more complicated concept.

The reading at "4th grade level" is at yet another level of difficulty in French if i understand correctly. That may be where some of English' complexity has an origin.

I routinely explain to students and the curious that English is less a "core language" than it is the gradual accretion of words from many languages; primarily French and Germanic during the centuries of iterative conquest of the British Isles and the intermingling of the conquerors and the conquered in the market place, with the inclusions of Latin, Greek, Norse, Celts/Gaelic and various other lexicons through various means. It makes for a rather simplistic explanation of a complicated issue, but serves to resolve the "why are none of the language rules actually followed, and sometimes there are more numerous exceptions than adherents" family of questions.

In any case, if this is what you have an intense passion to build, keep at it!

Thank you for the kind words. Regardless of Euspell adoption, the 205k lexicon may be useful for other NLP software. I spent more than two years tagging every word with its common parts of speech.

Reply on news.ycombinator.com