Re:SymSpell: 1M times faster spelling correction
by thangalin on Sunday March 06, 2022 @02:58PM
(#30580662)
I integrated JSymSpell[0] into KeenWrite[1], which is based on SymSpell[2]. The algorithm provides numerous suggestions for "plannin'", with "planning" being the first hit[3]. Despite the English lexicon[4] not including all words for all localizations (US/UK) the results have been adequate for my use so far.
What open-source solutions have you been using that aren't working as well as you'd like?
[0]: https://github.com/rxp90/jsymspell
[1]: https://github.com/DaveJarvis/keenwrite/
[2]: https://github.com/wolfgarbe/SymSpell
[3]: https://i.ibb.co/Bs0W6w2/screenshot.png
[4]: https://github.com/DaveJarvis/keenwrite/tree/master/src/main...
Re:SymSpell: 1M times faster spelling correction
by hashingroll on Sunday March 06, 2022 @07:16PM
(#30582644)
The state of the art models these days would not be able to be bundled with browsers. These are typically neural models [0] and require specialized hardware to run, and also have high memory requirements.
The models in browsers are light-weight to be able to run on most devices. Might be similar to those used in smartphone keyboards like GBoard [1].
[0] https://www.microsoft.com/en-us/research/blog/speller100-zer..., https://cloud.google.com/blog/products/g-suite/everyday-ai-b...
[1] https://ai.googleblog.com/2021/10/grammar-correction-as-you-...
Does anyone have any insight into what's holding this back?