- Added more languages: am, ast, awa, bho, fo, ht, lb, sa, tg, ti, tl, zsm, zh-hans, zh-hant
- Restrict numpy version to <2
- Added support for Serbian cyrillic and latin dictionaries.
- Changes in Slovak and Slovene similar languages.
- Added more languages: af, ar, az, be, bn, cy, et, fa, fi, ga, gu, he, hi, hu, id, kk, kn, ky, lt, lv, mn, ml, mr, ms, ne, pl, pt, ru, so, sv, ta, te, th, tr, tt, uk, ur, uz
- Fixed bug with character encoding that resulted in some sentences not being evaluated for certain languages.
- Fix issue that was preventing non-latin words from being evaluated.
- Improved removal of punctuation of evaluated tokens
- Conservative mode is now less conservative:
- Raised error threshold
- Tag targetted language in case of tie, if error rate is 0
- Now using CyHunspell.
- Added automatic tests.
- Icelandic mistakeable languages.
- Default dictionaries are now installed via pip as a dependency.
- Download of dictionaries in
fastspell-downloadis deprecated.
- Trigger fasttext download in fastspell-download command.
- Use specific github tag/release to store dictionaries.
- Automatic download of Hunspell dictionaries.
- Changeable configuration.
- Migrate to pyproject and src/ code structure
- Support for mixed scripts.
- Lowercase text for FastText prediction.
- Serbo-Croatian script detection.
- Serbo-Croatian mode.