Skip to content

v2.1.0: Improved Unicode and mixed-script detection, plus short text accuracy

Latest

Choose a tag to compare

@peterc peterc released this 05 Sep 15:09
· 1 commit to master since this release

v2.1 improves Unicode and mixed-script Japanese detection and adds a built-in preference for widely spoken languages on short texts. The public API and default minimum length remain unchanged.

  • Now prefers English, Chinese (Mandarin), Hindi, Spanish, and French for short texts. Eligible trigram candidates receive a point bonus through 10 letters, fading to zero at 50 letters. Script routing and candidate restrictions still apply.
  • Recognize uppercase Georgian, extended Greek, and supplementary Han letters using Unicode script properties.
  • Combine Han and kana when detecting Japanese.
  • Exclude digits, punctuation, and emoji from script votes and minimum-length counting; treat them as trigram separators while preserving combining marks.
  • Expand regression coverage for short texts and Unicode handling.

Short-text preferences are a heuristic: ambiguous fragments can still be misclassified. Statistical detection still returns undetermined below 10 letters by default.

Full changelog: v2.0.1...v2.1.0