Skip to content
Kotoshu Kotoshu 言修

ko 한국어 n. — full feature

Korean

Hangul eojeol, jamo-aware

Specimen

선생님이 학생에게 사랑을 가르칩니다

1. 생선님 선생님

What arrives on demand

dictionary
Hunspell dictionary with full affix morphology and compounding
frequency
No Kelly list under this code; ranking leans on the dictionary and model
semantic
FastText ONNX embedding model for context-aware reranking — available as fluency (~15 MB, the default), full (120 MB), or mini (~3 MB), chosen at setup
keyboard
Keyboard proximity on the Dubeolsik-2Set layout

Automatic detection is included — a document in this language is found before it is checked.

Set it up

gem install kotoshu
kotoshu setup ko

# optional, for semantic reranking
gem install onnxruntime
kotoshu setup ko --model                  # fluency, the default (~15 MB)
kotoshu setup ko --model --tier full      # maximum accuracy (~120 MB)

Notes

  1. 1.

    The Hangul tokenizer extracts eojeol — syllable blocks with combining jamo attached — so Korean words check whole.

  2. 2.

    Keyboard-proximity suggestions use the Dubeolsik 2-set layout (KS X 5002).

  3. 3.

    The dictionary carries 101,598 entries; the aff alone is 10.6 MB of ICONV rules, so Korean stays gem-only in the browser playground.

  4. 4.

    The fastText ONNX model is available for semantic reranking.