aidetect/langid.py

"""langid.py — lightweight language detection (ko vs en).

We only need to route to the right LM, so a fast Hangul-ratio heuristic is
enough and needs no dependency. Falls back to `langdetect` if installed and
the heuristic is ambiguous.
"""
from __future__ import annotations
import re

_HANGUL = re.compile(r"[\uac00-\ud7a3]")
_LATIN = re.compile(r"[A-Za-z]")


def detect_lang(text: str) -> str:
    """Return 'ko' or 'en'. Korean if Hangul chars dominate alphabetic chars."""
    n_ko = len(_HANGUL.findall(text))
    n_en = len(_LATIN.findall(text))
    if n_ko == 0 and n_en == 0:
        return "en"
    if n_ko >= n_en:
        return "ko"
    if n_en > 0 and n_ko / (n_ko + n_en) > 0.15:
        # mixed but meaningful Korean present -> treat as Korean
        return "ko"
    return "en"