backend/app/services/language.py

"""
Lightweight French/English language detection — no dependencies, no LLM call.

Scores a text by counting high-frequency French function words and
French-specific accented characters. Designed for short chat messages where
heavyweight detectors are overkill. Defaults to English when ambiguous.
"""
import re

_FRENCH_WORDS = {
    "le", "la", "les", "un", "une", "des", "du", "de", "et", "est", "ce",
    "cette", "ces", "je", "tu", "il", "elle", "nous", "vous", "ils", "elles",
    "mon", "ma", "mes", "votre", "vos", "notre", "nos", "son", "sa", "ses",
    "pour", "avec", "dans", "sur", "pas", "ne", "que", "qui", "quoi", "où",
    "comment", "pourquoi", "quand", "quel", "quelle", "quels", "quelles",
    "bonjour", "salut", "merci", "oui", "non", "voudrais", "veux", "peux",
    "pouvez", "envoyer", "envoie", "courriel", "objet", "sujet", "besoin",
    "aimerais", "souhaite", "souhaiterais", "démo", "démonstration", "être",
    "avoir", "faire", "aussi", "très", "plus", "moins", "donc", "alors",
    "était", "sont", "été", "fait", "déjà", "encore", "ici", "là", "chez",
    "entreprise", "société", "écrire", "rédiger", "moi", "toi", "lui",
}

_ENGLISH_WORDS = {
    "the", "a", "an", "and", "is", "are", "was", "were", "be", "been", "it",
    "this", "that", "these", "those", "i", "you", "he", "she", "we", "they",
    "my", "your", "our", "his", "her", "their", "for", "with", "in", "on",
    "not", "what", "which", "who", "where", "when", "how", "why", "would",
    "could", "should", "can", "will", "want", "need", "like", "please",
    "hello", "hi", "thanks", "thank", "yes", "no", "send", "email", "mail",
    "subject", "demo", "demonstration", "write", "company", "about", "to",
    "of", "do", "does", "have", "has", "get", "me", "us", "them", "also",
}

_FRENCH_ACCENTS = "àâäçéèêëîïôöùûüÿœ"


def detect_language(text: str) -> str:
    """Returns 'fr' or 'en'. Defaults to 'en' when the signal is weak."""
    lower = text.lower()
    words = re.findall(r"[a-zà-ÿœ']+", lower)

    fr_score = sum(1 for w in words if w in _FRENCH_WORDS)
    en_score = sum(1 for w in words if w in _ENGLISH_WORDS)
    # Accented characters are a strong French signal in this EN/FR context.
    fr_score += sum(2 for ch in lower if ch in _FRENCH_ACCENTS)

    return "fr" if fr_score > en_score else "en"


def language_name(lang: str) -> str:
    return "French" if lang == "fr" else "English"