MSA فصحى
EGY مصري
LEV شامي
GLF خليجي
MAG مغاربي
ENG
Heuristic ready
Text
—
How it works
- Heuristic: word/phrase markers per dialect `core.js:8` (EGY إزيك/كده، LEV شو/بدي، GLF وش/أبغى، MAG بزاف/ديال) + EN ratio `core.js:34`.
- Embeddings: exemplar prototypes `core.js:14` (3 per dialect) embedded via `Xenova/multilingual-MiniLM-L12-v2` `app.js:44` → cosine to input → `embeddingScores` `core.js:66`.
- Fusion: weighted `h*0.35+e*0.65` `core.js:79` → top dialect `core.js:84`. Sentence-level breakdown via same heuristic.
- Private: WebAssembly in browser, PWA offline after model cached. No server.
- Limitation: short texts (<8 words) noisy; code-switch flagged as lower confidence.