PaperAgent

Jun 15, 2026 · Artificial Intelligence

ML-Embed’s 3D‑ML Framework Breaks the Three Barriers of Multilingual Embeddings

The paper presents ML-Embed, a 3D‑ML framework that tackles the high computational cost, language‑coverage imbalance, and research opacity of multilingual text‑embedding models by introducing MEL, MLL, and MRL techniques, releasing a 50 M‑sample dataset covering 282 languages, and achieving SOTA on nine MTEB benchmarks while remaining fully open‑source.