SEA-Embedding: Open and Reproducible Text Embeddings for Southeast Asia
It addresses the lack of reproducible and robust text embeddings for Southeast Asian languages, which are underserved by current models.
SEA-Embedding is a fully open and reproducible text-embedding pipeline for Southeast Asian languages, achieving state-of-the-art results on the SEA-BED benchmark by studying data composition, training objective, and base encoder initialization.
Text embeddings are fundamental to many downstream applications, making robustness important for real-world NLP. However, most recent state-of-the-art embedding models are not reproducible because they rely on closed or undisclosed training data, and they remain insufficiently robust for Southeast Asian languages. We present SEA-Embedding, a fully open and reproducible text-embedding pipeline for Southeast Asian languages trained only on publicly available data, and use it to study three core factors of robust embedding design: data composition, training objective, and base encoder initialization. SEA-Embedding achieves state-of-the-art results on SEA-BED while enabling systematic and reproducible analysis of robust text embeddings for the region.