arxivcs.CLcs.AIcs.LG2026-07-05
Beyond Multilingual Averages: MTEB-PT, a Benchmark for Portuguese Sentence Encoders
Lucas Hideki Takeuchi Okamura, Alexandre Alcoforado, Anna Helena Reali Costa
Portuguese remains underrepresented in text embedding evaluation, despite being one of the most widely spoken languages in the world. As a result, embedding models are often selected based on English or multilingual metrics, while their effectiveness in Portuguese remains unclear…