Lost in Literalism: How Supervised Training Shapes Translationese in LLMs | Read Paper on Bytez