1 paper
Yiqi Liu, Flavjo Tomori, Nafise Sadat Moosavi +2
Feedback presentation shapes how LLMs use evaluator judgments during translation refinement. We introduce a machine translation protocol that keeps the evaluator score, main issue,…