Error Analysis Prompting Enables Human-Like Translation Evaluation in Large Language Models F i n d i n g s - A C L