What do Large Language Models Need for Machine Translation Evaluation?