An Empirical Study of In-context Learning in LLMs for Machine Translation F i n d i n g s - A C L