Apple has co-authored a study proposing a new method that could make large language models (LLMs)better for non-English users.