Yes it does! :) Sometimes we translate it in English and back and use the same model, but otherwise yes some are trained in other languages directly and work the same way!
i am looking forward to a follow-up, explaining how to conceptualize this shaky phenomenon you refer to as "real understanding". or why would you insist llms have no potential for such capability?
Simply, the "knowledge" of LLM's is not an "awareness of concepts" but a statistical analysis of copious amounts of text. The statistical analysis results in a prediction model that is primed by your input to generate a string of words. Each word is chosen and strung sequentially depending on the preceding text body. If we really wanted to compare the process to a less technical and more relatable human action, it's much closer to a "stream of consciousness" (improv) than it is to understanding and elaborating.