What you described is known as "sequence-to-sequence learning", or seq2seq. Google it or ask ChatGPT. It's a very common NLP technique that can do almost everything you mentioned.

Of course the utility will depend on the training data you collect. For example to go from corporate speak -> english you would need thousands of pairs of "translated" sentences. Or you could use an LLM to translate for you, paying as you go