Hi. I recently had Chatgpt translate some text from Korean to Hungarian but I noticed that it uses the English language Revised Romanization rules to transcribe names and places. Most languages actually have their own rules on how to transcribe non-Latin languages. For example, the Hungarian Academy of Sciences developed such rules for non-Latin languages, described in a book called “Keleti nevek magyar helyesírása”. All the rules of this book have been transferred to the Hungarian Wikipedia, which is easy for gpt to read and learn from, and gpt-4 demonstrated in our latest session that it can apply the rules when it translates for me.
My suggestion is that where available, such official transcription rules should be implemented, at least for the major non-latin alphabets like Cyrillic, Greek, Korean and languages like Japanese, Thai and Chinese.
This could make gpt more useful when used for translation or text aggregation/summarizing tasks.