Question1 is the input element has the aria-describedby attribute related to the dt element has id. And Question2 is the input element has the aria-labelledby attribute related to the dt element has id.
The combination of version of the OS and the screen reader and the browser is as follows.
The screen reader read by Japanese language. I translate it to English.
Question1 is no problem, so NVDA read to label texts after texts of the dt element. But, Question1 is problem, so NVDA didn't read label text In the case I use aria-labelledby attribute as described above.