Latin-script text (English and similar) is edited as real, selectable text. For scripts the PDF's font can't encode — like Devanagari, Chinese or Arabic — the edit is rendered as an image instead, so it displays correctly but the text itself is no longer selectable in that spot.

It replaces it — the original text is removed from the page's content and the new text is drawn in its place, rather than being pasted over the old text as a separate layer sitting on top.