medical dialogue generation (MDG) has gained increasing attention due to its
substantial practical value. Previous works typically employ a
sequence-to-sequence framework to generate medical responses by modeling
dialogue context as sequential text with annotated medical entities. Whil