Yeah makes complete sense, I guess the rationale there is it's hard to find application specific human to human conversation datasets