Training a model to handle both scripted responses with known transcripts and open-ended responses without predetermined text.