A Comparative Study of Compound Critique Generation in Conversational Recommender Systems