May I Ask a Follow-up Question? Understanding the Benefits of Conversations in Neural Network Explainability