Exploring the Influence of Relevant Knowledge for Natural Language Generation Interpretability