Towards Trustworthy Natural Language Explanations For Recommender Systems