Counterfactual Language Reasoning for Explainable Recommendation Systems