KDD 2025 - Boosting Explainability through Selective Rationalization in Pre-trained Language Models