Combining Pre-Trained Language Models and Features for Offensive Language Detection