A study on the relevance of semantic features extracted using BERT-based language models for enhancing the performance of software defect classifiers.