Evaluation Metrics for Text Data Augmentation in NLP

Recent surveys on data augmentation for natural language processing have reported different techniques and advancements in the field. Several frameworks, tools, and repositories promote the implementation of text data augmentation pipelines. However, a lack of evaluation criteria and standards for method comparison due to different tasks, metrics, datasets, architectures, and experimental settings makes comparisons meaningless. Also, a lack of methods unification exists and text data augmentation research would benefit from unified metrics to compare different augmentation methods. Thus, academics and …

architectures augmentation comparison cs.ai cs.cl data datasets evaluation evaluation metrics experimental frameworks implementation language language processing metrics natural natural language natural language processing nlp pipelines processing promote repositories standards surveys tasks text tools