Paradigm Shift in Natural Language Processing

Abstract In the era of deep learning, modeling for most natural language processing (NLP) tasks has converged into several mainstream paradigms. For example, we usually adopt sequence labeling paradigm to solve a bundle such as POS-tagging, named entity recognition (NER), and chunking, classification like sentiment analysis. With rapid progress pre-trained models, recent years have witnessed rising trend shift, which is solving one NLP task in new by reformulating task. The shift achieved great success on many becoming promising way improve model performance. Moreover, some these paradigms shown potential unify large number tasks, making it possible build single handle diverse tasks. this paper, review phenomenon shifts years, highlighting that different