In this work, we try to decipher the internal connection of nlp technology
development in the past decades, searching for essence, which rewards us with a
(potential) new learning paradigm for NLP tasks, dubbed as reStructured
Pre-training (RST). In such a paradigm, the role of data wi