[desku]

Here's a paper on how BERT (a large Transformer model trained using self-supervised learning) implicitly learns the traditional NLP pipeline: https://arxiv.org/abs/1905.05950