BERT Explainability

In this post we are going to explore few methods towards the explainability of BERT, and why it may be worth the time. BERT is an open source machine learning framework for natural language processing (NLP). BERT is designed to help computers understand the meaning of ambiguous language in text by using surrounding text to establish context. BERT, which stands for Bidirectional Encoder Representations from Transformers, is based on Transformers, a deep learning model in which every output element is connected to every input element, and the weightings between them are dynamically calculated based upon their connection (in NLP, this process is called attention). Let me explain the problem in a layman term first before delving into details and throwing a lot of technical and Machine learning jargons at you. So to put simply, if we have text columns in our dataset along with numerical columns and if we want to understand how that text column's content is contributing in our predictions (what words, bigrams, trigrams are playing important role) then how can we do it?