On the Explainability of Natural Language Processing Deep Models

What is it about?

In this survey, we focus on textual classifiers, language models, transformers and embeddings. We answer the question on whether the attention mechanisms are explainable or not. We also show how such models can be explained post-training and how their architectures can be updated to allow for transparency. We also recognize three stages where explainability comes in handy: input-level, language representations and decisions or outcomes. We show a case-study on neural machine translations.

Featured Image

Photo by Brett Jordan on Unsplash

Photo by Brett Jordan on Unsplash