As a NLP practitioner, whose research filed is Information Extraction (IE), I recently found that more and more IE tasks have been SOTAed by Question Answering (QA) models, which raises questions: why do QA models work so well? Do QA model architectures inherently approach IE tasks better than those traditional architectures? To answer these questions, in this blog, I survey four papers, which covers three IE tasks Name Entity Extraction (NER), Relation Extraction (RE) and Event Extraction (EE). I compare those proposed models with other competitive non-QA models and reveal whether QA models are as magical as they have been advertised.

As Sebastian Ruder suggested in his “10 Tips for Research and a PhD”, I wanna start my blogging here to share my experience in Ph.D studies and document interesting NLP knowledge. Appreciate all the constructive feedback, and hope everyone can learn and grow here.