A Survey on Bias in Deep NLP

Deep neural networks are hegemonic approaches to many machine learning areas, including natural language processing (NLP). Thanks the availability of large corpora collections and capability deep architectures shape internal mechanisms in self-supervised processes (also known as “pre-training”), versatile performing models released continuously for every new network design. These networks, somehow, learn a probability distribution words relations across training collection used, inheriting potential flaws, inconsistencies biases contained such collection. As pre-trained have been found be very useful transfer learning, dealing with bias has become relevant issue this scenario. We introduce formal way explore how it treated several terms detection correction. In addition, available resources identified strategy deal NLP is proposed.