Inspiration
In recent years, pre-trained models have substantially advanced the state of the art in numerous NLP tasks. Despite this progress, doubts are emerging regarding the true linguistic capacities of these neural networks. This projects aims to test these doubts on BERT, a state of the art pre-trained model.
To do so, I fine-tune BERT on MNLI, a benchmark data set for natural language inference, and test the model on HANS, a heuristic analysis data set. This inspiration for this project comes from the paper introducing HANS: https://arxiv.org/pdf/1902.01007.pdf.
Log in or sign up for Devpost to join the conversation.