A BERT Baseline for the Natural Questions
This provides a strong incremental baseline for researchers working on the Natural Questions dataset, improving performance metrics.
The paper tackles the problem of question answering on the Natural Questions dataset by introducing a BERT-based baseline model, which reduces the gap to human upper bound by 30% for long answers and 50% for short answers.
This technical note describes a new baseline for the Natural Questions. Our model is based on BERT and reduces the gap between the model F1 scores reported in the original dataset paper and the human upper bound by 30% and 50% relative for the long and short answer tasks respectively. This baseline has been submitted to the official NQ leaderboard at ai.google.com/research/NaturalQuestions. Code, preprocessed data and pretrained model are available at https://github.com/google-research/language/tree/master/language/question_answering/bert_joint.