See https://github.com/jammygrams/Pea-QA for details on model training (with narrativeqa dataset)