CS224N: Assignment 4 Reading Comprehension

The Dynamic Coattention Network [1] is an end-to-end neural network model for question answering. The model consists of a coattentive encoder that captures the interactions between the question and the context paragraph, as well as a dynamic pointing decoder that alternates between estimating the start and end of the answer span. Due to the extensive training time needed by the complex decoder in this model, this project explores the possibility of achieving comparable QA performance using the same neural network model, albeit with a simpler decoder structure. We discover that a complex decoder structure is necessary for excellent QA performance. Our model achieves an F1 score of 55.1 and an EM score of 42.4.