Machine Comprehension Using Robust Rule-Based Features and Multiple-Sentence Enhancing

In this project, we designed multiple featurelizers to extract information and answer multiple choice reading comprehension questions. Given a triple of passage question and answer, the featurelizer will generate a set of features which are designed using robust NLP tools. We then feed generated features into a neural network classifier which gives a probability score for each answer. The features we used are improved sliding window, key word distance, syntax feature, word embeddings, multiple sentences and coreference resolution.