The document describes a rule-based question answering system that takes in a question and story and uses Stanford NLP tools and custom rules to identify the sentence containing the answer. It classifies the question type, calculates word matching scores between the question and sentences, and applies rules to narrow down and score sentences. It outputs a report on test questions with the predicted answer and sentence. The system was improved over time by analyzing reports, adding rules to better identify answers for different question types, and incorporating named entity recognition.