As in the original paper, we added the possibility to perform question answering using only Quasimodo, i.e. we do not have an underlying language model. We used the same algorithm: Given a question, an answer and a knowledge base, we generate a set of features based on the connections between the words in the knowledge graph. Then, using the same training data as in Quasimodo, we train a linear classifier to predict a score for each answer. We reused the code provided with Quasimodo and added an interface to ask a question and give possible answers. Then, the system provides a score for each answer and displays them.