Text this: Visual Question Answering Using Semantic Information from Image Descriptions