Text this: Multimodal Neural Networks in the Problem of Captioning Images in Newspapers