Neural Generation of Diverse Questions using Answer Focus, Contextual and Linguistic Features

Question Generation is the task of automatically creating questions from\ntextual input. In this work we present a new Attentional Encoder--Decoder\nRecurrent Neural Network model for automatic question generation. Our model\nincorporates linguistic features and an additional sentence embedding to\ncapture meaning at both sentence and word levels. The linguistic features are\ndesigned to capture information related to named entity recognition, word case,\nand entity coreference resolution. In addition our model uses a copying\nmechanism and a special answer signal that enables generation of numerous\ndiverse questions on a given sentence. Our model achieves state of the art\nresults of 19.98 Bleu_4 on a benchmark Question Generation dataset,\noutperforming all previously published results by a significant margin. A human\nevaluation also shows that these added features improve the quality of the\ngenerated questions.\n