Learn to Ask What You Don’t Know

Asking questions relates to the cognitive ability of language comprehension and context understanding. For that reason, question generation is a challenging topic in Natural Language Understanding. In this work, we propose a task called “question generation with masked target answer,” which emphasizes asking questions from text passages without providing a target answer. Compared to other related question generation tasks, our task demands rigorous language comprehension and closely resembles the question asking ability of humans. We then propose various sequence to sequence-based models leveraging additional information about the text, such as its part of speech and named entity recognition(NER) tags. Results show that the proposed models perform on par with other related question generation tasks, despite lacking the key answer phrase.