Towards Topic-to-Question Generation

Abstract

This paper is concerned with automatic generation of all possible questions from a topic of interest. Specifically, we consider that each topic is associated with a body of texts containing useful information about the topic. Then, questions are generated by exploiting the named entity information and the predicate argument structures of the sentences present in the body of texts. The importance of the generated questions is measured using Latent Dirichlet Allocation by identifying the subtopics (which are closely related to the original topic) in the given body of texts and applying the Extended String Subsequence Kernel to calculate their similarity with the questions. We also propose the use of syntactic tree kernels for the automatic judgment of the syntactic correctness of the questions. The questions are ranked by considering both their importance (in the context of the given body of texts) and syntactic correctness. To the best of our knowledge, no previous study has accomplished this task in our setting. A series of experiments demonstrate that the proposed topic-to-question generation approach can significantly outperform the state-of-the-art results.

Proceedings of the 9th International Natural Language Generation conference

“…In future work, we plan to benchmark our systems with other knowledge-rich QG systems such as Olney et al(2012), Becker et al(2010) and Chali and Hasan.(2015). We want to quantitatively evaluate the advantage of using domain knowledge over relying on content analysis of the input document alone.…”

“…An alternative to the phrase-structure parse is the SRL (semantic role label) parse which identifies for each predicate in a sentence, its associated arguments and modifiers, and specifies their semantic roles. A QG system can then extract arguments and modifiers for question construction (Mannem et al, 2010;Lindberg et al, 2013;Chali and Hasan, 2015). These systems are able to generate a wider variety of questions than the phrase structure approach and are not as closely bound to the sentence source text.…”

“…Internally, the summariser identifies key concepts, links the concepts and selects the important ones through concept graph analysis. Chali and Hasan (2015) employed similar sentence simplification and transformation pipeline as the knowledge-poor system proposed in Heilman (2009). However, the system performed topic modelling to identify subtopics of the document.…”

“…Our approach belongs to knowledge-rich category and is most similar to Becker et al (2010) and Chali and Hasan (2015). However, these two systems do not take domain knowledge into consideration when selecting the target concepts.…”

Infusing NLU into Automatic Question Generation

“…An alternative to the phrase-structure parse is the SRL (semantic role label) parse which identifies for each predicate in a sentence, its associated arguments and modifiers, and specifies their semantic roles. A QG system can then extract arguments and modifiers for question construction (Mannem et al, 2010;Lindberg et al, 2013;Mazidi and Nielsen, 2014;Chali and Hasan, 2015). These systems are able to generate a wider variety of questions than the phrase structure approach and are not as closely bound to the sentence source text.…”