Paragraph-level Neural Question Generation with Maxout Pointer and Gated Self-attention Networks

TLDR

Question generation, crucial for QA and conversational assistants, has seen neural seq2seq models outperform rule-based systems, yet these models struggle with long paragraphs, often requiring full paragraph context for high-quality questions. This study introduces a maxout pointer mechanism with gated self-attention encoder to overcome the challenges of processing long text inputs for question generation. The proposed architecture combines a maxout pointer network with a gated self-attention encoder to effectively handle paragraph-level inputs. The model surpasses prior approaches on sentence-level inputs and achieves a state-of-the-art BLEU_4 score of 16.3 on paragraph inputs, up from 13.9.

Abstract

Question generation, the task of automatically creating questions that can be answered by a certain span of text within a given passage, is important for question-answering and conversational systems in digital assistants such as Alexa, Cortana, Google Assistant and Siri. Recent sequence to sequence neural models have outperformed previous rule-based systems. Existing models mainly focused on using one or two sentences as the input. Long text has posed challenges for sequence to sequence neural models in question generation – worse performances were reported if using the whole paragraph (with multiple sentences) as the input. In reality, however, it often requires the whole paragraph as context in order to generate high quality questions. In this paper, we propose a maxout pointer mechanism with gated self-attention encoder to address the challenges of processing long text inputs for question generation. With sentence-level inputs, our model outperforms previous approaches with either sentence-level or paragraph-level inputs. Furthermore, our model can effectively utilize paragraphs as inputs, pushing the state-of-the-art result from 13.9 to 16.3 (BLEU_4).