<ul><li>The authors propose a two-step process for template-based question generation from retrieved sentences.</li><li>Using a single Wh-word for all cases reduced performance, while entity-based Wh-word selection performed better.</li><li>Performance improves when the retrieved sentence matches the query and the original context in at least one additional named entity.</li><li>There is an optimal dataset size of around 50,000 examples for improved performance.</li></ul>

5 minutes Review: (Template-Based Question Generation from Retrieved Sentences for Improved…

Discover more