Our dataset exceeds the size of existing task-oriented dialog corpora, while highlighting the challenges of creating large-scale virtual wizards. It provides a challenging test bed for a number of tasks, including language comprehension, slot filling, dialog status monitoring, and response…