Learning by Asking for Embodied Visual Navigation and Task Completion