Visualizing attention zones in machine reading comprehension models