Which way is `right'?: Uncovering limitations of Vision-and-Language Navigation model