Can this framework be used to detect stance in unseen and unlabelled custom datasets given a hypothesis and a premise?