Towards Uncertainty Unification: A Case Study for Preference Learning