Towards Anchoring SelfLearned Representations to Those of Other Agents