Diverse data! Diverse schemata?

One of the key value propositions for knowledge graphs and semantic web technologies is fostering interoperability, i.e., integrating data across different themes domains. But why do we aim at interoperability in first place? A common answer to this question that each individual source only contains partial information about some phenomenon interest. Consequently, combining multiple diverse datasets provides a more holistic perspective enables us complex questions, e.g., those span between physical sciences social sciences. Interestingly, while these arguments are well established go by names, variety realm big data, seem less clear whether same apply on level schemata. Put differently, want but also schemata or single one rule them all?