Recent advances in object-centric process mining necessitated the standardization of object-centric event data (OCED). An IEEE taskforce has developed a "meta-model" for OCED, but there is no existing reference implementation or automated techniques to transform legacy data into OCED. This task requires domain-specific knowledge about the semantics of the legacy data in order to make explicit how events act on various inter-related data objects and their attributes. We propose a semantic header that defines how extracted legacy data maps to OCED concepts and the domain-specific reference ontology using PGschema. We automatically translate the header into database queries to construct an event knowledge graph that is compliant with OCED and the domain ontology using a declarative extract-load-transform approach. The approach has been implemented and demonstrated on 7 real-life datasets, making it one of the first attempts to make OCED operational.