Mapping Relational Database Constraints to SHACL (Extended Version)

Abstract

Most structured data today is still stored in relational databases, which makes it important to provide a translation between relational and semantic data. A relational to RDF mapping, such as R2RML [14], provides a way to view existing relational data in the RDF data model through declarative mappings. While relational to RDF mapping translates relational instance data to RDF, it does not specify any translation of existing relational constraints such as primary and foreign key constraints. Since the introduction of R2RML, interest in RDF constraint languages has increased and SHACL [17] has been standardised. This raises the question of which SHACL constraints are guaranteed to be valid on a dataset produced by a relational to RDF mapping. For arbitrary SQL constraints and relational to RDF mappings, this is a hard problem, but we introduce a number of restrictions on the mappings that allow us to introduce a constraint rewriting for relational to RDF mappings that faithfully transfers SQL integrity constraints to SHACL constraints. We define and prove two fundamental properties, namely maximal semantics preservation and monotonicity.