Understanding the Semantic SQL Transducer

TL;DR

The paper addresses the need for semantic modelling to bridge business and technical perspectives in data management and governance. It introduces the Semantic SQL Transducer, a SQL-based tool that provides a materialised lossless conceptual view of a relational source and supports bidirectional mappings between the source schema and a conceptual schema, all on standard SQL. Key contributions include a formal design framework with first-order schemas, lossless transformation patterns (vertical and horizontal decomposition), and a canonical abstract relational model in $6^{th}$ normal form$ (CARM), plus support for multiple conceptual models (ERD, ORM, UML Class Diagrams, Property Graphs, Knowledge Graphs). The approach promises improved auditability, governance, data preparation semantics, and interoperability in a knowledge-centric data stack that reduces semantic and data-management debt.

Abstract

Nowadays we observe an evolving landscape of data management and analytics, emphasising the significance of meticulous data management practices, semantic modelling, and bridging business-technical divides, to optimise data utilisation and enhance value from datasets in modern data environments. In this paper we introduce and explain the basic formalisation of the Semantic SQL Transducer, a well-founded but practical tool providing the materialised lossless conceptual view of an arbitrary relational source data, contributing to a knowledge-centric data stack.

Figures (6)

• Figure 1: The Modern Data Stack - source: Fivetran (MDSCON 2020)
• Figure 2: A Semantic Data Stack
• Figure 3: The Semantic SQL Transducer abstract architecture
• Figure 4: The Semantic SQL Transducer of the example
• Figure 5: The conceptual schema of the example (in the ORM notation)