In modern data engineering, the primary obstacle to progress is rarely a lack of data, but rather the "Semantic Gap"—the disconnect between raw, low-level data and the high-level conceptual understanding required for decision-making. As Massimo Romanelli and Norbert Reithinger describe, this gap exists because different systems use different "languages" (schemas) to describe the same physical reality.
To solve this, researchers have developed a framework for Semantic Interoperability. Instead of building custom adapters for every pair of interacting systems, data is "lifted" into a shared, formal ontology. This creates a universal translation layer where disparate data sources can be integrated, queried, and reasoned over as a single, cohesive knowledge base.
The Mediated Schema as a Universal Translator
The core intuition behind this research is the move from point-to-point integration to a hub-and-spoke model. In traditional systems, if you want a mobile assistant to answer a question using five different databases, you must write code to handle five different formats.
José Luis Ambite and Vinay K. Chaudhri pioneered the use of a central mediated schema. This schema acts as a shared vocabulary. When a query is made, a manager orchestrates the request across diverse systems by mapping the central terms to the specific internal representations of each source. This allows for what Jyotishman Pathak and Vasant G Honavar call "semantic interoperability": the ability for systems to collaborate despite using different naming conventions or database structures.
Lifting Data into Formal Ontologies
For this integration to work, raw data must be "lifted" from its native format (like JSON, XML, or SQL) into a semantic format, typically RDF (Resource Description Framework). Michael I. Luger and Ying Ding 0001 demonstrate this in the music industry, where they "lift" isolated archives into a unified "MusicOntology."
This process involves more than just changing file formats; it involves mapping data to Domain Ontologies. For example:
- In Medicine: Joel T. Dudley and Atul J. Butte used SNOMED-CT as a "common denominator" to link genomic gene expression profiles with clinical electronic health records.
- In Agriculture: Darya Lapo and Roberto Monsorno map heterogeneous sensor streams (weather, soil, moisture) to a formal ontology to give apple growers actionable insights.
- In Engineering: Ian Horrocks 0001 and Henson Graves use ontologies to unify data from mechanical, electronic, and software engineering disciplines during aircraft design.
Anchoring with Upper-Level Ontologies
To ensure that even different domain ontologies can talk to each other, researchers often anchor them to Foundational (Upper-Level) Ontologies like BFO (Basic Formal Ontology) or DOLCE.
As Shusaku Egami explains, these upper-level ontologies provide a standardized foundation for fundamental concepts like space and time. By aligning a specific domain (like air traffic management or informed consent) with these foundational structures, researchers ensure that the data remains interoperable across entirely different scientific fields. Marcelline R. Harris utilized this approach to create the Informed Consent Ontology (ICO), ensuring that patient data could be shared across different hospital systems and research biobanks without losing its legal or clinical meaning.
From Raw Streams to Automated Reasoning
The ultimate goal of this semantic mapping is to move from observation to automated reasoning. When data is structured semantically, machines can "understand" the context of the information.
Michael Bruenig and Jane Hunter 0001 applied this to bushfire safety. By converting raw sensor streams into a standardized graph format, their system doesn't just record temperature; it reasons about the relationship between wind, humidity, and fuel load to estimate fire risk in real-time. Similarly, Anika Schumann and Marco Luca Sbodio developed a system that diagnoses why traffic is congested by linking bus locations with city event schedules and roadwork data through a shared semantic framework.
By prioritizing shared definitions over centralized storage, this research direction enables a "Semantic Data Fabric" (Martin Kaltenböck)—a world where data from any source can be instantly discovered, integrated, and put to work.