Semantic Interoperability Models

• Semantic Interoperability Models are formal design patterns that ensure data retains its intended meaning through alignment of ontological and propositional structures.
• They integrate heterogeneous metadata, schemas, and logical frameworks to enable automated reasoning and scalable cross-domain data reuse.
• Various frameworks, including CBIM, EMMO, and DCAT-AP+, demonstrate practical applications in AEC, research data management, biomedical informatics, and digital twin systems.

Semantic interoperability models are formal mechanisms, design patterns, and computational workflows that enable heterogeneous data, metadata, and processes from disparate information systems, tools, or domains to be meaningfully integrated, queried, and reused. Distinct from syntactic interoperability, which aligns formats and protocols, semantic interoperability ensures that exchanged information retains its intended meaning and can be understood, processed, and reasoned about automatically across boundaries of software, organizational context, and domain-specific conceptualization.

1. Conceptual Foundations and Formal Principles

The core of semantic interoperability rests on explicit, formally defined mechanisms for representing the meaning and reference of data. Two central axes are commonly distinguished:

• Terminological interoperability concerns itself with alignment or mapping of terms, vocabularies, and ontological entities such that their meanings (intensional content) and—when required—their referents (extensions) can be established as “the same” or suitably related. Formal distinctions are made between ontological (identity-of-meaning, often via owl:sameAs) and referential (identity-of-reference, e.g., owl:equivalentClass) mappings (Vogt et al., 2024).
• Propositional interoperability addresses the alignment of entire data statements, including their internal schema (slots, fields) and the logical regime governing their use. This involves schema interoperability (crosswalks aligning the positions and constraints of different data shapes) and logical interoperability (shared or suitably aligned inference frameworks).

Semantic interoperability relationships are thus often encapsulated as mappings, alignments, or crosswalks—between terms, schema slots, and TBox/ABox logical structures. The formal foundation ranges from Description Logic (as in OWL), through RDF(S), first-order logics, and category-theoretic schemas (Nagy et al., 23 Jan 2026), to formal workflows for mapping between n-ary statements and minimal units of meaning (Vogt et al., 2023, Vogt et al., 2024).

2. Representative Frameworks and Models

A wide range of semantic interoperability models has been proposed for different domains, frequently combining ontologies, knowledge graphs, machine learning, and rule-based formalizations.

• Graph-Based Knowledge Models: In the built environment, CBIM generates a knowledge graph from object geometry using ML-driven classification, rule-based relationship inference, and semantic enrichment with geometric attributes. The pipeline delivers a directed attributed graph that serves as an interoperability substrate across BIM-tool versions (Wang et al., 2023).
• Ontology-Driven Integration: Both the European Materials and Modelling Ontology (EMMO) and the Digital Engineering Framework for Integration and Interoperability (DEFII) use OWL ontologies as schema-level backbones to align disparate engineering models, enable automated reasoning, and serve as an authoritative source of truth. The design space of EMMO is parameterized along axes (semiotics/physicalism, triad/dyad, modal/non-modal) with explicit recommendations for alignment, extension, and integration (Horsch et al., 2020, Horsch et al., 2020, Dunbar et al., 2022).
• Reference Schema Patterns and Interlingua: The Rosetta Statement metamodel and associated frameworks (Rosetta Stone) model all semantic content as structured n-ary statements, mirroring human-readable natural language. Each statement is typed by a lightweight pattern, with slot constraints and direct mapping to minimal reference vocabularies such as Wikidata. This design drastically reduces the combinatorial cost of schema alignment, supports dynamic labels, and enables lay-domain experts to define and query structured data without ontology expertise (Vogt et al., 2024, Vogt et al., 2023).
• Layered Application Profiles and Contextual Extensions: Interoperability across research domains is advanced through layered profiles (e.g., DCAT-AP, DCAT-AP+), which offer an extensible provenance-aware core (activities, agents, entities, qualitative/quantitative attributes) that can be specialized for disciplines (e.g., ChemDCAT-AP for chemistry/catalysis) while remaining compatible via inheritance, mixin, and slot-mapping mechanisms such as LinkML (Stroemert et al., 2 Feb 2026).
• Category-Theoretic and Mathematical Foundations: Recent advances show that treating schemas and mappings as categorical objects yields O(n) specification complexity for n ontologies, with bidirectional migrations and automatic query rewriting through colimits. Categorical Query Language (CQL) is used to construct and translate between canonical schemas for IFC, BRICK, and RealEstateCore in building interoperability scenarios (Nagy et al., 23 Jan 2026).

3. Methodologies for Alignment, Mapping, and Enrichment

Semantic interoperability models employ both manual/knowledge-engineering and automated methodology components:

• Ontology Alignment and Crosswalks: Formal mapping mechanisms are used to align entities, properties, and schema slots between domain ontologies and reference or top-level ontologies (e.g., EMMO, Wikidata, BFO). Methods include subclass/equivalence axioms, graph-transformation rules, and point-to-point slot crosswalks (Horsch et al., 2020, Vogt et al., 2024).
• Semantic Annotations and Blocks: In PLM, formal semantic annotations link model elements to ontology concepts, specifying their equivalence, subsumption, or other relations, and use explicit structural blocks to aggregate subgraphs into single semantic relationships, supporting inference and inconsistency detection (Liao et al., 2014).
• Hybrid Machine Learning and Rule-Based Pipelines: CBIM’s approach combines random forest classifiers for object type identification with deterministic rule sets for physical relationship extraction (e.g., hosting, adjacency), followed by geometric attribute calculation for semantic enrichment of graph nodes (Wang et al., 2023).
• Layered and Modular Architecture: Models such as DEFII and DCAT-AP+ exploit multi-layered architectures (canonical + application ontologies; provenance core + specialized extensions), modularizing ontologies and mapping interfaces to improve scalability, maintainability, and domain adaptation (Stroemert et al., 2 Feb 2026, Dunbar et al., 2022, Berges et al., 2024).
• Declarative Mapping and Query Translation: Categorical/functorial approaches (CQL), as well as tools such as SPARQL2XQuery+Xs2OWL for bridging XML and RDF datasets, leverage declarative mapping rules and formal logic to enable correct-by-construction bidirectional migrations and semantics-preserving query translation across format and schema boundaries (Bikakis et al., 2016, Nagy et al., 23 Jan 2026).

4. Application Domains and Demonstrated Impact

Semantic interoperability models have been validated in domains including:

• Architecture, Engineering, and Construction (AEC): CBIM and BIM-to-BRICK frameworks facilitate precise mappings between geometry-centric models (Revit, IFC) and operational/semantic ontologies (BRICK), supporting automated digital twin construction and data exchange across BIM, BMS, and IoT sensor networks (Wang et al., 2023, Vittori et al., 2023).
• Research Data and FAIR Digital Objects: Layered DCAT-AP+ profiles, together with schema inheritance via LinkML, enable comprehensive, provenance-aware cataloging of cross-domain research data. ChemDCAT-AP demonstrates significantly improved data integration and reusability between chemistry and catalysis (Stroemert et al., 2 Feb 2026).
• Computational Molecular Engineering: EMMO-aligned marketplace frameworks provide the ontological backbone for integrating simulation software (VISO), workflow capture (OSMO), and variables (VOV), supporting reasoning and discoverability across codes, data types, and workflow steps (Horsch et al., 2020, Horsch et al., 2020, Horsch et al., 2019).
• Biomedical and Health Informatics: Formal mapping modules, such as DB2OntoModule and path-mapping mechanisms, facilitate OWL2-based alignment of proprietary schemas to canonical EHR ontologies, enabling message-level interoperability and automated assertion transfer across applications and standards (HL7, openEHR) (Berges et al., 2024).
• Digital Engineering and Cybersecurity: DEFII’s MISD-driven interfaces expose ontology-aligned ASTs, supporting tool-agnostic data ingestion, automated reasoning, and structured query access in cybersecurity and cyber-physical engineering contexts (Dunbar et al., 2022).
• Industrial Digital Twins: Large-language-model pipelines that structure raw datasheet text into semantic nodes (name, definition, affordance, value, type, unit, source) demonstrate the use of generative AI for semantic unit extraction and standardized digital twin model generation in Industry 4.0 AAS contexts (Xia et al., 2024).

5. Evaluation, Scalability, and Limitations

Semantic interoperability models report various metrics and practical outcomes:

• Quantitative metrics: CBIM achieved 100% classification accuracy, F1 score ≈ 0.99 for relationship extraction, and full model consistency under isomorphism checks for BIM downgrading scenarios (Wang et al., 2023). Categorical approaches reduce mapping complexity from O(n²) to O(n) for n schemas (Nagy et al., 23 Jan 2026).
• Qualitative assessment: Rosetta Statement patterns substantially lower the technical barrier for domain experts, enable n-ary statement modeling with versioning and cognitive interoperability benefits, and are incrementally extensible to full logical reasoning patterns by ontology engineers (Vogt et al., 2024).
• Performance constraints: Some models note that tool-specific (e.g., Revit- or IFC-dependent) implementations limit generalizability, and that ontology/annotation maintenance requires governance as schemas and domain concepts evolve (Wang et al., 2023, Liao et al., 2014). Categorical and modular approaches require formal mapping validation and management of composition-induced path dependencies (Nagy et al., 23 Jan 2026).

6. Outstanding Challenges and Directions

Key limitations and research directions highlighted in current models are:

• Absence of universal ontologies/schemata: The FAIR 2.0 framework asserts there is no global “best” ontology. Exposing and aligning the relevant terminological and schema mappings is required, rather than enforcing a monolithic solution (Vogt et al., 2024).
• Tool and Semantic Layer Integration: Adaptation to new tools and workflows, as in DEFII and BIM-to-BRICK, requires modular mapping modules and interface specifications, both to keep up with evolving application APIs and to support new domain ontologies (Dunbar et al., 2022, Vittori et al., 2023).
• Automated and explainable mapping: LLM-powered and retrieval-augmented agent architectures suggest potential for semi-automated semantic node extraction, but accuracy and explainability depend strongly on the underlying model and domain knowledge, with RAG methods particularly beneficial for less capable base models (Xia et al., 2024).
• Governance and registry infrastructure: FAIR 2.0 and Rosetta frameworks identify the need for scalable, open, and well-governed registries of mappings, schema definitions, and operation services. Distinguishing statement types and versions (assertional, contingent, prototypical) remains an open problem (Vogt et al., 2024, Vogt et al., 2024).
• Extensibility and Reasoning: The extension of lightweight, cognitively familiar semantic statement models to rigorous logical reasoners (e.g., OWL2, SHACL, SWRL) requires the development of pattern crosswalks and consistent versioning protocols, so that both domain-expert and ontology-engineer needs are met (Vogt et al., 2024).

7. Comparative Table of Exemplar Semantic Interoperability Models

Model/Framework	Formal Backbone	Key Methods	Primary Domain(s)
CBIM	Attributed Graph	ML classification, rule extract	BIM, AEC
DCAT-AP+ / ChemDCAT-AP	RDF/OWL, LinkML	Layered profiles, schema extension	Research data, chemistry
EMMO & Marketplace	OWL2, Mereosemiotics	Alignment, design space tuning	Materials modelling, computational science
Rosetta Statement	Patterned n-ary schema	Minimal unit, interlingua, version	Knowledge graphs, FAIR data, any domain
Categorical (CQL)	Category theory	Functors, colimit integration	Building data, cross-ontology integration
DEFII	OWL2, GraphDB, MISD	Tool-agnostic interfaces, reasoning	Digital engineering, cybersecurity

The diversity in formalizations and workflow components across models underscores the principle that semantic interoperability is not a one-size-fits-all proposition; architectural, logical, and governance dimensions must be co-designed to suit both technical requirements and the evolving needs of domain experts and automated reasoning clients.