Ontology

Ontologies rely on description logics to provide unambiguous, mathematical interpretations of concepts. Unlike a database schema that defines structure, formal semantics define the meaning of relationships, enabling a reasoner to infer new, logically valid facts from explicitly stated ones. This eliminates the ambiguity inherent in natural language labels.

Every resource is assigned to a distinct class within a hierarchical structure. This is not mere tagging; it involves defining necessary and sufficient conditions for class membership. For example, a Cardiologist is not just a label but is formally defined as a Physician who performs a CardiacProcedure. This precision enables complex querying and validation.

Axioms are the rules that govern the domain. They constrain the interpretation of relationships, specifying characteristics like:

• Transitivity: If A is part of B, and B is part of C, then A is part of C.
• Domain/Range: The subject of a hasDiagnosis property must be a Patient.
• Disjointness: A BenignTumor cannot also be a MalignantTumor. These rules are the engine for automated consistency checking.

An ontology represents a consensus view agreed upon by a community of domain experts, not a single application's perspective. This commitment to a shared model is what enables true semantic interoperability between disparate systems. Two independent software agents can exchange data and have it retain its precise, intended meaning without custom point-to-point mapping code.

The ontology's classes and properties form the T-Box (terminological box), while specific real-world data points populate the A-Box (assertional box). An assertion like Patient_123 hasBloodPressure 140_over_90 instantiates the schema. A reasoner can then classify that patient instance as a member of the inferred class HypertensivePatient based on the axioms defined in the T-Box.

A core characteristic is the ability to run a reasoner—a program that derives logical consequences from the axioms. This performs two critical functions:

• Consistency Checking: Ensures no logical contradictions exist (e.g., an organ being classified as both a Tissue and a MedicalDevice).
• Classification: Automatically computes the complete subclass hierarchy, finding that a Finger is a Digit which is an AnatomicalStructure, even if not all links were manually asserted.

A structured data model that instantiates an ontology by representing specific entities as nodes and their relationships as edges. While the ontology defines the classes and properties (the schema), the knowledge graph holds the factual data (the instances). This separation enables deterministic factual grounding for AI systems, ensuring that a language model retrieves a specific 'John Doe' node rather than hallucinating a person.

The computational task of identifying and merging records that refer to the same real-world entity across different data sources. A robust ontology is critical here, as it provides the semantic context to determine that 'J. Smith' in one database and 'Jonathan Smyth' in another are the same :Person based on shared, uniquely identifying properties like :hasORCID or :hasEmail.

The task of converting a natural language utterance into a machine-understandable logical form, such as a SPARQL query. This process relies entirely on the target ontology to map phrases like 'drugs that cause headaches' to the correct classes (:Drug) and properties (:hasSideEffect) in the knowledge graph, bridging the gap between human language and structured data retrieval.

A technique for learning low-dimensional vector representations of entities and relations. While these models often learn from graph topology alone, the logical axioms in an ontology can be injected as a loss function to create more semantically aware embeddings. This ensures that the vector for a :Dog is geometrically constrained to be a subclass of the vector for :Mammal, preserving the formal taxonomy in the latent space.