Ontological Semantics

Architecture of Ontological Semantics#

The key components required to build a knowledge framework leveraging Ontological Semantics — the theoretical grounding that has enabled the current field of Large Language Models — which help us create knowledge out of existing sources in a systematic manner.

@startuml
!theme plain
skinparam defaultFontSize 13
skinparam component {
  BackgroundColor #EBF5FB
  BorderColor #2980B9
  FontColor #1A252F
  RoundCorner 10
}
skinparam database {
  BackgroundColor #FEF9E7
  BorderColor #F39C12
}
skinparam rectangle {
  BackgroundColor #EAFAF1
  BorderColor #27AE60
  RoundCorner 8
}
skinparam arrow {
  Color #2C3E50
  FontColor #555
  FontSize 11
}

title Architecture of Ontological Semantics

package "Static Knowledge Sources" {
  database "Ontology\n(World Model)" as ONT
  database "Fact Repository\n(Domain Facts)" as FACT
  database "Lexicon\n(Word Forms & Meanings)" as LEX
  database "Onomasticon\n(Named Entities)" as ONOM
}

component "Knowledge Representation\nLanguage" as KRL

rectangle "Semantic Analyzer\n(Large Language Model)" as LLM

rectangle "Use Cases" as UC {
  component "Text Summarization" as TS
  component "Question & Answering" as QA
  component "AI Agents" as AG
}

ONT --> KRL : "defines concepts"
FACT --> KRL : "instantiates"
LEX --> KRL : "maps words → meaning"
ONOM --> KRL : "resolves named entities"

KRL --> LLM : "structured meaning\nrepresentation"

LLM --> TS
LLM --> QA
LLM --> AG
@enduml

The Four Static Knowledge Sources#

Ontology — the World Model#

The ontology defines the concepts that exist in a domain and the relationships between them. It is the schema of the knowledge graph: what nodes (concepts) and edge types (relationships) are valid. Without an ontology, a knowledge base is just a bag of facts with no shared meaning.

Fact Repository#

Stores instantiated knowledge — specific, asserted facts about the world derived from texts, experiments, or expert input. Facts are the populated rows of the ontology’s schema.

Lexicon#

Maps surface word forms to their semantic meanings. A lexicon allows the system to recognize that “kinase inhibitor,” “enzyme blocker,” and “phosphorylation suppressor” may all refer to the same ontological concept, enabling robust natural language understanding across varied terminology.

Onomasticon#

A specialized lexicon for named entities: people, organizations, places, products, and standards (e.g., “ICH Q10,” “S88,” “FDA”). The onomasticon lets the system resolve ambiguous proper nouns to canonical knowledge graph nodes.

Knowledge Representation Language#

The Knowledge Representation Language (KRL) is the formal grammar that ties the four sources together. It allows a system to derive meaning from raw text by:

1. Parsing surface language through the Lexicon
2. Resolving named entities through the Onomasticon
3. Grounding concepts against the Ontology
4. Asserting new facts into the Fact Repository

This pipeline is what transforms unstructured text into structured, queryable knowledge graph entries.

Semantic Analyzer — Large Language Models#

The Semantic Analyzer uses the KRL-structured knowledge to drive user-facing interactions. Modern LLMs play this role, but a key distinction applies:

@startuml
!theme plain
skinparam defaultFontSize 13
skinparam rectangle {
  RoundCorner 10
  FontColor white
}

rectangle "**Black Box LLM**\n(no visible ontology)\n\nGenerates answers\nCannot explain reasoning\nKnowledge is opaque" as BB #C0392B
rectangle "**Glass Box (Ontological Semantics)**\n(explicit ontology + fact repository)\n\nGenerates answers\nCan cite sources\nKnowledge is auditable" as GB #27AE60

BB -right[hidden]-> GB
note bottom of BB : Hallucination risk\nLow traceability
note bottom of GB : Ontology-grounded\nHigh explainability
@enduml

The Glass Box advantage is critical in regulated domains like pharma, manufacturing, and clinical research — where every knowledge claim must be traceable to its source.

Supported Use Cases#

Applied Example: This Site’s Taxonomy as an Ontology#

The STEAM knowledge graph on this site is itself an instance of ontological semantics in practice:

• Ontology nodes: Science, Technology, Engineering, Arts, Mathematics
• Ontology edges: enables, implements, models, formalized by, grounded in
• Lexicon: tags and categories on each page
• Onomasticon: persona names, standard names (ICH, S88, S95)
• Fact Repository: each page’s related_concepts frontmatter

See the full graph: STEAM Knowledge Graph →