Ontologies: A Ten-Lecture Course

Ten lectures, twenty to thirty minutes each, about four hours ten total at a narration pace of one hundred and forty-five words per minute. Written to be listened to rather than read. Transcripts live in transcripts/ as plain text files, numbered in listening order.

Length follows the sources. Each lecture runs exactly as far as the notes listed for it carry it, and no further, so the files differ in length by a third or more. Shipped word counts:

Every claim traces to a note in this repository. The source notes are listed per lecture below, and the transcripts name their sources aloud where it matters.

Design constraints

• Solo lecturer. One voice, spoken prose, no bullets, no code, no URLs, no section headings inside the transcript. Anything a narrator cannot say out loud has been cut or rewritten.
• One sitting per file. Twenty to thirty minutes at about 145 words per minute. There is no word target. Quality is the constraint and length is the by-product: a lecture stops when its notes are spent.
• Self-contained openings. Each lecture starts by saying where it is in the arc, because listeners on a walk lose the thread between sessions.
• One recap beat. Each lecture ends by stating the single claim it wants retained, then stops. No summary-of-the-summary.
• Acronyms expanded on first use in every lecture, since listeners may start anywhere.

The arc

Lectures one to three establish what the word means and where it came from. Four to seven are the engineering tradition, ending with how you would actually build one. Eight is the criticism. Nine and ten are Palantir and the present. The criticism deliberately lands after the listener knows how the machinery works, so it argues with something concrete.

Lecture plans

1. Three Things People Mean by “Ontology”

Claim to retain: The word names a philosophical discipline, a specific theory of what exists, and a machine-readable file. Most confusion in this field is one of those three senses being swapped for another.

Beats: the airport-conversation problem; sense one, the branch of philosophy; sense two, a particular philosopher’s ontology; sense three, the artifact; Gruber’s definition read slowly; why “explicit” and “shared” are the load-bearing words; what an ontology is not (a database schema, a taxonomy, a knowledge graph); the promise of the series.

Sources: notes/gruber-ontology-definition.md, notes/ontology-word-history.md, notes/skos-simple-knowledge-organization-system.md, notes/scigraph.md, SYNTHESIS.md.

2. From Aristotle to Wolff: Where the Word Came From

Claim to retain: Ontology began as the search for the single true account of being, and the engineering version quietly dropped that ambition while keeping the vocabulary.

Beats: Aristotle’s “being qua being” and why first philosophy is not physics; the two-thousand-year gap before the word; Lorhard 1606; Goclenius 1613 and the Clauberg misattribution; Wolff 1730 and the demonstrative method; general versus special metaphysics as the ancestor of upper versus domain ontologies; Kant’s demolition; Husserl’s formal ontology; what survived the transfer to computing and what did not.

Sources: notes/ontology-word-history.md, notes/bfo-basic-formal-ontology.md, notes/common-core-ontologies-cco.md.

3. Quine, and Ontology as a Property of Theories

Claim to retain: To be is to be the value of a variable — Quine’s own wording, not the familiar “bound variable” version, which is Boolos’s later title. Ask not what exists, but what your model must quantify over to be true.

Beats: the problem of non-being; the move from world to theory; the criterion stated plainly; worked example on an employee schema; why the criterion is silent about truth; why that silence makes ontology alignment permanent rather than embarrassing; the structural resemblance between Quine’s criterion and Guarino’s construction, which cites neither the other.

Sources: notes/quine-ontological-commitment.md, notes/gruber-ontology-definition.md, notes/ontology-matching.md, notes/alignment-api.md.

4. The Cathedral: Cyc, and the Birth of the Engineering Sense

Claim to retain: The computer-science sense of “ontology” was invented to solve a sharing problem between knowledge-based systems, and its founding project bet everything on a critical mass it never reached.

Beats: expert systems and the knowledge-acquisition bottleneck; Lenat and Shepherd starting Cyc in 1984 knowing the odds; what codifying common sense requires — causality, time, space, substances, intention, belief; the peanut butter and table example; a million hand-crafted axioms; the DARPA Knowledge Sharing Effort and why sharing, not reasoning, produced the definition; Gruber 1993; Borst’s addition of “shared”; the lesson about deferred payoff.

Sources: notes/cyc-lenat-1995.md, notes/gruber-ontology-definition.md, notes/cyc.md, notes/description-logics-dls.md.

5. The Semantic Web: A Vision and the Stack It Built

Claim to retain: Every awkward feature of the standards is an answer to a requirement in the 2001 article, and most of those requirements assumed an open web of strangers.

Beats: the Pete and Lucy scenario read out; what it demands technically; RDF as the merge-without-coordination data model; the URI as global identity; why triples and not tables; RDFS as the small vocabulary; OWL as the expressive one; SPARQL as the query layer; SHACL and ShEx as the validation layer that arrived late and mattered more than expected; serializations as spelling rather than substance; the four phases and where the energy went.

Sources: notes/semantic-web-2001-vision.md, notes/rdf-resource-description-framework-1-1.md, notes/rdf-schema-rdfs-1-1.md, notes/owl-2-web-ontology-language.md, notes/sparql-1-1.md, notes/shacl-shapes-constraint-language.md, notes/turtle.md, notes/json-ld.md.

6. How the Machinery Actually Works

Claim to retain: Open-world semantics is the single idea that most often breaks a database engineer’s intuition, and every strange behaviour of a reasoner follows from it.

Beats: description logic as the mental model; concepts as sets, roles as pairs; TBox and ABox; what a reasoner computes — subsumption, membership, consistency; the classification example; the open-world assumption spelled out with a worked case; no unique name assumption; why absence proves nothing; the expressivity and complexity trade; OWL profiles EL, QL, RL and what each buys; why validation with SHACL is a different job from inference with a reasoner.

Sources: notes/description-logics-dls.md, notes/owl-2-web-ontology-language.md, notes/elk.md, notes/hermit.md, notes/openllet.md, notes/shacl-shapes-constraint-language.md, notes/dl-query-class-expression.md.

7. Building One for Real

Claim to retain: An ontology project is a governance project with a file attached. The modelling is the easy half.

Beats: competency questions before classes; layered reuse and why BFO excludes scientific terms; mid-level ontologies; picking a foundation and what DOLCE, gist, and gUFO each commit you to; design patterns and why the pattern catalogue is not a certification; identifiers and the label-carries-meaning rule; Protégé and the editor spectrum; automation with ROBOT and documentation with WIDOCO; alignment as its own discipline; version and profile compatibility as an operational fact; the scope-creep failure mode.

Sources: notes/bfo-basic-formal-ontology.md, notes/common-core-ontologies-cco.md, notes/gist.md, notes/gufo.md, notes/dolce-descriptive-ontology-for-linguistic-and-cognitive-engineering.md, notes/ontologydesignpatterns-org.md, notes/protege.md, notes/robot.md, notes/widoco.md, notes/obo-foundry.md, notes/elot.md, notes/agreementmakerlight-aml.md, notes/logmap.md.

8. The Criticisms That Landed

Claim to retain: Three critiques, three different layers. Incentives, applicability, and consequences. The third is the one the field still has not answered.

Beats: Doctorow’s seven obstacles with the eBay and Nielsen examples; schemas are not neutral, washing machines both ways; what Doctorow concedes about implicit metadata; Shirky’s periodic table and noble gases; Dewey’s 200s and the Library of Congress on Asia and Africa; there is no shelf; the parable of the Yahoo ontologist; Shirky’s real preconditions restated as a checklist; Bowker and Star on the International Classification of Diseases, apartheid reclassification, and the Nursing Interventions Classification; torque; black-boxing; what each critique does and does not prove.

Sources: notes/doctorow-metacrap.md, notes/shirky-ontology-is-overrated.md, notes/bowker-star-sorting-things-out.md, notes/semantic-web-retrospective.md.

9. Palantir: The Ontology as an Operational Layer

Claim to retain: Palantir’s Ontology adds verbs and permissions to the nouns, which is exactly what the W3C stack cannot express, and it drops the logic, which is exactly what the W3C stack is for.

Beats: what Palantir actually says in its own architecture documentation; the fourfold integration of data, logic, action, and security; object types, objects, object sets, link types and their two sides; the dataset analogy Palantir itself offers; action types as transactions with parameters, validation, and side effects; functions and interfaces; writeback, so decisions become data; digital twin framing; the explicit claim that it is not a semantic layer; the head-to-head with OWL on open versus closed world, reasoning, and federation; why the agent story makes actions the important half.

Sources: notes/palantir-ontology.md, notes/owl-2-web-ontology-language.md, notes/shacl-shapes-constraint-language.md, notes/gruber-ontology-definition.md.

10. The Politics of an Installed Ontology, and What Comes Next

Claim to retain: Once an ontology drives actions inside an institution, its categories stop describing the world and start producing it. Everything in lecture eight becomes a legal question.

Beats: the Danish POL-INTEL study and its two senses of ontology; the getaway-vehicle example; platformisation and the redistribution of skill; the German constitutional court ruling of February 2023 and the bystander problem; why this is Bowker and Star with a contract attached; what a responsible review of an operational ontology asks; then the present — can a language model replace an ontology, build one, use one; the three answers; the guarded pattern from ELOT; closing verdict on when to build one and when not to.

Sources: notes/palantir-ontology-critique.md, notes/bowker-star-sorting-things-out.md, notes/llm-ontology-debate.md, notes/ontolearner.md, notes/ontoaligner.md, notes/elot.md.

Listening in ElevenReader

Upload the ten files in transcripts/ in order. They are plain text with no markup, so nothing is read aloud that should not be. See VOICE.md for the narrator setup.