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

About 19 minutes spoken, 2,815 words. Plain text for a reader app: 02-aristotle-to-wolff.txt

This is the second of ten talks about ontologies. Last time we separated three senses of the word: the philosophical discipline, a particular theory of what exists, and the machine-readable file. Today we trace where the word itself came from, over about two thousand three hundred years, and we finish with a question worth carrying through the rest of the series: when computer science borrowed this word, what came with it, and what got quietly left behind.

Let me warn you at the start that this lecture has more dates in it than the others.

Start with Aristotle, in the fourth century before Christ. Aristotle does not use the word “ontology.” Nobody does for another two millennia. What Aristotle does is identify the subject. In Book Four of the Metaphysics he describes a discipline he calls first philosophy, and he distinguishes it from every other science by the way it takes its object.

Here’s the distinction, and it’s cleaner than most people give it credit for. Every particular science studies things under a restriction. Physics studies things insofar as they’re subject to motion. Medicine studies bodies insofar as they’re healthy or sick. Each one cuts a slice out of reality and works on the slice. First philosophy takes no slice. It studies things insofar as they are — being qua being, in the standard translation. It asks what’s true of everything that exists, purely because it exists.

That sounds impossibly abstract until you notice that Aristotle immediately gets concrete. He wants to know: what are the fundamental kinds? What is it for something to be a substance rather than a property? Is a horse the same kind of thing as the colour of the horse, or the horse’s height, or the horse’s running? His answer is the Categories, a list of the basic ways a thing can be said to be.

Now, why should an engineer care about a list from the fourth century before Christ? Because that’s an upper ontology. That’s exactly the move that BFO, the Basic Formal Ontology, makes today. BFO is a small upper-level ontology that deliberately excludes physical, chemical, biological and other terms belonging to the specialised sciences. Its whole job is to supply domain-neutral categories that a specialised ontology can then sit beneath, and more than five hundred and fifty ontology-driven projects use it that way. Aristotle is doing the same job with the same motivation, which is that if you don’t fix the top-level kinds, every domain will invent its own and they’ll never fit together.

So the subject is ancient. The word is not.

Aristotle’s commentators, all through the Middle Ages and into the early modern period, recognised that being as such was a distinct object of study. But none of them thought a new word was needed. They said metaphysics, and that was that.

Then, at the start of the seventeenth century, somebody coined it.

One candidate is sixteen oh six. A German philosopher named Jacob Lorhard published a work called Ogdoas Scholastica, and in it the Latin word ontologia appears — the science of being. That’s the appearance Britannica credits.

Seven years later, in sixteen thirteen, Rudolf Goclenius printed the word in his Lexicon philosophicum, a philosophical dictionary. And priority between those two is contested. Ferrater Mora, writing without Lorhard in front of him, says the first instance occurs in Goclenius. So which date you get depends on which reference work you open.

There’s also a persistent misattribution worth knowing about. A whole line of respectable historians — Eucken, Gilson, Pichler, Wundt, Heimsoeth — name Johann Clauberg as the first philosopher to use the term. Ferrater Mora’s rebuttal is a date. Clauberg published Ontosophia only in sixteen forty-seven, by which time the word was already in use.

Now to the man who actually made the word stick.

Christian Wolff, the German rationalist, published Philosophia prima sive ontologia in seventeen thirty. First Philosophy, or Ontology. That’s the book that put the word into general circulation, and Wolff’s definition is the one worth memorising because it’s the high-water mark of the ambition: ontology is scientia entis in genere, quatenus ens est. The science of being in general, insofar as it is being.

Two things about Wolff’s programme matter for us.

The first is his method. Wolff proceeds demonstratively — rationally, deductively, from first principles, in the manner of a geometry textbook. Definitions, axioms, theorems. This is ontology as a proof system.

And underneath the method sits a claim about us. On Ferrater Mora’s account, Wolff took the human intellect to be capable of knowing the essence of things. That, as I read it, is the load-bearing assumption. Without it the deductions still run, but they stop being deductions about being and become deductions about your own definitions. And that ambition, that you could derive the structure of reality by deduction, is exactly the ambition that description logic and OWL, the Web Ontology Language, inherit at the level of technique while abandoning at the level of scope. A modern reasoner does deduce theorems from axioms. It just doesn’t claim the axioms are true of reality. It claims they’re true of your model, which is a much smaller and much more defensible claim.

The second thing is Wolff’s organisational scheme, and this one is a direct ancestor of how ontology repositories are laid out today. Wolff divided metaphysics into two parts. General metaphysics, which is ontology, deals with being as such — the categories that apply to anything whatsoever. Special metaphysics deals with particular kinds of being: the soul, bodies, God. General first, then special.

Look at what that maps onto. General metaphysics is the upper ontology: BFO, or DOLCE — the Descriptive Ontology for Linguistic and Cognitive Engineering — or the Suggested Upper Merged Ontology. Special metaphysics is the domain ontology: the Gene Ontology for gene function, CIDOC CRM, the Conceptual Reference Model, for cultural heritage, the Financial Industry Business Ontology for finance. And between them, modern practice has inserted a layer Wolff didn’t have — the mid-level ontology, like the Common Core Ontologies, which supply reusable notions such as agents, events, measurements, and facilities that are too specific for the top and too general for any one domain.

That three-layer structure, upper, mid, domain, is the standard architecture of ontology engineering, and it’s Wolff’s distinction with an extra floor.

I should flag something about that mapping, because I’ve now made it twice and I don’t want it to pass as history. It’s my mapping. None of the sources I’m working from claims that the engineering layers descend from Wolff. What the sources give me is the shape at both ends and nothing in between. At one end, Britannica’s Wolff: general metaphysics applies to all things, special metaphysics to particular kinds of thing. At the other end, Guarino, Oberle and Staab on modern practice: the primary purpose of top-level ontologies, they say, lies in providing a broad view of the world suitable for many different target domains. Same shape, no documented line of descent. Reinvention is at least as likely as inheritance. What I’ll claim is only this — the split gets made twice, by people with different problems, and that is worth noticing.

Since I keep saying upper ontology as though it were obvious what one buys you, let me say what the two I’ve named actually commit to.

The Basic Formal Ontology describes itself as a small upper-level ontology intended to support information retrieval, analysis, and integration across scientific and other domains. Read that as a list of jobs rather than a claim about reality. Find things, examine things, join things up. I said a moment ago that it excludes the specialised sciences, and the project states that exclusion as a warning to its own users: don’t expect it to supply specialised scientific terms, because those are outside its scope on purpose. So what it commits to is deliberately small. A shared upper-level structure for your domain ontologies to align around. The domain terms remain your problem, and that’s the deal rather than a shortfall.

The Common Core Ontologies commit to something more awkward, because a mid-level has two neighbours to answer to instead of one. It’s a suite of eleven ontologies of logically defined generic terms and relations, and it extends the Basic Formal Ontology, which its documentation calls an ISO-standard top-level — ISO being the International Organization for Standardization. The modules cover time, artifacts, currencies, geospatial entities, information entities and so on, and the classes are the ones you would otherwise rebuild in every project you ever work on: person, date, employment, nickname, measurement.

Then comes the part that tells you what it really commits to. Its authors explicitly encourage users to publish their own domain extensions rather than expanding it indefinitely with domain content. It is not meant to absorb content specific to individual domains. That content belongs in extensions, maintained separately, sitting underneath.

So each floor is defined as much by what it refuses to hold as by what it holds, and the refusal is enforced editorially — a governance board, a contributing guide, a standing instruction to take your domain content elsewhere. That last framing is mine again. The sources give me the policy; calling it the engineering form of general-versus-special is me.

So we have the word, we have the ambition, and we have the architecture. Now we get the demolition.

Immanuel Kant went after general metaphysics directly, and that attack is why Wolff’s programme stops being the mainstream. What exactly it amounts to is a philosophy course of its own, and the one-line gloss I’m about to give you is mine, not something my sources settle. As I read it, Kant’s charge is that Wolff mistook a feature of our thinking for a feature of the world.

I raise it because that charge has a direct engineering echo, and the echo is the useful part.

Mistaking a feature of the modelling apparatus for a feature of the modelled world is precisely the failure mode of ontology engineering. You build a model of shipping. It has a class called Delay. You get very good at reasoning about delays. And after two years, nobody in the organisation can think about a late shipment except as an instance of Delay with a cause property, because that’s the only shape the system will accept. The category has stopped describing your operations and started constraining them. That’s Kant’s objection, arriving on a Tuesday, in a stand-up meeting. We’ll see it again in lecture eight under a different name, when Bowker and Star call it torque, and again in lecture ten when a Danish police platform turns a car in a registry into a getaway vehicle.

After Kant, the word survives but changes hands.

In the early twentieth century Edmund Husserl, the founder of phenomenology, picks it up. Husserl calls Wolff’s general metaphysics formal ontology, and he contrasts it with special, regional ontologies — the ontologies of particular domains of being. Note that vocabulary. Formal ontology. Regional ontologies. If you’ve spent time in this field, you have just heard two phrases you already know. You probably thought computer scientists coined them in the nineteen-nineties. They didn’t. Martin Heidegger took the word in a different direction, arguing that the whole tradition had forgotten to ask what being means, and grounding the question in the analysis of human existence, which he called Dasein.

I mention Heidegger mostly to draw a boundary. Nothing in the Heideggerian line feeds into computer science. If you go looking for the philosophical roots of OWL you’ll find Aristotle, Wolff, and Quine. The phenomenological branch is a genuinely different tradition with a genuinely different question, and confusing the two produces a lot of bad writing.

One thing I skipped in that sprint through the centuries is worth going back for, because it turns up in modern practice wearing different clothes. It’s the argument about universals, and it has a genuinely surprising payoff.

The question is: when we say Socrates is human and Plato is human, is there a thing, humanity, that both of them share? Three answers were on offer. The realists said yes, universals exist independently of the particular things. The conceptualists said universals exist, but only in the mind. And the nominalists said there is nothing but individuals plus the names we give them — “human” is a word we apply to similar things, and there’s no further entity.

That argument ran for centuries, and the reason I’m telling you about it is that it’s the same argument two engineers have when one says “we should have a Customer class” and the other says “there is no such thing as a customer, there are just people who bought something, and ‘customer’ is a label we apply.”

Both of them are right, in the sense that neither position has won. The General Formal Ontology, a modern top-level ontology, deliberately offers several kinds of category — universals, concepts, and symbolic structures — precisely so that it can stay open across realism, conceptualism, and nominalism. That’s how unsettled the question still is. And the choice has consequences in the model. If you’re a realist about customers, Customer is a class, and a person either is one or isn’t. If you’re a nominalist, “customer” is a role that a person plays with respect to a transaction, and the right model has people, transactions, and a role that relates them.

That second modelling choice, incidentally, is the more flexible one, and it’s why upper ontologies treat roles as a first-class notion. The General Formal Ontology gives detailed accounts of roles and functions. So does gUFO, the lightweight form of the Unified Foundational Ontology, which separates a taxonomy of individuals from a taxonomy of types and makes role one of the type kinds. The reason is that treating a role as a class produces a specific and very common bug. If Customer is a subclass of Person, then a person who stops buying has to change class — and in my experience that is the change these models handle worst. Whereas if being a customer is a role a person plays in a relationship, they can start and stop playing it without becoming a different entity.

That is a live, practical, expensive modelling decision, and the medieval debate about universals is exactly where the arguments for both sides were worked out. Nobody tells engineers this, and I think they’d model better if someone did.

So what actually made the crossing into computing?

Three things came across. The vocabulary came across: ontology, category, formal, upper and regional, the whole lexicon. The architecture came across: general before special, upper before domain. And the deductive method came across: axioms, definitions, and a machine that derives theorems from them, which is what a reasoner is.

One thing did not come across, and it’s the one that matters most. Wolff’s ontology was meant to be the single true account of being, derived by reason, correct for everyone. An engineering ontology is nothing of the kind. It’s chosen for a purpose. Gruber’s phrase for these things is that they are designed artifacts, formulated for specific purposes and evaluated against objective design criteria. Notice what does the judging in that sentence. Design criteria. Not correspondence with being. There are many possible models of the same domain, and the good ones are good relative to a set of questions somebody needs answered. Genesereth and Nilsson said it in the line I quoted last time: a conceptualization is a simplified view of the world that we wish to represent for some purpose. For some purpose. Wolff would have found that sentence unbearable.

And here is where the trouble comes from. The vocabulary still carries Wolff’s ambition even though the practice abandoned it. When a vendor says “we have modelled the ontology of your enterprise,” the word is doing work that the artifact cannot support. It sounds like a discovery about your business. It is a set of choices about your business, made by people, for reasons, that could have gone otherwise. Every time you hear the word used with a definite article — the ontology, as though there could only be one — a claim of objectivity is being made that nothing in the engineering practice earns.

So, the claim to keep from this lecture.

Ontology began as the search for the one true account of being, and its whole vocabulary was built for that job. Engineering kept the vocabulary, kept the layered architecture, kept the deductive method, and quietly dropped the claim to uniqueness. Most of the overclaiming you’ll encounter in this field is that dropped claim sneaking back in through the word itself.

Next time, the philosopher who found a way to make ontology tractable, by refusing to answer the question at all.