On “Machine Readable” Ontologies
My official “I don’t get it” but maybe “we all do”
Professional writing, for me, usually takes a wide U-Turn or 3-Point Turn around a problem space. Provocative, pointed, focused on a central issue, but informative, conversational, abstract for the philosopher, and leaving the punch-line till last with a ta-da! that hopefully resonates. It’s what I do.
However.
When I read about RDF (Resource Description Framework), which I am more than happy to work with, and how it is “machine readable” I oft harper back to a lesson that I learned many years ago from a research scientist, Clifford Heath, who had a totally different view on the world to most and staked his claim in forging a new understanding of what “machine readable” actually is.
In an AI age, where machine readable means a machine reading just about any text that you can imagine, Mr Heath had a vision beyond his time, or at least one that stemmed from solid foundations of Controlled Natural Languages (CNLs).
You see…what Mr Heath did was simple. He said, if Object-Role Modeling (ORM), a very structured and concise conceptual modelling language for ontologies and databases, can have its ORM diagrams (that contain natural language predicates) converted into natural language (of any language)…as Controlled Natural Language…then you can write a parser to parse that CNL and create ORM diagrams or store an ORM model within the constructs of the ORM metamodel. Mr Heath’s CNL is called the Constellation Query Language.
Simple.
At least until you try and write such a parser. Which is exactly what Mr Heath did, and what FactEngine (my business) has done with the FactEngine Knowledge Language, following the breakthrough science of Mr Heath.
Let’s have a look at that opening salvo of database/ontology design statements in FactEngine Knowledge Language, a Controlled Natural Language ontology language, again:
Natural Language, easy to read, machine readable…human readable (per Mr Heath) and easy to get your head around.
Now let us look at just the first 4 lines in Resource Description Framework:
@prefix : <http://example.org/schema#> .
# Line 1: Each Lecturer in a Timeslot is in a Room (Ternary Relationship) :TimetableBooking1 a :TimetableBooking ;
:lecturer :Lecturer1 ;
:timeslot :Timeslot1 ;
:room :Room1 .# Line 2: Each Lecturer likes another Lecturer (Binary Relationship) :Lecturer1 :likes :Lecturer2 .
# Line 3: Each LecturerLikesLecturer involves at most one Lecturer :LecturerLikesLecturer1 a :LecturerLikesLecturer ;
:involves :Lecturer1 ;
:involves :Lecturer2 .# Line 4: Each TimetableBooking involves at most one Lecturer :TimetableBooking1 :involves :Lecturer1 .
What Mr Heath was saying, and we can all agree that the world is moving towards natural language understanding in a big way with AI…is that we need to shift our vision of what is machine readable in such a way that we do not confuse ourselves with otherwise laziness in terms of not-putting-the-effort-into-writing a beautiful parser.
That might sound like a harsh indictment of RDF (Turtle variant above), however if I had a choice, and I argue that if you had a choice, you would much rather read, write, transmit and use statements like these:
That is all we are saying. How many thousands of hours, if not millions of hours per year, can be saved by moving to a paradigm where we accept that computers are smarter than we think, and that so are humans. We can write parsers that interpret controlled natural language…machine readable does not mean something has to be complex, has-to look like machine-level-code, or have to confuse people.
And sometimes the philosopher needs to give way to the person inside who says, “Enough already!…there are easier ways to do things. All it takes is time and effort to write a beautiful parser and ultimately to treat Information Technology staff and the end customer with the respect of appreciating that most people are smart enough to get it”. We all get natural language…but we don’t all get RDF.
That respect is to say, “You know what? Sure…we can create a language that confuses people, takes up five-fold gigabytes of storage, takes a long time to learn and get right…or we can create languages that you understand, that you can enjoy usings, that are equally machine readable and simply do the job…and I will be happier for the experience, because you, as a customer and consumer of that quality, will be happier too”.
At least we think so.
Thank you for reading. As time permits, I will write more Object-Role Mdoeling, on Mr Heath’s Constellation Query Language and FactEngine’s Knowledge Language.
=============End===========
