On Tue, Nov 27, 2012 at 3:31 PM, Diane Hillmann <[log in to unmask]> wrote:
P.S. BTW, one of the best things about RDA is the relationships, all of which have URIs:
Since this is not code4lib, I will not reproduce my seattle lightning talk word for word :-)
If a property has the same meaning, but has different values when asserted about objects of different types, then you can use OWL Restriction Classes to express this.
Restriction classes may seem incredibly counterintuitive at first, but after a while they will seem merely credibly counterintuitive. So instead of diving straight in, I'll describe a simple example in simple English.
Example:
(1) Everything which is eaten by something is a food.
(2) Everything that a carnivore eats is an animal. Everything that eats an animal is a carnivore.
(3) Everything that a herbivore eats is a plant. Everything that eats a plant is a herbivore.
(4) Every fly is an animal. Every antelope is an animal. Every flytrap is a plant.
(5) Fred is a fly. Audrey is a flytrap. Ann is an antelope. Noel is a grass. Agatha is an antelope.
(6) Audrey eats Fred. Ann eats Audrey. Ann eats Noel.
Let's think about how this can be translated to OWL, then consider what other sentences are entailed.
Translating into OWL
(1) is easy to handle; it's saying something that is true for everything that is eaten by anything. This corresponds to a basic RDFS range restriction:
ObjectPropertyRange(:eat :food)
(2) is where things start to get harder. We want to say some things about eating that only apply to carnivores and animals. Let's look at the first part: "Everything that a carnivore eats is an animal.".
In a Description Logic like OWL, what we have to say is that
"the class of things that are eaten by some carnivore is a subclass of animal."
SubClassOf(ObjectSomeValuesFrom(ObjectInverseOf(:eat) :carnivore)
The second part works similarly - "the class of things which eat some animal is a subclass of carnivore"
SubClassOf(ObjectSomeValuesFrom(:eat :animal)
:carnivore)
The translation of the herbivore is similar.
SubClassOf(ObjectSomeValuesFrom(ObjectInverseOf(:eat) :herbivore)
:plant)
SubClassOf(ObjectSomeValuesFrom(:eat :plant)
:herbivore)
(4,5, and 6) are simple class, instance, and property value definitions, so I won't go in to them here.
Entailments
From our example, we can infer the following information
Agatha is an animal
Ann is an animal and a herbivore.
Audrey is a plant, a carnivore, and a food.
Fred is an animal and a food.
Noel is a plant and a food.
We know that Noel is a plant because he is eaten by Ann, who ate Audrey, making an a herbivore, and everything she eats a plant.
But see what happens if we add a sentence
(7) Ann eats Agatha.
Now we get the following inferences:
Agatha is a plant, an animal, and a food.
Ann is an animal, a carnivore, and a herbivore.
Audrey is a plant, an animal, a carnivore and a food.
Noel is a plant, an animal, and a food.
Fred is still an animal and a food.
This is an example of how the term "restriction class" can seem a bit of a misnomer. Under the OWL/RDF semantics, positive assertions license more inferences, rather than serving as constraints. Unless you specify that something is false, it won't be assumed to be so.
Agatha being eaten by Ann (who we already know is a herbivore) means that Agatha must be a plant. Conversely, Ann eating an animal means that she is a carnivore, which means that grassy Noel must be an animal.
If we add an assertion
(8) no carnivore is a herbivore,
If we try to run a reasoner, it will kick. Expunging the whole Ann eating Agatha incident will bring things back to the status q. a.
Simon
