Friday, January 20, 2006
[361] ß parallel discussion in “national debate” bead thread
This is part of a discussion
that will be moved to a Wiki page soon.
The Second School of Semantic Science, on
founding
[133] ß Judith Rosen’s recent
communication
Communication from the BioPAX community (Allen)
Hi Paul,
I have a few minutes so I thought I would respond to one of your notes.
> From http://www.bcngroup.org/beadgames/generativeMethodology/128.htm
>
> Alan, you mentioned in your recent note that OWL has a retrieval
> (OWL inference) that uses shared types, shared properties... and
> flexible combinations.
Have you seen this OWL specification work
> in more than one instance?
Do you know “HOW” this spec works or
> will work in theory?
>
> you (Alan) state:
>
> "There are other ways of establishing correspondences between
> things in the SW, e.g. shared types, shared
> properties, and relatively flexible combinations of such defined
in
> the OWL spec.
Paul
Here's what I mean. By establishing correspondence, I mean a computational way of computing some pairing of concepts. In the framework of OWL, we can look at what computational means using what we have available and from that see what kind of correspondences we can have.
Some examples:
Membership in a class can be computed. So correspondences between members of a class can be made, based on the class.
Properties relate individuals, so this another way to make correspondences.
Because OWL let's you define necessary and sufficient conditions to be a member of class (within a limited vocabulary), we can say these conditions are another way to form correspondences. So for example one may define the class of all things that have a "name" property whose value is "Alan". This can establish a correspondence between people named Alan, by putting them in the same class, even if they were not initially asserted to be in this class.
By not having the unique name assumption, OWL lets things be called by different names. By using nominals we can make rules which establish which individuals are to be considered to be the same.
A nominal in DL language is a class with an enumerated set of instances. If you make a class defined by necessary and sufficient conditions, which is a nominal with a single instance, then all individuals which satisfy the conditions will be considered to be the same. (as if sameAs was asserted between them). By using importing different ontologies on to the same set of instances one can even have that determination vary according to the need at hand. à (comment at [136])
OWL allows statements involving anonymous individuals. So for instance we can state that Alan and Paul share some friend, even if we don't know who that friend is.
OWL allows "general concept inclusions", which are relations between anonymous classes. So one can state that "the set of things with three sides" is the same set as the "set of things that have three angles".
OWL is good at expressing uncertainty. Saying something is a member of a class simply states what you know and leaves the rest unknown (unless some more information that further narrows it down). If you say that Human is subclassed into Man and Women, and you have an instance of Human, then all you know is that it might be a man or a women (or something else). If you add a covering axiom (saying that the set of Humans is exactly the union of the set of Men and Women) then you know that it might be a man or a women or both. If you say further that Men and Women are disjoint then you know it must be either a Man or a Women, but not both, and that you don't yet know which.
What you get from OWL that is quite useful is that it can check, in a bounded amount of time, that what you have stated is either consistent or inconsistent. Which I think is rather helpful.
You ask "Have you seen this OWL specification work in more than one instance?"
This depends on what you mean by work. I have seen each of the things
I describe above an action using the pellet reasoner. I have not built a complex ontology that takes advantage of all this yet. That's the work for BioPAX. We'll see how far we can go, and then decide what to do when we have a problem. So far we haven't taken advantage of OWL effectively, so that's the first thing to do. And remember, the goal of BioPAX isn't to simulate the world or do common sense reasoning, it's goal is to effectively communicate biological knowledge at some level. Any modeling is to the end of making that communication more effective. This is a lesser goal than some other ontology projects.
Regards,
Alan