Thursday, February 23, 2006
The Second School of Semantic Science
ontologyMapping Glass Bead Games
On the limits of the OWL standard à [184]
Reading material [1]
Reading material [2]
Reading material [3]
Summary of the discussion up to this point à [186]
Mapping between RDB, Object Oriented Modeling and OWL
Communication from Irene
Hi Paul,
It is my opinion two way (round trip) mappings between some of these paradigms are problematic.
Specifically, I am talking about UML (objects) and OWL. In my recent conversation with Dean on this topic, we've ended up concluding that this particular mapping may be even theoretically impossible (again, I am talking about two-ways). Or at least so complex, that it is practically impossible.
For many reasons our inclination is to take a pragmatic approach. Currently, we are looking at translations in the following ways:
1. RDB to OWL - this translation is interesting because we want to connect ontologies to the databases for either database integration or to offer more flexibility applications running on top. Our approach is to generate a proxy ontology from the db schema, connect it to a 'real ontology' (this could actually be very small) connections are done using either just RDFS or OWL (depending on the situation), use the proxy to translate ontology queries into SQL. [4]
What we do here is make each table a class and each column either a data or object property. We also do some special post processing to 'hide' (screen out) classes that are there purely for database normalization purposes.
2. UML to OWL (and XML Schema to OWL) - I see this as really a legacy model migration need. For example, sometimes UML models really represent domain knowledge and have been developed as such (with no intention to be ever transformed either manually or programmatically into the object code).
There may be some other uses and I have heard people talk about these. But, so far, I am yet to hear a clear (non fuzzy) description. I believe it may eventually emerge. Personally, I am more interested in how the use of knowledge models changes the way applications are developed -makes them thinner, more generic, etc.
For now, we have made some choices on the UML to OWL translation. We are still fine tuning them based on the different UML models we have tried. We believe that in most cases there will always be a need for some human intervention to turn these translated artifact into useful RDFS and/or OWL models. [5]Therefore, our goal for translation is not to be perfect (if there is such a thing), we just want to offer a useful utility that simplifies and streamlines this process.
No, we have not looked at using Topic Maps. You've probably already got the feeling from what I have written so far, that we are NOT trying to create a universal interlingua. [6]
On the other question you are raising, we are not DL fanatics, far from it. For example, we often use datalog and rules when and if needed. We understand RDFS/OWL very well and we do like them as standards. Having said this, I acknowledge they have shortcomings - then again I am yet to meet a perfect standard.
Why do we like them? One important reason is that unlike pretty much anything else, they have clearly defined semantics. In other representations, people may spend days and month discussing things like 'what does it mean for something to be a subclass, etc'. In RDFS/OWL, this is defined in clear and finite ways in terms of inferences each statement entails. These serve as complete definition. There are other reasons as well. [7]
I believe, when all is said and done, many new languages, paradigms and technologies initially seem difficult. Often, as tools, resources, best practices, etc. catch up, they get adopted and no longer feel difficult. It takes time for such transition to happen.
As you know, marketplace has its own dynamics and forces, with the technical merit being just one factor. For example, only yesterday I talked to a very experienced and prominent XMLer who (as many XML gurus) was singing praises to RelaxNG over XML Schema. And then in the next sentence, he said - 'there is no way RelaxNG will ever get a traction in the enterprise'. His conclusion had to do with the marketplace realities, not the technical foundation.
Regards,
Irene
[1] http://dip.semanticweb.org/documents/ECIS2005-A-Methodology-for-Deriving-OWL-Ontologies-from-Products-and-Services-Categorization.pdf
[4] Comment from Paul Prueitt: We feel that this is doable when the ontology is very stable. In many cases, this stability is being achieved in various bioinformatics ontologies such as BioPAX.
[5] The requirement for human inspection during mapping of one information model to another one is a Second School foundational concept.
[6] One wonders from a theoretical perspective why Topic Maps with its accommodation of the notion of interpretation as being required to realize the full understanding of “meaning” or “semantics”.
[7] We have come to understand this value to OWL that you have just stated.