Tuesday, February 21, 2006
Generative Methodology 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]
Related Comment from John Sowa à [193]
Dialogue about ontology mapping
Raj,
Your insight [191] leads you to ask questions similar to those I, and others, have asked.
I am not sure about all of the details regarding this question of mapping.
see:
http://www.bcngroup.org/beadgames/generativeMethodology/186.htm
I see two consequences of the debate regarding invertible mappings between OWL ontology and RDBMS schema.
One is a better understanding of what Descriptive logic based ontology "reasoning" is and IS NOT.
The second is a better understanding of what is mean (and NOT MEANT) by the OWL literature when the phrase "open world assumption" is used.
One should ask: what is your definition of open world assumption? what is your definition of reasoning?
We have asked these questions because these definitions might, and I stress might, be confusing not only you (and me) but almost everyone.
The open world assumption as stated in the OWL/Protege literature is about the ability to add new structure to an ontology (in certain cases) and to add information that may or may not be complete (as is often required of a database schema). Data added may also to an OWL ontology that is not formatted correctly. This is a function of XML. XML allows data that does not fit the “schema” of an OWL contraction to exist without causing the OWL construction to fail.
What the open world assumption as stated in the OWL/Protege literature is not is an assumption that any new structure may be added to an existing OWL ontology.
For example, a class within a subsumption hierarchy cannot be removed completely unless that removal correctly modifies all dependencies on that class. A question arises as to how this complete modification can occur. More critically however is something like the "evolution" of a new class from an instance.
I firmly believe that changes in an OWL ontology will eventually be instrumented so that there are naturally evolving categorical structure within the classes, properties and constraints - but at present this evolution is not the sameAs what one finds in ecological or biological reaction networks... such as what BioPAX is modeling. As a consequence, OWL can function as an Information Model for data and process exchanges within distributed scientific communities, but OWL cannot function is if mathematics in regards to an ability to model completely the causality (entailments) that are occurring in nature. The usefulness of the Information Model is very important.
Regarding "reasoning". OWL ontologies equipped with Racer, FACT++ or Pellet function is a program over an index.
The logic is deductive and flows according to deterministic rules. In theory, these rules can be defined as stored procedures working with relational data. If the OWL ontology is fixed, such as BioPAX level 2, and there are reasons to flow the ontology with the instance data into a Oracle database, then this can be done. Why not? In both cases, one has a simple (non-complex) finite state machine.
The debate over mapping OWL to RDBMS (including the OWL indexing) will gain momentum because the RDBMS is an incumbent technology.
Some feel that this debate will soon result in actually constructed proofs that major OWL ontologies can be translated into RDBMS schema and performances compared. OWL has problems both with the persistence as well as the efficiency of the organization of data and its indexing. These are well known issues. A behavioral match between a RDBMS and an OWL ontology with instances will not have these same efficiency and persistence problems.
In the area of web services, this is particularly important since efficiency and persistence is compounded with the need for a standard (ie fixed) information model).
Ontology of the descriptive logic type may be excellent for designing how data are to be organized and indexed in standard information modules. But the design stage may be followed by a persistence paradigm that we are only now discovering.
Now I have to stress that my understanding my be incorrect at more that one point, so I ask that if someone sees how and why I have made an error that this person make clear what the error is and perhaps this will change my mind.