30 June 2005

exegesis on formal ontology - or where I learn to sleep quietly at night again

Formal Ontologies are necessary.

I've said it, and I'm glad I said it!

The past several weeks, ever since becoming acquainted with Clay Shirky's diatribe against ontologies and digital marking (tagging) techniques (which are, at least according to Tim Berners-Lee, required for the semantic web), I have felt a deep blight covering part of my weltanschauung. I believe that within the area of data interoperability, and system composability, there is definitely a need for a formal specification of conceptualization (ontology). Where I believe that I and Mr. Shirky (whom I usually agree with, and enjoy reading) must part ways is this. When a group of individual people are attempting to all communicate, from different backgrounds, understandings and capabilities then ontology is not possible to concieve of on the part of the "speaker" (or in the case of posted information, the "producer"). This is because what is said is percieved of by someone with a different understnading. Because of this, when people talk, they are constantly refining, and re-refining their shared ontology. This is done through not only restatement of ideas in different terms, but also in inference and context. These things the human mind is capable of, where computers (not currently, if ever) are not.

This is where the idea of a "folkonomy" comes from. A folkonomy is a populist taxonomy derived from the bottom up, where people constantly contribute (the refining and re-refining), in the sense of a wiki. Perhaps with this, there will be a taxonomy (if not an ontolgy) that there can be general consensus on.

However, in the world of system interoperability, the situation is much different. Even though two computer systems have an ontology defining their data being interchanged, the two systems do not "understand" the data being exchanged. What they are capable of doing is understanding where in the ontology (which is a formalized structure defining the classes and instances of data being exchanged, and the allowable relationships between that data) the data resides. That is the level of "understanding" that a formal ontology brings to system interoperability. Does the ontology have to be semantically perfect? It has to be perfect enough to accomodate all the ways in which the participating systems can make use of the data (hence, all the possible properties and concepts making up the entities and relations being described). Does it have to be complete? That depends on what you mean by complete. It has to be complete enough to accomodate all of the aspects of data being interchanged, but it does not necessarily have to be a complete view of the whole universe of discourse (as would be required if the ontology were for a free communication mode, such as speech).

Note that this does get more and more complex when we speak of an ontology that will accomodate agile system to system connectivity and interoperability - such as when we have a large number of systems, which may or may not be known of when the ontology is derived. Any of these systems can be arranged to interoperate with each other on demand, and have to do so through the worldview afforded by the ontology - then it must be more "perfect", and more "complete" (as I've defined them above).

This where sufficiency comes into play.

So, Mr. Shirky, wherever you are tonight, I am now at peace with your writing. I believe that there is a home for metadata, tagging, and formal ontologies. But perhaps (and this coincides with my philosophical leanings) it is not achievable for general communications using the paradigms that we currently employ and are aware of. However, for the purposes of system interoperability, I believe that a "perfect" enough, and "complete" enough ontology is achievable with the defining and tagging technologies of today.

My weltanschauung is now complete again.



Blogger joe said...

First, Chuck, good post! You've got to know there's some absolutely great work here! (and of course elsewhere on your site!)
I also wanted to let you know I've been using the phrase "fractal ontology" in a slightly different sense (less directly technical in an AI) but not in an entirely dissimilar spirit.
We are, finally, interested in many of the same questions... I'm also curious about the role an "ontology" (schema or taxonomy of knowledge representation) would have to play in a linguistically-competent AI. You seem to believe the "tagging" problematic is currently insoluble (i.e., that no additional meaning inheres when a schema is imposed upon raw data); but am I right in saying that you see "sufficient" connectivity and inter-operability as a solution, or at least a workaround? Let me know your thoughts...
Glad your weltanschuungs' back to normal. Take care, buddy...

02 April, 2007 08:19  

Post a Comment

<< Home