Contrast
between Semantic Web and Anticipatory Web principles
Related
discussion in InOrb technology bead thread .
Communication from Irene
Polikoff -> .
Beginning of notes indexed by
{n} where n = 1 - 5
1.
Would a 'good enough' common upper ontology provide benefits? {1}
Or rephrased, is there any reason why we would not want the Army, DoD, or
Federal government to adopt a common upper ontology (for systems
needing data/semantic interoperability ) . {2}
2.
Can we achieve cross-domain semantic interoperability without a common
upper ontology? {3} (Note: Any two ontologies can be
mapped, but this becomes an n-squared problem, which can't scale
up.) {4}
I usually
say that ANY upper ontology is better than none. But of course, I mean any
reasonably good one. An abomination would not be useful, but luckily there
aren't any of those.
Ultimately, no, you cannot. You can limp along for
some time without either an upper ontology or even a reference ontology for
your specific domain (i.e., the latter would enable you to map two domain
ontologies that are about roughly the same domain) with implicit compacts
(nodding our heads that yes, this is what we mean and we agree to this), but it
will hurt, and eventually you're going have to get your feet looked at, and the
ailment remedied. We don't want to generate a new generation of
"conceptual/semantic" stovepipes, but that is a real possibility.
Can we do everything immediately? No. Will upper
ontologies (and other reference ontologies) themselves evolve? Yes. But in
general, you don't wait -- at least, you don't wait if you want to minimize
cost, mitigate risk, and have more effective data and applications. You plan
now. There are a range of options/tradeoffs, but doing nothing, avoiding
evaluation, planning, etc., is probably the worse and most costly decision.
Dr. Leo
Obrst
The MITRE
Corporation
Upper ontology can be used weakly as a point of reference. Reconciliation of metadata or data can be made subject to real time reification and to community negotiation as in the SchemaLogic server technology. However, the origin of mental events is always formative and involves a mutual induction between the invariance across a vast memory of experience and the anticipations that are often situated in an environment (where no ontology exists – a system’s environment is “outside the system”.) So one can think about the formation of situated ontology in real time. The first way to do this is completely within human community or by an individual, what I called “descriptive enumeration” (Slide 5 – 7 K-Gov 2002 presentation).
The “good enough” ontology does not have to be itself standardized, if the process of forming ontology is well understood within a community of practice and the community is interested in seeing what has not been seen before. We do not standardize the sentences we are allowed to use in social discourse, do we? In human social discourse, what becomes “commonly” known are the usages of words and phrases, but in a social environment where nuance and novelty is easily detected and understood. Ontology usage can have a similar generative nature.
The issue is straight jacketing of what can be said or done with fixed upper ontology. In the related domain of Fixed Upper Taxonomy, the Anticipatory Web approach is defined using a social manipulation of a small set of words in a controlled vocabulary in order to help in the communication. I am positive that Drs Obrst and Hendler have something to say about the need for agility when using Fixed Upper Ontology; I just do not know what it is. It would be productive to hear this.
The standardization can come at the substructural level and on the data encoding forms (examples Orbs or hash tables), or transmission forms (examples XML or data compression tables), that content is placed into. What Jim Schoening’s question seems to address is the assumption that taxonomical or subsumption relationships can be usefully written into stone. So if something happens, one asks, “Is this military or civilian”? Contextualization can suffer from this type of standardization. Is this not true?
Are Semantic Web advocates standardizing knowledge representation to be hierarchical only, or can simple graphs be used that are not assuming these hierarchical relationships? What about OWL and the dependency on first order logics, does this become standardized? Again, I know that there are answers that can be given to these questions, but I just do not know what these answers are.
Understanding that there is substructural invariance in social discussion is something that has been hard to make computer science out of - at least that is the impression that I get. But the issue here is not about computer science; it is about natural science and in particular the social and cognitive sciences. Remember that the periodic table for physical chemistry was not available until the physical science evolved to the point where an understanding of atomic structure simplified chemistry. Social system behavior, or human behavior may have similar decompositional science. However, if the standards for ontology expression explicitly do not recognize Anticipatory Web designs, then neither funding will flow to this new science nor will very much progress be made.
The importance of standards has to be weighed by the limitations that standards do place on practice. Are there alternatives to standards as advocated by W3C, OWL and RDF communities? Yes, the first alternative is the Topic Map standard.
But Topic Maps have to be seen as a real standard, not depending on RDF, and not something that Jim Hendler claims to be reducible to RDF. It is not.
The Topic Map standard is not reducible to RDF in exactly this sense that parts of ontologies can be reified by the human in the loop in real time to provide a clear and transparent representation of what is actually being communicated between humans in the situation – rather than something else. But one uses what I call mutual induction (between machine representation of structure and human understanding of function) to reify the graph constructions that then act as cognitive primers. No RDF, and no first order logics in how the human mental events occur. But there can be OWL services used in the production of Orb constructions and Orb constructions can be read into RDF format. This is a simple matter.
The trick here is to separate structure from meaning. One does not want a standard meaning for a word, or for ontology construction. One needs only standardize the forms used to communicate structure from one user to another user. Human mark-up language standards work demonstrates some advanced practice in producing standard formalism that is then used to mark-up a model of human behavior. Perhaps we are working far to hard on this standardization thing?
The ontology merge problem may be set up incorrectly in most people’s minds. Certainly this has been the biggest visual problem with Cyc ontologies. Micro theories seem to help. However, the Anticipatory Web approach creates a set of sub-structural invariances, encoded simply as ordered triples having the form
< a, r, b >
where a and b are locations and r is a relationship variable.
A new PowerPoint Presentation on this is