comments on OWL Last Call drafts

I have been thinking about the OWL Last Call drafts for some time and 
I have been troubled by a general feel of unease.  While the use 
cases provide a broad range of interesting and useful scenarios, my 
general feeling is that OWL as presented will not be an effective 
means to accomplish these ends.  I would be more comfortable if I 
could point to specific shortcomings that I could suggest be 
corrected (such as the n^2 problem with disjoint classes - why not 
have a mutuallyDisjoint construct that is applicable to a list of 
arguments?), but my issues go deeper than this.  In discussions with 
others (including some outside SAIC), I have encountered a pervasive 
feeling that OWL might be interesting but will not have significant 
impact on real problems.  Any examples in the drafts that go beyond 
the trivial are complicated and tedious, and it is unlikely that 
anyone beyond a diehard would have the motivation to do any 
significant ontology capture in this format.

What appears to be a significant shortcoming is the lack of explicit 
support for real life problems that involve interacting ontologies. 
While there is significant mention of ontology reuse and combining 
independently generated ontologies, the drafts seem to miss the point 
that independent ontologies have an overlap in coverage that in most 
cases cannot be captured as a one-to-one equivalence but rather show 
an incomplete and possibly inconsistent interpretation of a domain. 
This is the information that gives value to ontology combination and 
reuse. OWL has no means to capture this.  For example, if I have one 
ontology where my name is captured as "Ken Laskey" and I have a 
second ontology where nameFirst captures "Kenneth" and nameLast 
captures "Laskey", how does OWL represent this correspondence at the 
class level?  Note, even at the individual level, sameIndividualAs 
may not be sufficient because a Google search identifies more than 
one Ken Laskey, and it is not feasible to annotate each individual.

Moreover, search is evidential and inductive in nature. I need not 
only to be able to do searches, but also to represent the partial 
correspondences I find in the results.  How does OWL support the need 
to show an accumulation of evidence towards a conclusion rather than 
trivial cases where the answer is close-ended?  In a real problem, I 
need to be able to accumulate evidence about whether a new mutation 
of SARS is appearing or whether I have a local increase in lethality 
of the standard virus.  The ontology and the individual information I 
need is much more fragmented than that which I would use to choose a 
wine.

There are other concerns I have about OWL.  We are creating a means 
of capturing information for use by other processing engines but I do 
not see that we have a clear understanding of what these engines are 
and what content and form of information these will need for their 
tasks.  For example, the multimedia use case refers to a knowledge 
fragment "typically made of mahogany".  How would a processing engine 
draw this conclusion, store this conclusion, or use it in further 
processing or search?  I know that there are many examples of how I 
can specially craft an ontology to conceptually support such tasks, 
but without a range of examples on how engines would actually do this 
work, we do not know what information needs to be captured at the 
meta level represented by OWL or how it should be efficiently 
represented.

Now in response to many of my concerns, one could say we have to 
start somewhere and OWL is as good as (if not better than) most 
alternatives.  However, in work we have done at SAIC, we have seen 
that when an ontology is created by one group of people (however 
knowledgeable) for use by a wider community, the ontology reflects 
the task priorities of the "first one in" and results in frustration 
to those outside the process because it constrains use from a 
different perspective.  Rather than facilitating discussion, the 
initial view often stifles growth.  All too frequently, the result is 
that the ontology is used only by its initial adherents.  One could 
argue that this is what has happened with RDF.  A Working Group (even 
one that endeavors to be inclusive) by nature codifies the context in 
which it will do its work.  Given we all acknowledge that many 
different ontologies represent various views of a domain, what 
attributes of OWL overcome shortcomings of a single ontology created 
by a single group and make it the single sufficient basis for a 
general-use standard across the Web?

Finally, in several places in the drafts, it is noted that "tool 
support" will be required, such as for maintaining consistency for 
merged ontologies.  However, I am concerned that there will be no 
impetus to create tools unless OWL ontologies can be shown to be 
useful without the tools and that tools will provide needed leverage 
for people who find value in the manipulations the tools do.  Large 
developers will not create tools just because the tools can do neat 
things unless there is some rationale for a larger market.  Do we 
have significant information that vendors will embrace OWL, either to 
create tools or, more importantly, for use in their internal 
infrastructure?

With these Last Call drafts, we are heading towards standardizing 
something which does not clearly show the ability to enable solutions 
on the scale of the Web.  The level of complication and the lack of 
general robustness of artifacts built on OWL make it unlikely it will 
be used for more than focused problems on a limited scale.  While 
this may be useful for further research, moving forward to give the 
imprimatur of a standard to something which is unlikely to have wide 
and significant impact would only diminish the significance of W3C 
Recommendations.  For this reason, I strongly encourage that the use, 
purpose, and syntax of a web ontology language be rigorously 
reconsidered.

Ken Laskey
-- 

    ****************************************
  *   Ken Laskey				 	*
  *   4001 Fairfax Drive	phone:	703-276-4804	*
  *   Suite 300		fax:	703-524-1643	*
  *   Arlington, VA 22203                           	  	*
    ****************************************