Tony --

Adrian wrote> A problem with RDF-based inference is that it quickly becomes human-opaque, because the notation is machine-oriented and the inferences are, well, intricate.
 
Tony wrote in reply> Important to note that RDF does not have any native 'notation' - it has a data model which lends itself to multiple notations (or serializations) - some of which may be less opaque than others. As noted elsewhere the problem with RDF notations is not XML or whatever, it is the simple fact that resources, properties, etc are identified by URIs - which themselves can easily become non-human-transcribable (contrary to the principles laid out in RFC2396, sect.1.5).

Adrian continues the discussion> Yes, if URIs were more human readable, that might help.

But IMHO, more is needed.   Consider the following inference step from [1]:

Paper is related by fact#:title to  An Overview of RDF Query Languages 
Paper is related by fact#:author to __Description1
__Description1 is related by rdf:_2 to aeb
aeb is related by fact#:name to  Andreas Eberhart 
not: there is a known email address for the person with home page aeb
--------------------------------------------------------------------------------------------------------------------------------
Andreas Eberhart  is an author , with email (unknown) , of  An Overview of RDF Query Languages 

If we were given only several hundred triples and the conclusion, it would be quite hard to see how the first led to the second.  And this is just a simple tutorial example.  Consider the kinds of complex inference that would be needed to conclude from millions of triples, say,

the feature A000253 is a pcgene (a gene where at least one of the parts is mRNA)

Just figuring out how inferences from triples lead to such a conclusions could take up much of the working day, and would be error-prone.

On the other hand, if we represent the knowledge that is needed as English-like inference rules, one can just ask for an explanation/proof that the conclusion follows from the knowledge and the data.  The inference above is a step in such an automatically generated explanation,

Does anyone on the list know of other work in this direction?

                                           Cheers,  -- Adrian Walker


[1]  The demo called RDFQueryLangComparison1 at www.reengineeringllc.com
-- 

Internet Business Logic  --  online at www.reengineeringllc.com

Reengineering LLC,  PO Box 1412,  Bristol,  CT 06011-1412,  USA

Phone 860 583 9677     Mobile 860 830 2085     Fax 860 314 1029