Re: [xml-dev] Generating New Knowledge by Deductive Reasoning using Sche

[Stephen Green <stephengreenubl@gmail.com>]

Thanks Rick. Great stuff to look into. I'm especially interested in how
best to output the report such that the report can be 'queried' by some
subsequent Schematron rules. Now it occurs to me that there are two
features which might need to be a lot more mature so that we can have
a kind of logic ecosystem supporting automated deduction alongside
statements in prose which are not necessarily for automated deduction:

1. The assertions could be more general statements of 'facts' rather than
being limited to test assertions. (Can a Schematron assertion/rule include a
'statement' of fact which is not necessarily a assertion to be tested?)

2. The output from both assertion-driven 'tests' and more general 'deductions'
based on the more general 'statements' can be marked up in the same markup
so that they provide input for further tests and deductions

I worked with OASIS TAG TC on something similar, as you know, but even
here there may need to be more progress or external development such that
'test' and 'deduction' results can share the same markup with the
input. i.e. that
markup may need to be generalised beyond 'assertions' to include other kinds
of statements too so that any output can be represented with the same markup.
Plus the 'deduction' side of things put forward by Roger could perhaps do with
more support.

Ontology languages might need adaptation too to support their use in both
representing categories of 'contexts' and in providing input for deductions
(so that the deduction 'Fred Blogs is the robber' includes that 'Fred Blogs is
a criminal', since 'robber' is a subclass or subcategory of
'criminal', etc). Having
Schematron allow such 'is-a' relationships regarding its contexts or allowing
category ontologies to be written such that Schematron and other such tools
can read them might help create this ecosystem too. Or is that already part
of Schematron?

Best regards

Steve
---
Stephen D Green




On 21 November 2010 11:45, rjelliffe <rjelliffe@allette.com.au> wrote:
> On Fri, 19 Nov 2010 14:56:35 +0000, Stephen Green
> <stephengreenubl@gmail.com> wrote:
>>
>> A very nice example, Roger. Maybe what would make it look more
>> useful would be to demonstrate how to chain the deductions so that
>> the deductions can themselves add to the 'knowledge' and further
>> deductions be made based on those initial deductions. This might then
>> provide an improved parallel methodology to the more typical OWL
>> or RDF knowledge-base / triple-store / inference-engine approach.
>>
>
> Just use the <sch:let> expressions more. You can have arbitrarily long
> chains of deductions, and pull all the logic out of the assert or report
> tests.
>
> <sch:let name="speeder-driversLicenseNumber" value="."/>
>
> <sch:let name="GunLicense"
> value="for $i in
> collection('GunLicenseFolder?select=*.xml;recurse=yes;on-error=ignore')
> return $i/GunLicense[.//Person/driversLicenseNumber eq
> $speeder-driversLicenseNumber]"/>
>
> <sch:let name="matching-drivers-license"
> value="$speeder-driversLicenseNumber eq
> $GunLicense//Person/driversLicenseNumber" />
>
> <sch:let name="matching-gun" value="$GunLicense/registeredGun/Gun/serial eq
> doc('Robbery.xml')/RobberyEvent/evidence/Gun/serial" />
>
> <sch:let name="has-gun-license"  value="count($GunLicense) eq 1"/>
>
> <sch:let name="only-person-licensed"  value="count($GunLicense//Person) eq
> 1" />
>
> <sch:let name="is-bad-guy"  value="$matching-drivers-license and
> $matching-gun and $has-gun-license and $only-person-licensed" />
>
> <sch:let name="is-our-bad-guy"  value="parent::Person/child::name =
> document('licensed-to-kill.xml')/agents/agent/name "/>
>
> <sch:report test="$is-bad-guy  and not( $is-our-bad-guy)" >
> Therefore, <sch:value-of select="parent::Person/child::name"/> is the
> Robber.
> </sch:report>
>
>
> So you can add extra deductions on the same subject.  (The let expressions
> are scoped to schema, pattern, or rules, but I don't know that this is a
> disadvantage.)
>
> Another way to approach this is neural net fashion: you have one Schematron
> schema generating individual deductions, then subsequent schemas as layers
> generating deductions based on the results of layer before (eg the SVRL
> output.)  (Schematron supports an attribute "flag" to help declare the found
> properties of a pattern/rule/assertion too (such as "incomplete data" or
> "security risk found".)  You can physically combine the patterns for the
> different layers into the same text file using the phases mechanism, if
> appropriate too. (Of course, command-line parameterization has been
> available for the last decade with Schematron, however a WWW technology may
> be more interested in configuration from XML documents at some URLs rather
> than just command line arguments.)
>
> By the way, there has been a version of Schematron that operated on RDF not
> XML: Schemarama used the Squish query language. However, I imagine that
> there are a lot of research tools available for RDF.
>
> Cheers
> Rick Jelliffe
>
>