Re: SUO: Conformance and Interoperability
Chris --
The reason two interoperable systems must draw the same
inferences from the same data is that the "definition" (or
"partial definitions" or "specification") of the concepts in
each system is, in my view, dependent absolutely on the
set of inferences that can be drawn, at least to the level
of inferencing that both systems will employ along their
inferencing chains. So to be sure that two systems are
indeed talking about the "same things", they must draw
the same inferences from any assertion about those things.
If, for example, one system says that a car *must* have
a motor and another says that a car *may* have a motor, then
an object in the real world which is a car whose motor
has been removed for repair will give different inferences
in the two systems. If both systems have only *complete* cars
in their instance database, then this difference will not be
obvious, but as soon as a motorless car shows up, they will
differ as to how to classify it and how to reason about it.
I don't think that practical systems can be depended upon
to be able to specify every instance of the concepts that
they will have to deal with, and that is the reason why
intensional rather than extensional definitions are useful.
When intensional definitions differ in any way (even if only
five or six links down a chain of inference), there is no
guarantee that some new object of interest will be considered
as the "same thing" in both systems.
As I mentioned, this is the "best case" of interoperability.
There are other, looser but possibly useful notions of
"interoperability". Certainly, systems that do not arrive
at identical inferences in every case may nevertheless be
able to usefully interchange information, and might be said to be
"interoperable", at lest for certain purposes. A second-best
case may be where one system (SYS-2) may have a part that is
identical to another (SYS-1) but also contain some additional
data and reasoning components. One can say that SYS-2 could
arrive at the same inferences as SYS-1, and therefore "know"
exactly what SYS-1 is talking about, but may then be able to
draw its own differing inferences, knowing that they may be
different from those of SYS-1. If a good Standard ontology
becomes available, with domain-specific extensions, this may
be a common situation in industrial data interchange.
There may well be good business reasons why neither of these
"interoperability" scenarios is practical for some users of
a standard ontology, but the users need to be aware that they
may well come to different conclusions from the same data by
using different ontologies, and that this may have significant
effect on their goals.
[CP]
> However consideration of business practice in this area might be
> useful.
Yes, but until there actually exists a good standard
ontology that all businesses can use, I think it would be
rash to assume that they would not leap at the opportunity for
genuine and reliable interoperabilty if it were available
to them. I don't think there is any such right now, and
I suspect that business practices that make do without
a standard ontology cannot be taken as evidence that it
would not be immensely more useful than the divergent
ontologies now in use.
Pat Cassidy
===================================
Chris Partridge wrote:
> Pat,
>
> I find your comment below strange - from a business point of view.
>
> "My understanding of the best case of interoperability is when two
> different systems presented with the same data derive the same
> inferences."
>
> There seems to be no reference to whether the data in the two
> systems refers to the same things. Typically in business systems
> the objective is to ensure that the systems are talking about the
> same things - e.g. USDollars not ITLira or even USDollars not
> lines, or bolts not nuts. If you look at standards for
> interoperating systems such as SWIFT or OFX there is no consideration
> of inferences, ditto engineering models such as EPISTLE. So a commercial
> requirement for interoperability is a reasonable guarantee that the
> two systems are talking about the same 'things'.
>
> If the SUO is intended to be merely for interoperability of
> ontologies, then deriving the same inferences may be a useful goal.
>
> However consideration of business practice in this area might be
> useful. One conclusion may be that they have got it totally wrong.
> Another may be that it makes sense to use other mechanisms than
> merely axioms.
>
> Regards, Chris
>
>
>
> -----Original Message----- From:
> owner-standard-upper-ontology@majordomo.ieee.org
[mailto:owner-standard-upper-ontology@majordomo.ieee.org]On
> Behalf Of Patrick Cassidy Sent: 26 October 2001 19:42 To: West,
> Matthew R SITI-ITPSIE Cc: Standard-Upper-Ontology (E-mail) Subject:
> Re: SUO: Conformance and Interoperability
>
>
>
> Matthew West has questioned the possibility of interoperability using
> a common Upper Ontology:
>
> [MW]
>>
>> In my view, there is only a benefit if using the SUO provides
>> interoperability between the different ontologies that use the
>> SUO as their core. So let us look at the degree of
>> interoperability that can be guaranteed between such ontologies.
>>
> . . .
>>
>> Bottom line: The use of a common upper ontology guarantees nothing
>> about different lower ontologies. Therefore having a conformance
>> statement that suggests otherwise is misleading.
>>
>
> My understanding of the best case of interoperability is when two
> different systems presented with the same data derive the same
> inferences. The requirements I envision for enabling such an
> outcome are either:
>
> (1) both systems use exactly the same ontology and inferencing
mechanisms.
>
>
> or . . .
>
> (2) both systems use the classes and relations of a standard upper
> ontology to define any new classes and relations, with all local
> extensions compatible with the Standard in not generating any
> contradictions. When interchange of data is desired, each system
> must transmit to the other system any newly defined classes and
> relations as well as specifying known instances of those new
> classes, and also transmit any new inferencing mechanisms not used
> by the other system. If the combined system with both sets of
> extensions cause contradictions to be generated, then some measures
> must be taken to eliminate the contradictions by changing one or
> both extensions.
>
> Method 2 answers Matthew's question:
>>
>> The question revolves around whether there is any benefit in
>> having ontologies that use the SUO as its core but with local
>> extensions being able to claim conformance for the whole
>> ontology, rather than just their use of the SUO.
>>
>
> Yes, if two ontologies do define their extensions using a common
> upper ontology there is a process available to achieve complete
> interoperability, but if they use different upper ontologies, no
> such process is available. The corollary is that, the smaller the
> extensions are as a fraction of the total ontology, the easier it
> will be to achieve true interoperability by the above process.
> There may be methods to map one ontology to another, but that will not
> achieve the level of interoperability achievable by use of a
> common upper ontology with such a reconciliation process.
>
> As has been pointed out previously, any one set of upper concepts
> can be used to derive different sets of new concepts that have
> properties differing from each other. True interoperability can
> only be achieved if those developing their own extensions are
> willing to share them with others with whom they want their systems
> to interoperate. A lesser degree of interoperability may still be
> achieved if extensions are not shared or prove contradictory, but
> this should surprise no one, and it seems to be the kind of case that
> Matthew has described. What is unclear from Matthew's note is
> whether he thinks that some form of mapping between incompatible
> domain ontologies that do *not* use the same upper ontology will be
> just as easy as a mapping between two domain ontologies that *do*
> use a common upper ontology. I would expect that mapping, where
> necessary, will be easier, even between contradictory ontologies, when
> a common upper ontology is used. Is there any experience to
> suggest otherwise?
>
> The ease of developing contradictory extensions is precisely the
> problem that has led me to argue on several occasions that,
> although we are concerned first with developing a Standard
> **Upper** Ontology, the definition of more specialized concepts
> that could be included in optional modular extensions should not be
> discouraged. It serves two useful purposes -- to make the intended
meanings
> of the higher categories more precise (by reducing ambiguity about
> the instances that fall under each class); and to provide common
> off-the shelf modules for those for whom they are adequate,
> reducing the chances that contradictory extensions will be
> developed by those who desire interoperability. The cutoff point
> for the "Upper" and the start of the "extensions" can be debated
> after some stability and agreement on the highest levels of the
> upper-most ontology have been achieved.
>
> What is still not known, and is a major question that development of
> a Standard Upper Ontology can help to answer, is in how many cases
> the different and contradictory treatments of the same concept in
> different ontologies have arisen simply because there was no
> widely-used standard ontology, rather than because their users felt
> that the different concept definitions were **absolutey
> necessary** to achieve the performance they desired. Pat Hayes has
> given examples of contradictory time theories. The question of
> 3-D versus 4-D objects has also been discussed, but I am still
> uncertain in that case that there is no way to express 4-D concepts
> with 3-D concepts that include time intervals explicitly in the
> assertions; in such a case they should be able to coexist in a
> single ontology without contradiction. I would very much like to
> see examples that cannot be translated between the two
> representations. Still, there is a widely-held opinion that *some*
> contradictory theories will be needed to satisfy the requirements
> of different users, and there is a corresponding widely-held
> opinion (I think) that the SUO should make some provision for
> including alternate theories (and also possible worlds and
> fictional contexts). The best form for this "lattice" of theories
> should, I think, be a major thread in these discussions. My
> feeling is that in order to claim to be a "standard" upper ontology,
every effort should be made even in the case of alternate
> incompatible theories to minimize the differences, by reducing
> them to the absolute minimum that different users insist
> cannot be reconciled. This should maximize interoperability
> between incompatible theories, which may then have to rely
> on some form of mapping to bridge the residual differences.
>
> In addition to Adam's requirements for "conformity", I
> would add that every leaf node of the SUO should have
> at least one and preferably multiple instances defined,
> and every conforming ontology must have those instances
> retained as instances of the original classes, though they
> may be distributed among newly-defined subclasses. This
> will reduce the chances of contradictions in ontologies
> using the SUO with different extensions.
>
> Pat Cassidy
>
>
> =============================================
> Patrick Cassidy
>
> MICRA, Inc. || (908) 561-3416
> 735 Belvidere Ave. || (908) 668-5252 (if no answer)
> Plainfield, NJ 07062-2054 || (908) 668-5904 (fax)
>
> internet: cassidy@micra.com
> =============================================
>
>
>
>
> West, Matthew R SITI-ITPSIE wrote:
>
>
>>Dear Colleagues,
>>
>>I have been arguing for a very narrow conformance statement
>>whilst Adam has been arguing for something more wide
>>ranging. In this note I will try to explain why a more
>>wide ranging conformance statement achieves almost
>>nothing and is potentially misleading.
>>
>>The question revolves around whether there is any benefit
>>in having ontologies that use the SUO as its core but with
>>local extensions being able to claim conformance for the
>>whole ontology, rather than just their use of the SUO.
>>
>>In my view, there is only a benefit if using the SUO
>>provides interoperability between the different ontologies
>>that use the SUO as their core. So let us look at the
>>degree of interoperability that can be guaranteed between
>>such ontologies.
>>
>>One thing I hope we would agree on by now is that it is
>>possible to have different conceptions of the world. At
>>the level we are at we have already found that there can
>>be a 3D or a 4D view. If the SUO stopped at Chris Menzel's
>>set theory, it would be perfectly possible for two groups
>>to work independently, one on a 4D ontology, one on a 3D
>>ontology. There would essentially be no interoperability
>>between these two ontologies, except where the set theory
>>concepts defined in the common module were used. In fact
>>even this would be minimal, because almost the first act
>>of developing a 4D ontology would be to create a concept
>>of a set with unchanging membership, which would then
>>be the primary concept of set/class used in the rest
>>of the ontology.
>>
>>On the other hand, these two ontologies can be integrated
>>by creating a mapping between them that says things like
>>if I have a continuant then I also have a 4D object of a
>>certain type, etc. I can then interoperate between 3D and
>>4D conceptions.
>>
>>Well this is a dramatic example, but in fact the chances
>>of any two groups independently developing an ontology
>>that is identical where it overlaps - even in the axioms
>>- let alone the labels is sufficiently remote as to be
>>fanciful.
>>
>>Bottom line: The use of a common upper ontology guarantees
>>nothing about different lower ontologies. Therefore having
>>a conformance statement that suggests otherwise is
>>misleading.
>>
>>
>>You will gather that I feel quite strongly about this,
>>and this is not without reason. I have been working in
>>ISO TC184/SC4 now for some 8 years. One of the standards
>>being developed there is ISO10303 - Standard for the
>>exchange of product model data. This was developed
>>(starting in the late '80s) precisely on a core that
>>was incorporated into different models that included
>>extensions for specific requirements. Many of the
>>developers of these standards expected that they
>>would be interoperable because they had a common core.
>>They were disappointed. The only parts that ended up
>>interoperable were those that were harmonised as part of
>>the development.
>>
>>This caused sufficient concern that in 1995 SC4 set up
>>a new Working Group, WG10, with one of its principle
>>tasks being to investigate why interoperability was
>>not achieved and what it takes to ensure interoperability.
>>
>>I am the Deputy Convenor of WG10, and I have lead this
>>work strand. The results are incorporated in ISO18876.
>>http://www.iso18876.org/
>>
>>A definition of lunacy: executing the same process twice
>>and expecting a different outcome.
>>
>>
>>Matthew West
>>Principal Consultant
>>Shell Information Technology International Limited
>>Shell Centre, London SE1 7NA, United Kingdom
>>
>>Tel: +44 20 7934 4490 Other Tel: +44 7796 336538
>>Email: matthew.r.west@is.shell.com
>>Internet: http://www.shell.com
>>http://www.matthew-west.org.uk/
>>
>>
>>
>
>
>
--
=============================================
Patrick Cassidy
MICRA, Inc. || (908) 561-3416
735 Belvidere Ave. || (908) 668-5252 (if no answer)
Plainfield, NJ 07062-2054 || (908) 668-5904 (fax)
internet: cassidy@micra.com
=============================================
> > every effort should be made even in the case of alternate incompatible
> theories to minimize the differences, by reducing them to the
> absolute minimum that different users insist cannot be reconciled.
> This should maximize interoperability between incompatible
> theories, which may then have to rely on some form of mapping to
> bridge the residual differences.
>
> In addition to Adam's requirements for "conformity", I would add
> that every leaf node of the SUO should have at least one and
> preferably multiple instances defined, and every conforming
> ontology must have those instances retained as instances of the
> original classes, though they may be distributed among
> newly-defined subclasses. This will reduce the chances of
> contradictions in ontologies using the SUO with different
> extensions.
>
> Pat Cassidy
>
>
> ============================================= Patrick Cassidy
>
> MICRA, Inc. || (908) 561-3416 735 Belvidere
> Ave. || (908) 668-5252 (if no answer) Plainfield, NJ
> 07062-2054 || (908) 668-5904 (fax)
>
> internet: cassidy@micra.com
=============================================
>
>
>
>
>
> West, Matthew R SITI-ITPSIE wrote:
>
>
>> Dear Colleagues,
>>
>> I have been arguing for a very narrow conformance statement whilst
>> Adam has been arguing for something more wide ranging. In this
>> note I will try to explain why a more wide ranging conformance
>> statement achieves almost nothing and is potentially misleading.
>>
>> The question revolves around whether there is any benefit in
>> having ontologies that use the SUO as its core but with local
>> extensions being able to claim conformance for the whole
>> ontology, rather than just their use of the SUO.
>>
>> In my view, there is only a benefit if using the SUO provides
>> interoperability between the different ontologies that use the
>> SUO as their core. So let us look at the degree of
>> interoperability that can be guaranteed between such ontologies.
>>
>> One thing I hope we would agree on by now is that it is possible
>> to have different conceptions of the world. At the level we are
>> at we have already found that there can be a 3D or a 4D view. If
>> the SUO stopped at Chris Menzel's set theory, it would be
>> perfectly possible for two groups to work independently, one on a
>> 4D ontology, one on a 3D ontology. There would essentially be no
>> interoperability between these two ontologies, except where the
>> set theory concepts defined in the common module were used. In
>> fact even this would be minimal, because almost the first act of
>> developing a 4D ontology would be to create a concept of a set
>> with unchanging membership, which would then be the primary
>> concept of set/class used in the rest of the ontology.
>>
>> On the other hand, these two ontologies can be integrated by
>> creating a mapping between them that says things like if I have a
>> continuant then I also have a 4D object of a certain type, etc.
>> I can then interoperate between 3D and 4D conceptions.
>>
>> Well this is a dramatic example, but in fact the chances of any
>> two groups independently developing an ontology that is identical
>> where it overlaps - even in the axioms - let alone the labels is
>> sufficiently remote as to be fanciful.
>>
>> Bottom line: The use of a common upper ontology guarantees nothing
>> about different lower ontologies. Therefore having a conformance
>> statement that suggests otherwise is misleading.
>>
>>
>> You will gather that I feel quite strongly about this, and this is
>> not without reason. I have been working in ISO TC184/SC4 now for
>> some 8 years. One of the standards being developed there is
>> ISO10303 - Standard for the exchange of product model data. This
>> was developed (starting in the late '80s) precisely on a core that
was incorporated into different models that included
>>extensions for specific requirements. Many of the
>>developers of these standards expected that they
>>would be interoperable because they had a common core.
>>They were disappointed. The only parts that ended up
>>interoperable were those that were harmonised as part of
>>the development.
>>
>>This caused sufficient concern that in 1995 SC4 set up
>>a new Working Group, WG10, with one of its principle
>>tasks being to investigate why interoperability was
>>not achieved and what it takes to ensure interoperability.
>>
>>I am the Deputy Convenor of WG10, and I have lead this
>>work strand. The results are incorporated in ISO18876.
>>http://www.iso18876.org/
>>
>>A definition of lunacy: executing the same process twice
>>and expecting a different outcome.
>>
>>
>>Matthew West
>>Principal Consultant
>>Shell Information Technology International Limited
>>Shell Centre, London SE1 7NA, United Kingdom
>>
>>Tel: +44 20 7934 4490 Other Tel: +44 7796 336538
>>Email: matthew.r.west@is.shell.com
>>Internet: http://www.shell.com
>>http://www.matthew-west.org.uk/
>>
>>
>>
>
>
>
--
=============================================
Patrick Cassidy
MICRA, Inc. || (908) 561-3416
735 Belvidere Ave. || (908) 668-5252 (if no answer)
Plainfield, NJ 07062-2054 || (908) 668-5904 (fax)
internet: cassidy@micra.com
=============================================
>> >>was incorporated into different models that included extensions
>> for specific requirements. Many of the developers of these
>> standards expected that they would be interoperable because they
>> had a common core. They were disappointed. The only parts that
>> ended up interoperable were those that were harmonised as part of the
>> development.
>>
>> This caused sufficient concern that in 1995 SC4 set up a new
>> Working Group, WG10, with one of its principle tasks being to
>> investigate why interoperability was not achieved and what it
>> takes to ensure interoperability.
>>
>> I am the Deputy Convenor of WG10, and I have lead this work
>> strand. The results are incorporated in ISO18876.
http://www.iso18876.org/
>>
>>
>> A definition of lunacy: executing the same process twice and
>> expecting a different outcome.
>>
>>
>> Matthew West Principal Consultant Shell Information Technology
>> International Limited Shell Centre, London SE1 7NA, United Kingdom
>>
>> Tel: +44 20 7934 4490 Other Tel: +44 7796 336538 Email:
>> matthew.r.west@is.shell.com Internet: http://www.shell.com
http://www.matthew-west.org.uk/
>>
>>
>>
>>
>
>
>
--
=============================================
Patrick Cassidy
MICRA, Inc. || (908) 561-3416
735 Belvidere Ave. || (908) 668-5252 (if no answer)
Plainfield, NJ 07062-2054 || (908) 668-5904 (fax)
internet: cassidy@micra.com
=============================================