Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: SUO: Unanimous consent




Dear Pierre,

See comments below.


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.west@shell.com
Internet: http://www.shell.com


> -----Original Message-----
> From: Pierre Grenon [mailto:pierre.grenon@ifomis.uni-leipzig.de]
> Sent: 01 September 2003 01:16
> To: West, Matthew R SITI-ITPSIE
> Cc: standard-upper-ontology@ieee.org
> Subject: RE: SUO: Unanimous consent 
> 
> 
> Dear Matthew and Jay,
> 
> some comments
> 
> > > -----Original Message-----
> > > From: Jay Halcomb [mailto:jhalcomb8@attbi.com]
> > > Re: the discussion of acceptability of SUO starter documents, 
> > > and the lack
> 
> [...]
> 
> > > Since we currently lack them, this is an opportunity for 
> the group to
> > > formulate some at least minimally necessary criteria for 
> > > acceptability of
> > > officially sanctioned starter documents. I suggest, to begin 
> > > with, that 
> > > 1)
> > > some formal specification of the language of an ontology 
> > > should be required,
> > 
> > MW: I agree.
> 
> My first thought is that we need to declare a common language 
> or a family
> thereof. CL comes to mind. Anything like a FOL language 
> should do enough to get
> to the core and approximate a good level of formalization. 
> OTOH, I am receptive
> to the overload claim. However, remark that a proposer of a 
> document should
> provide this service instead of requesting the whole group to 
> get familiar with
> yet another syntax. 
> 
> You cannot expect a group to form, to do the translation job 
> in logical syntax
> (which ought to occur, so it seems to me), then axiomatize 
> the intuitions
> expressed in a documentation before it starts wondering where
> are the putative inconsistencies, gaps and so on. The SUO WG 
> is supposed to do
> the second kind of tasks on a starter. At a later level of 
> maturity of the
> candidate, issues of adequacy and acceptability might arise 
> in addition.
> (Although some such debates should arise when deciding 
> whether a candidate
> starter has good chances to success if further developed 
> within this group.) 

MW: A problem of fixing on a single language is that it
inevitably has a viewpoint and intended purpose that might
not match your purpose. I don't think a KIF ontology makes
a very good database specification for instance.
> 
> On second thought, I think the language might not be such an 
> issue, especially
> if, as it seems, all we care about is a syntax (in the 
> context of a starter).
> From that standpoint, the main advantage of formal logical notation is
> readibility (over EXPRESS, say) and absence of ambiguity 
> (over English, say). 

MW: I would say EXPRESS was significantly more readable than KIF.
> 
> Actually, I might not whine at a document presenting tersely 
> in English the
> following elements of the theory:
> 
> 1) gloss of primitive vocabulary (including meta-language as needed)
> 
> 2) definitions
> 
> 3) axioms
> 
> Options: Examples, Notes explaining informally the background 
> intuition. 

MW: I agree. In fact I think this should be the real start point.
Formal axiomatisation should follow. I.e. write the specification,
then write the code.
> 
> > MW: This is available - at least sufficient to build compilers.
> > EXPRESS is specified in ISO 10303-11.
> 
> Do you have to buy that? 

MW: I think so.  You shold be able to find it in your University
Library though. Alternatively you could try looking in:
http://www.matthew-west.org.uk/Documents/princ03.pdf

P51 and following gives sufficient introduction to the graphical
form that the meaning should be clear. The graphical form maps
to a subset of the lexical form, but you miss very little if
you ignore what is not represented graphically, since we use
a subset of the language.
>  
> > > and 2) at least some formal axiomatization of content should 
> > > be required,
> > > together with an explicit indication, even if informal, of how
> > > axiomatization is intended to be done. 
> > 
> > MW: There is some formal axiomatisation. Each line on the diagrams
> 
> I'm leery about visual languages. They tend to hide the true 
> conplexity of the
> axiomatization and the semantics is far from being obvious 
> usually, if there's
> one. More generally, we've had a couple of PDF files with diagrammatic
> representations of BFO around for more than a year. Would you 
> find this an
> acceptible document? I hope not.

MW: EXPRESS is a quite different language than that used by the BFO.
It has both graphical and lexical form. The meaning of each is
well specified. Also we have used the language in a very regular
way, so once you get the hang of things you can make progress very
quickly.

MW: Having said that, tranlsating into other forms would be part of
the programme of work I would propose. Though I would probably prefer
to hold the ontology in a database and generate the different language
forms.
> 
> > represents an axiom, some other marks do too. I can appreciate you
> > may have as much difficulty determining what they mean as I do with
> > KIF. Don't expect sympathy though.
> 
> Then try English, just being rigorous. It reads well in your 
> papers. 

MW: Nice to hear.

> But
> really, S-KIF (SUMO) is manageable, CycL even more intuitive 
> imho (just strings
> of letters). That kind of things is close enough to plain 
> English with a
> slightly weird syntax and impoverished grammar, you can even 
> curse in CycL.
> Maybe that's a non native speaker illusion. 

MW: EXPRESS is no harder than these to understand. But as I have said
above, I think English should come first.
> 
> The point is that axioms should not be buried as intuitions, 
> be left implicit
> or mixed up with informal gloss in the documentation. They 
> should all be
> extracted and put naked in a section of a document. They can 
> be commented, if
> you use a formalism, they should probably be paraphrased.

MW: I agree. I see that as the next part of the work.
> 
> > MW: But let us try to quantify this:
> 
> I hope this is irony. Seems to me that we're speaking about 
> something that
> precisely can't be meaningfully quantified. 

MW: Thank you. Exactly the point I was trying to make. What I
see is everyone making judgements based on their own experience
and expectations (including you and me). We just don't have (yet)
enough truly common ground.
> 
> > 1. How many concepts should a starter document have:
> >    a) as a minimum?
> 
> 1?
> 
> According to your ISO habits, it seems that I could start 
> with one concept,
> say, individual and claim I have an exhaustive ontology. I 
> guess John Sowa has
> probably discussed that kind of ontologies somewhere.  

MW: Even zero, but the intention to have some would be fine
for a New Work Item. This would be unusual though.
> 
> Seriously, if you are clear about your intuitions or the 
> theory you propound,
> all the concepts that appear / are used / are necessary (not 
> sure which
> criteria are the right ones) in the theory you put forward to 
> provide adeqaute
> understanding / expressiveness should be declared. 

MW: I think that that is the job of the work stream, not of
an initial document.

MW: I wonder if part of the problem here is a clash of 
mentalities. It seems to me that what you are suggesting is
somewhat an academic approach, where the starter document is
equivalent to an academic paper - mostly the product of one
person, or a small team, delivering essentially a finished
product.

MW: This is very different from how I have seen standards
being developed, which operate very much on a quality 
improvement basis. So if you have a stated purpose as your
start point, you can quite literally start with a blank
piece of paper and raise issues against it to improve it
to better meet the purpose. As the issues are resolved
perhaps new ones arise, but eventually you get a document
that measure against the purpose has no major technical
defects, and probably then it is accepted as a standard.

Nearly everything you have said in this thread could be
formulated as an issue against the document I presented.
I probably agree with most of them. And then we have
a base line to move forward from.
> 
> >    b) as a maximum? 
> 
> The question is meaningful only as concern the defined 
> concepts. I'd say, only
> define what allows your to express more tersely the axioms of 
> your theory. No
> need to provide a list of concepts by boolean combination if 
> your theory
> conatins axioms which would do the job. (Just an example, I 
> hope no SUO does
> that, it might work for set theory, but this is dreadful in ontology.)

MW: Actually I was thinking of the upper boundaries of 
what makes an upper ontology, but I agree with your points
here too.
> 
> > 
> > 2. What ratio of axioms to concepts should a starter document
> >    contain?
> 
> Axioms in the starter should reflect your present 
> understanding and mastery of
> your theory. How many principles are there in your theory 
> that you would care
> enough to stipulate? 

MW: OK. So given that I am cautious, the number I have chosen should 
be OK then. Not that there aren't others not stipulated yet.

> Now, you don't have to shoot for minimality in the case of a 
> starter proposal,
> I think. 

MW: I agree.

> Finding a better axiomatization would be the job of 
> the WG, however,
> the point is to find a 'better' one, not to find one, which 
> supposes that
> there's one. This would presumably require more understanding 
> of a suitable
> logic. 

MW: I think raising issues of where the theory is missing key
elements (as is undoutedly the case here) is also appropriate.

> In that connection, it might be that you feel that some 
> principles ought to be
> deducible from other axioms. I would list them as axioms in a 
> starter proposal
> with such a mention. It would be useful for three reasons:
> - this would give a notion of what you consider useful to 
> state (even despite
> the fact that you feel that what you state is not so 
> fundamental as not to be
> deducible)
> - this would provide hints concerning the logic you assume 
> (it's not trivial
> for a proposition to be a theorem)
> - this would facilitate the job of the group when it comes to 
> work on the
> axiomatization

MW: I think all of this is work for an approved project. Not
necessarily content of a starter document. I even think it is
bad for it to be content of a starter document, because it means
there will be too much investment in it that means that there
will tend to be both a lack of group ownership and a resistance
to make changes.

MW: I think you are too much deliverable focussed than process
focussed. It is operating the process that results in a standard,
that is how consensus is achieved.
> 
> > > Finally, 3) that some
> > > characterization of the logical system or systems intended to 
> > > be employed in
> > > automated reasoning over the ontology should be given.
> 
> Jay, 
> I think that we can't expect that from a starter document, 
> unless the document
> itself makes claims about the logic. If you request the three 
> numbered items
> above, you have a theory. What you do with the theory, how 
> appropriate it is,
> which logic it requires and how the axioms might need to be 
> rewritten could be
> seen as development of a candidate standard within this group. 
> 
> But, I agree this is something we want the standard to 
> specify. It seems to me
> that a specification of a full logic should be a basic 
> elelemnt of the standard
> documents.
> 
> > MW: An EXPRESS based system ensures that the axioms specified are
> > conformed to by data sets that claim to be models of the ontology.
> > Not sure if that counts as automated reasoning. If not, why is 
> > automated reasoning a necessary purpose? I can appreciate it might
> > be where your interest lies.
> > 
> > MW: As it happens my intention would be to develop the material
> > from something that had been initially developed for one purpose,
> > to something that had wider applicability, but I don't see why
> > it should be necessary to complete that before it can be a starter
> > document (makes an oxymoron of the "starter" part to me).
> > > 
> > > Each of the 3 criteria above (particularly the formal 
> > > aspects) will have to
> > > be met at some point anyway, if a formal upper ontology is to 
> > > fit into, say,
> > > the IFF framework, or into a lattice of theories. 
> 
> Yes, but these criteria seems to me global in the context of 
> the Sowa drift, I
> mean lattice of theory (or is it registry?) 

MW: A lattice of theories held in a Registry (roughly equals database).
> 
> [...]
> 
> > 
> > MW: I think it is important to be clear about what stage you are at,
> > but what I am hearing from some people is that you should 
> necessarily
> > start at around the CD level. I think that is a bad idea.
> 
> It was fine for the SUO WG to start with a New Item. 

MW: Fine. Then why so much fuss about completeness of axiomatisation.
> 
> Inside this group, you want to have as much as you can at any 
> stage. It might
> seem to you that what I suggest is of an high level of 
> completion, but in my
> understanding it is not. The main part of the job is review 
> which involves
> internal development and comparaison with other starters. 

MW: I don't see any need for comparison with other starters. They
started where they chose to, not where I chose.

> And 
> this is nothing
> like what I suggest for the starter which is barely explicitation.
> 
> The problem is that this group tends not to be productive and 
> to act as a mere 
> decisional body. You could be more radical than I am and, 
> observing the way
> this
> group seems to function, claim that there should be no 
> starter at all, only
> candidates for being standard. We would just vote on yes and no (no
> standardization process here anyway). 

MW: This would be precisely a step in the wrong direction.

> At some point, when 
> less people use the
> word 'ontology' or when most of us changed their carrer 
> objectives, we gather
> all the rescapees of the votes put them in a tree (top node 
> is the ontology
> with one concept, of course, there would be just one level of 
> leaves, all
> ontologies on a par) and call that the Registered Upper-Level Ontology
> Standard.
> 
> > MW: I think it would be good to establish some process like 
> this with
> > associated deliverables. We could adopt the same levels as ISO (but
> > rename them) as a start for this.
> 
> Oh, I agree with that if you mean turning this discussion 
> list into a working
> group, I'm all for it. 

MW: Precisely what I am trying to do. I think there is no hope of
significant progress if we do not.

> In that case, it makes sense to 
> qualify what I've said
> in the previous paragraph. But I fear we're speaking major 
> procedural debates
> for a few years here. 

MW: I hope not. I have found that a few basic principles agreed are
enough. The stages of a process, and the quality criteria and improvement
process for the deliverables, and what we vote on when are enough.

MW: I would be happy to draft something if there was sufficient
support for this approach.
> 
> 
> -- 
> Pierre Grenon
> IFOMIS Uni Leipzig
> Haertelstr. 16-18
> 04107 Leipzig
> http://people.ifomis.uni-leipzig.de/pierre.grenon/
> pgrenon@ifomis.uni-leipzig.de
> phone: 49(0)351971672
> fax: 49(0)3519716179
> a.Á&
>