[Top] [All Lists]

Re: [ontolog-forum] Relating and Reconciling Ontologies (legacy systems)

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: Mike Bennett <mbennett@xxxxxxxxxxxxxxx>
Date: Sun, 24 Apr 2011 12:50:02 +0100
Message-id: <4DB40E6A.3090400@xxxxxxxxxxxxxxx>
There's an important question in there I think: do ontologies 
show a truth or a view?    (01)

If we take the view that ontologies show a view of the truth, 
with the truth itself being "out there" in some Platonic way, 
then reconciling different ontologies becomes a matter of 
reconciling different views of the (same) truth.    (02)

Or to put it another way, different ontologies make different 
ontological commitments in relation to how they choose to model 
the truth (so explicitly, an ontology never claims to be a truth, 
only a representation of the truth). Those ontological 
commitments may vary according to differences in granularity, in 
terms of what features of the truth are seen as primary for this 
particular application (think Tube map versus street maps of 
London), and in other matters like dimensionality - but these are 
all conscious choices in how the truth is to be represented in 
the ontology, they are not disagreements about the truth.    (03)

All of which suggests that mapping and reconciling ontologies 
must be a finite and tractable task, for as long as those 
ontologies are all made by people who agree about the truth in 
the subject matter in question.    (04)

Mike    (05)

On 24/04/2011 12:28, Patrick Durusau wrote:
> David,
> On 4/23/2011 6:14 PM, David Eddy wrote:
>> Jack -
> <snip>
>>> A topic map values all world views
>> I'm not sure I agree with how this is said.
> +1!
> Topic maps, like any other map, have boundaries, either by design or
> necessity.
> Think about the London Tube map or a map of the New York subway system.
> Both are quite valuable for their intended purposes and despite being
> maps, you would not consult either one for Martian cartography.
> A topic map is constructed by one or more authors with mappings that are
> of interest to them or thought to be useful for others.
>> The world view of "a system" (an extremely slippery concept) is
>> actually quite constrained.  I worked on a 3,000 year old banking
>> application (Letters of Credit) that contains 808 distinct data
>> elements and 387 unique terms.
>> In the context of a/many legacy systems what I want is the ability to
>> easily find what's in Silo A (likely on a VERY narrow view).
>> That is... I want the ability to ask for "policy number" which I can
>> guess at&   be shown that since I'm in Silo A, one of the significant
>> technical names is M0101.
> Certainly it is the case now with software such as Talend and others, to
> create what I would call *blind* mappings that are point to point
> mappings between terms. That is there might exist a mapping between
> "policy number" and "M0101."
> But that is hardly useful unless you know one of those two terms. And
> should you want to extend that mapping, to say silo X, you really have
> no way to know why the mapping was done in the first place. It could be
> that it was what we are assuming, a mapping of policy number, but it
> could also be the mapping of a primary ID. Or some other mapping. Hard
> to say without more information.
> A topic map mapping can do more than simply point from one string to
> another, it can also include *disclosure* of what properties at both
> ends of the mapping must match in order for the mapping to occur.
> And each of those properties are also subjects which can have properties
> and so be identified in various ways.
> None of which is a magic talisman nor foolproof. But the disclosure of
> why a mapping was made offers the chance, albeit never a certain one,
> that you as a human reader will recognize a subject across a semantic
> gap and realize that is the same subject you are looking for.
> Think of the days when we all used the Reader's Guide to Periodical
> Literature and for every entry in the index, we simply trusted there
> were some properties that resulted in that entry being made.
> Topic maps offer the capability when we get to that entry to see what
> properties were seen as meriting that entry. We may or may not agree
> with it but we have a different (note I did not say better) opportunity
> to create more mappings based upon that information.
> If you think of topic maps as offering the capability to capture the
> basis for the mappings between systems that we do anyway, that comes
> closest to being a shorthand view of them.
>> Then if I say I also want to peer into Silo B, magically MSTR-POL-NO
>> is what I'm needing to know.  Silo C... contract_id does the trick.
>> In extreme circumstances, repeat 67 times (this might be something of
>> a UI challenge.)
> The challenge isn't a UI one but of discovery of those relationships and
> mapping them with properties to show why.
> There is no magic bullet that will substitute for some process of
> determining that M0101 (Silo A) and MSTR-POL-NO (Silo B) and contract_id
> (Silo C) all represent the same subject and to provide a basis for
> mapping between them.
> However, once that discovery is made, whether by you, a data analyst, an
> intelligence officer, a topic map can capture that as organizational
> memory and so the next person who comes looking, has an easier time of it.
> Noting that you only need view as much or as little of the mapping(s) as
> is needful.
> The real goal isn't 10,000 "hits" including blogs, tweets, webpages,
> ads, etc.
> It is to have one (1) "hit" with the correct answer (for some particular
> circumstance, ....all the usual qualifiers).
>> To date I have neither seen, nor heard ANY interest in this seemingly
>> mundane task from the world of ontologies.  John Sowa excepted.
> Well, I hate to defend ontologies, ;-), but you can think of ontologies
> as being one slice of a topic map pie that represents a particular world
> view and relationships within that world view.
> Nothing wrong with that and for some purposes, like relational
> databases, can be quite effective at particular tasks.
> It depends upon what you want to do.
> If you are trying to map across information systems in ways that allows
> for merger of information from different information systems, where
> conversion would result in information loss, then you most likely need a
> topic map.
> If you want to preserve your information domain as "your" domain for
> which you are the gatekeeper, then conversion using some basis known
> only to you is your best bet. (or that works with "your" software)
> Ontologies, save you the time of creating your own map of some domain,
> but at the expense of using its view of a domain. Which may or may not
> be a close enough fit of your view to be useful. But there is no denying
> they have been useful in a number of contexts. My only reservation is
> the notion that any of them represent some sort of "truth" rather than a
> "view" of a domain. If the latter, then we should be able to have
> different "views" of the same domain, mapped to each other.
> Hope you are having a great weekend!
> Patrick
> PS: In a shortened form: There are no universal maps just as there are
> no universal ontologies. There are maps and ontologies that are useful
> for particular purposes, times and circumstances. The more that is
> disclosed as the basis for any mapping, whether of ontologies or no, the
> easier its discovery and reuse by others.
>    (06)

Mike Bennett
Hypercube Ltd.
89 Worship Street
London EC2A 2BF
Tel: +44 (0) 20 7917 9522
Mob: +44 (0) 7721 420 730
Registered in England and Wales No. 2461068    (07)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J    (08)

<Prev in Thread] Current Thread [Next in Thread>