Re: [ontolog-forum] doing standards [was - Re: Webby objects]

[William Frank <williamf.frank@xxxxxxxxx>]

On Wed, Nov 28, 2012 at 11:34 AM, Amanda Vizedom <amanda.vizedom@xxxxxxxxx> wrote:

William,

I, at least, am not talking about an ontology without any NL involved. I'm talking only about the advantages, in at least some cases, of moving (the illusion of) NL out of the unique-concept-names.

Yes, I understand now the problem of people thinking a name OR label has to accurately express the meaning they want the concept to have.  An effective way of creating this shared perception is needed.

In fact, I think that NL rich and accurate enough to be used for good generation, visualization, parsing, disambiguation, indexing, retrieval, and all those great things is missing and wanted in many ontologies. 

Abosolutely! 

 

I say semi-NL, because to make concept names unique, somewhat disambiguating to core users, and syntactically acceptable, NL-based concept names end up somewhat artificial anyway, not exactly matching NL usage.

Sure, as in

DiscreteProcess
 

I say (the illusion of) NL, because what's captured in these names isn't lexically rich enough or accurate enough to support those great ontology uses I mentioned above. 

Concept names can't really do the work of supplying this semantic assignment role. One reason is the complexity, contextuality, and dynamic nature of how linguistic reference and usage really work. Another reason is the tension between the requirements on names - unique, short and manageable in developer tools, unambiguous -- and the nature of NL expressions of a concepts. 

So, the question for me is not NL vs no NL.  It's how the NL should be incorporated into the ontology. I think that in most cases, if you have diverse users (effectively different NL communities, over language, dialect, domain, role, *or time*), the NL aspect of the ontology needs to be handled in a way that supports processing according to NL principles. It also needs to support things like localization and change (don't want to change IDs because usage changes or new users step in - change/extend the lexical assertions as well. 

In short, the argument is not for less NL, it is for more and better.

To me, this is the best summary of the problem and the solution.
 

I say this as an ontologist who is *not* a  linguist.  I've worked on non-toy applications with significant NL mapping requirements, at significant scale. In the process, I've learned a lot about lexicalization of ontology concepts (that is, really, incorporating assertions that map concepts to lexical items) and what the requirements are for doing that in a way that supports good generation, visualization, parsing, disambiguation, indexing, retrieval, and all those great things.  The attempt to load the NL into the concept *names*, rather than using mapping assertions (label, usefor, or otherwise), doesn't get you a whole lot of actual semantics and does get you extra problems.  

I would like to see a set of guidelines to effect this kind of solution.  I have tried to do so by assiduosly insisting on highly structured, fullsome, entirely English definitions of domain concepts, as a preparatory step to constructing an ontology in a  formal language. 

 Thanks,

Wm

Best, 

Amanda

On Wed, Nov 28, 2012 at 8:06 AM, William Frank <williamf.frank@xxxxxxxxx> wrote:

I do not understand how, if machine ontologies are nothing but formal relations between expressions, they have any semantic content?
If the 'ontology' is just a formal system, then it could not be used for any pariticular purpose, unitl it is *assigned* an interpretation outside itself.

For example, suppose there was an ontology with six primtive terms, p1, ,,, p6, and 6 derived terms, d1, d6. d1, for example, is defined to be an entity consisting of exactly one of p1 to p6.

What is this ontology about?  Could it not be some subset of inorganic chemistry, or equally a parts explosion of a carborator?  How do we know what it is intended for unless it is attached to experience outside the system?

For example, John Sowa talks about how chemists use NL expressions like 'carbon dioxide' in very precise ways.  But there is more to organic chemistry than the interlocking of terms; it is mostly about  making observations, doing things in labs and predicting behaviors of real things.  

So, without any NL directly involved, how can even a machine do anything useful with a machine readable ontology?

 

_________________________________________________________________
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
 

_________________________________________________________________
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    (01)