Newsgroups: comp.ai
Path: cantaloupe.srv.cs.cmu.edu!rochester!cornellcs!newsstand.cit.cornell.edu!news.acsu.buffalo.edu!news.uoregon.edu!arclight.uoregon.edu!feed1.news.erols.com!news1.erols.com!news.bconnex.net!news2.insinc.net!onramp.ca!geac!itcyyz!sunil
From: sunil@itcyyz.ipsa.reuter.com (Sunil Khare)
Subject: Re: Ontological Patterns
Organization: Reuters Information Services (Canada) Ltd., Toronto
Message-ID: <Dyp75o.7Aw@itcyyz.ipsa.reuter.com>
References: <32513BDF.FE1@dkfz-heidelberg.de>
Date: Thu, 3 Oct 1996 11:41:47 GMT
Lines: 38

In article <32513BDF.FE1@dkfz-heidelberg.de>,
Johannes Link  <j.link@dkfz-heidelberg.de> wrote:
>The question I want to raise is: Are their similar tendencies/efforts in
>the field of (ontological) knowledge engineering to build up libraries
>of "ontological patterns"? 

Sorry, I don't know if there are any such libraries.

>interesting example deleted
>
>Well, the example is neither complete nor very detailed. Thus, I don't
>want to discuss its contents, but the idea of ontological patterns in
>general and - if the idea is not too silly - the form which could be
>used (or is already used somewhere?) to specify "ontological patterns".
>

I don't think the idea is silly. In my passing interest in natural language
processing, a similar idea has occurred to me. To map a sentence from surface
to d-structure, one technique is to use case-role types of mechanisms. The
definition of a verb could include a list of 'normal' roles
that the verb can accept, and where they belong in syntactic structure.

Here is the similarity: certain verbs, say verbs of motion, have common sets
of roles. Consider walk, run, swim ... in the most simple-minded way, they have
at least three roles: Agent, From-Location, To-Location. The Agent is fulfilled
by the syntactic subject, the from by a prep-phrase headed by 'from', and
the other by a 'to'-headed prep-phrase. Like you, I'm not interested in the 
particular example, but such a roleset seems like an "ontological pattern" to 
me that can be shared for these types of verbs. Perhaps, though, this is too
specialised for NLP compared to knowledge engineering 'at large'.

I also agree that object oriented notation lends itself well to knowledge
representation. Inheritance permits accumulated information to gather about
a concept at lower levels of the hierarchy. Multiple inheritance is even 
better. And over-riding inherited attributes permits 'special-cases'.

Regards,
Sunil
