ontolog-forum
[Top] [All Lists]

Re: [ontolog-forum] to concept or not to concept, is this a question?

To: Pat Hayes <phayes@xxxxxxx>
Cc: ontolog-forum@xxxxxxxxxxxxxxxx
From: Adam Pease <adampease@xxxxxxxxxxxxx>
Date: Fri, 15 Jun 2007 14:10:38 -0700
Message-id: <4673004E.7090009@xxxxxxxxxxxxx>
Hi Pat,    (01)

Pat Hayes wrote:
>> Pat,
>>
>>   This
>>  > seems to me to illustrate a wise observation by Doug Lenat: you might
>>  > have to have an upper ontology, but which one you have doesn't really
>>  > matter a damn, because there's very little useful to say at the upper
>>  > levels, and whatever you really want to say at the middle levels,
>>  > where all the actual content is, can be made to fit with just about
>>  > any upper level you like.
>>
>> This is oft quoted, but I think has more to do with marketing than
>> science.  In fact, we showed in 1999 on the HPKB project that about 1/3
>> of the terms in proofs and answers on a large test were from the Cyc
>> upper ontology.
> 
> Yes, but I wonder what this means exactly. You used metrics based simply 
> on a term occurring in the answer or proof. If an ontology makes an 
> upper-level distinction between, say, entities and qualities (I'm making 
> this up) and then uses its own terminology of 'entity' and 'quality' in 
> its middle-level axioms, then these terms will of course occur in many 
> proofs. This however does not establish that there is any actual need 
> for this distinction: the occurrences in the proofs may arise simply 
> from the middle levels of the ontology being carefully constructed to 
> maintain the distinctions it itself insists upon making in its upper 
> levels.
> 
> BTW, this is a very valuable study and I don't mean to denigrate it. And 
> I entirely concur with your conclusions there about the need for more of 
> this kind of work to be done.
>     (02)

This is a valid point.  We could go back to the data to try to remove 
mention of classes like TangibleThing, that don't add a lot of 
inferential value by themselves.  For what it's worth though, my sense 
was there were not a lot of cases like that.  I'd also say that in my 
current work terms like Entity and Physical rarely appear in practical 
inferences with SUMO.  And of course even if they did, if hundreds of 
terms from the upper level appear, the top dozen terms are such a small 
percentage, that the overall assessment of 1/3 still holds.    (03)

>>  I would agree that several reasonable upper ontologies
>> could be constructed, at least three have, and possibly any given middle
>> level content could be reformulated to use a different upper level, but
>> that doesn't mean that the upper content isn't immediately useful in any
>> practical inference.  Try doing without saying that one event happens
>> before another, for example.  That's upper level content
> 
> Then we are at cross purposes. I'd call that middle level; in fact, its 
> part of a temporal ontology describing relations between time-intervals.     (04)

That seems like an odd way to characterize it.  Cyc's BaseKB, SUMO (just 
the top 1000 terms) and DOLCE all include temporal relations.  But, if 
you want to call that mid-level, that's ultimately ok, now that I 
understand what you mean.    (05)

> Yes, that is about real things (events) and says something nontrivial 
> about them. But go up from 'event'. If something tells you that events 
> are occurrents, what use is that? A good rule of thumb is, if the 
> terminology being used to make the distinctions is arcane or obscure, 
> then the distinction is likely to be worth making only for formal 
> reasons, to satisfy a philosophical prejudice or in order to provide a 
> scaffolding for a more useful distinction. In most large-scale 
> ontologies, the top three or four levels of the classification seem to 
> have this quality.    (06)

I agree.  That's a very tiny part of Cyc or SUMO.  And when Cycorp talks 
about the upper level, surely they are talking about what used to be 
released as "Upper Cyc", which was the top few thousand terms.    (07)

So, I guess we're in agreement, the top few levels (I'd say maybe three, 
which in SUMO would mean only 13 concepts, shown below) don't make much 
difference, and the proof of that is how rarely they appear in practical 
inferences.  But the next 10 levels or so, which in Cyc and SUMO would 
still be termed part of the upper ontology, do make a big difference.    (08)

Adam    (09)

  Entity
         Abstract
             Attribute
             Graph
             GraphElement
             Proposition
             Quantity
             Relation
             SetOrClass
         Physical
             ContentBearingPhysical
             Object
             Process    (010)

> 
> Pat
> 
>> bound to be
>> present in any number of practical inferences.
>>
>> Cohen, P., Chaudhri, V., Pease A., and Schrag, R. (1999), Does Prior
>> Knowledge Facilitate the Development of Knowledge Based Systems, In
>> Proceedings of the Sixteenth National Conference on Artificial
>> Intelligence (AAAI-1999). Menlo Park, Calif.: AAAI Press.
>> http://home.earthlink.net/~adampease/professional/cohen-aaai99.ps
>>
>> Adam
>>
>> Pat Hayes wrote:
>>>>  Weighing in with Ingvar...
>>>>
>>>>  If I want to describe an organization*, is it the people, the 
>>>> facilities
>>>>  or the notion of an entity that performs some function in accordance
>>>>  with internal guidance* and external laws*? The ideas in the guidance
>>>>  and laws may be reduced to energy on phosphor or symbols in ink on 
>>>> paper
>>>>  but it's the ideas that matter. How do I describe the idea of a 
>>>> design*
>>>>  produced by the organization* that satisfies a customers mission
>>>>  statement* as they compete in a economic marketspace* to reach their
>>>>  annual goals*
>>>>
>>>>   * = subclass of Concept
>>>
>>>  Well now, I understand all of the above until that last claim. This
>>>  seems to me to illustrate a wise observation by Doug Lenat: you might
>>>  have to have an upper ontology, but which one you have doesn't really
>>>  matter a damn, because there's very little useful to say at the upper
>>>  levels, and whatever you really want to say at the middle levels,
>>  > where all the actual content is, can be made to fit with just about
>>>  any upper level you like. I know a fair amount about organizations,
>>>  designs, mission statements and annual goals, and enough about
>>>  economic marketplaces and internal guidance to follow what others are
>>>  saying, but I'm damned if I know anything worth writing down about
>>>  Concepts. And indeed, if I were asked to come up with a name for a
>>>  superclass of all those *'s, my reaction would be that they have
>>>  nothing whatever in common. I fail to see how an organization can
>>>  possibly be said to be a concept, in fact, or for that matter a
>>>  mission statement (I have actually held mission statements in my
>>>  hands at various times). Of course we can speak of a concept of an
>>>  organization, but that's not the same as the organization itself. Can
>>>  a concept have legal rights? Some organizations do.
>>>
>>>  Pat Hayes
>>
>> _________________________________________________________________
>> Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/ 
>> Subscribe/Config: 
>> 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 Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx
>>
> 
>     (011)

_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Subscribe/Config: 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 Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (012)

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