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

Re: Directions for SUO: monolithicity and Cyc




>
>>
>>An alternative thread of activity can be as Nicola recommends, to spend time
>>relating alternatives, and work towards a library of reference 
>>ontologies, one
>>of which will be Cyc.
>
>Taking Cyc as a whole "module" is not what I suggest; rather, we 
>should try to "steal" from Cyc as much as possible, adding more 
>stuff from our own if necessary.

Er...it might be better to use a different word, even in scare 
quotes. How about 'recycle'?

Pat Hayes

PS.  BTW, in the context of the RKF effort, I spent some time last 
summer trying to recycle a workable spatial ontology out of Cyc. It 
was an oddly frustrating experience. On the one hand Cyc has a rich, 
even elaborate, collection of concepts concerned one way or another 
with spatial relationships. Many of of them, though, were like the 
'information-bearing object' concept which I mentioned a while ago, 
ie they deliberately combined spatial aspects with other aspects. For 
example, Cyc has a whole lot of notions which combine shape with 
material aspects of a body, such as being 'sheetlike', ie both very 
thin compared to length and breadth, and also being flexible enough 
to wrap around something. Other concepts combine geographical space 
with legal or political concepts, or with functional properties such 
as being useable as a human dwelling. If one is seeking a 'spatial 
ontology', these concepts are frustrating, since one has to work 
quite hard to extract the purely spatial aspects of the axioms which 
describe them, and the end result is oddly incomplete and 
unsatisfactory. But this isnt really a critique of Cyc, since it is 
*designed* to have this 'overlapping' quality: part of the Cyc style 
is to try to identify concepts which have this mixed character (and 
which would resist being put anywhere in a classification tree: the 
Cyc inheritance heirarchy is extremely, even almost pathologically, 
non-tree-like; many of these concepts have dozens of inheritance 
paths leading to them) on the belief that these are the places where 
the axioms are densest, ie where one gets the most inferential bang 
for the conceptual buck. But it sure does make it hard to stea... 
sorry, to recycle. In the end, we decided it was easier to do it 
ourselves.

---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes