Hi Adam, Thanks for the reply.
On Fri, Nov 09, 2007 at 09:48:53PM -0800, Adam Pease wrote: > Linas, > My take is that it is a fact that there are different ways of carving > up metaphysics that are not mutually compatible, but which are > individually adequate. It's precisely why SUMO and Cyc can be different > and successful. There's no need to merge them. Pick one and do useful > things. I have, and by historical accident, it was opencyc. The system is a conversational-english system, its trying to chat with humans, and learn something about them, and remember what its learned. Several issues come up. The system "learns", it adds new assertions. So, before I invested much time "teaching" the system, I simply wanted to make sure I picked the "best" ontology (currently most complete, *and* likely to dominate in the future.). > Better yet, reuse > either SUMO or Upper Cyc and contribute open source content to the > formal knowledge available to researchers and practitioners. The things my system can learn are "joe talked about baseball last week", which is not exactly of general interest. But then, it might learn a lot of details about yankees games ... which is argueably "general knowledge" but still not even a mid-level ontology. > many people vehemently assert for many years how upper ontology is > impossible while some of us quietly go about building and applying them > to real problems in science and industry. I'm not doing that. However, my concern was real: knowledge about yankees games shouldn't be critically dependent on the specific upper ontology, and that is a real problem that needs a real solution. Ben G. did in fact suggest an answer, what may well be a real, viable answer, in a seperate email. He suggests that my problems are not with the ontology so much, as the use of predicate logic for reasoning. By using a non-predicate-logic system, such as Pei Wang's NARS, or Novamente's probabalistic reasoner, one can effectively decouple the upper ontologies (space and time) from the topic at hand (yankees games). For example: in my current system, there is a checker to see if an input statement is "plausible". When "X is Y" is asserted, the system tries to figure out if its even plausible. So if some prankster claims "a ball is a glove", my system, which is pretty ignorant about balls and gloves, tries to determine if its even plausible that they refer to the same thing. Well, yes, its plausible because both are instances of #$Thing. The deduction, that balls and gloves are both "things", while of some utility, isn't really of practical use here -- the idea-ness of "thing" is too far away from the topic of baseball to be useful. I am *hoping* that moving to a non-predicate reasoning system will allow me to control the scope of the ontology applied to a situation better. I'm also hoping that this kind of "scoping" will also help the system to eventually re-learn its own upper ontologies. So, after a long conversation about Einstein, space, and time, perhaps the system could develop a new onotolgy there, without trashing eveything it knows about baseball. So maybe my question should be rephrased: has anyone been able to say anything useful about "scoping" in ontology systems? --linas ----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=64287152-ae6dd4
