[Top] [All Lists]

Re: [ontolog-forum] Scheduling a Discussion [was: CL, CG, IKL and the re

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: "Amanda Vizedom" <amanda.vizedom@xxxxxxxxx>
Date: Tue, 15 Jan 2008 20:47:36 -0500
Message-id: <73158bfd0801151747s65bd267eo31b7573760525bed@xxxxxxxxxxxxxx>
Jumping in a little late here, having just caught up on my list reading...

I worked on the Cyc context representation a few years back. To be very clear, that's several affiliations ago. I don't know what has happened with it in the last few years, and I'm in no way speaking for the folks on the project now. However, I presented on it at Context '99, and while I'm not sure that my talk ever made it into distribution, the related "Dimensions of Context-Space" paper was made publicly available. If it's not still on their website and there is interest, I can contact some folks and see whether it can be dug up and provided. So, it's not quite as ultra-proprietary as someone suggested. ;-)

I can at least offer answers to some of the specific questions raised in this discussion. I'm going to use primarily present tense as it seems to me likely that this level of structure hasn't changed, but see the disclaimer above.

The ist predicate is used as the general "context" marker, as suggested. It doesn't invoke some singular context logic. In fact, one might say that it serves primarily to block some reasoning, rather than to invoke it!  A primary motivator for having contexts in Cyc was the need for inference barriers between assertions that would generated contradictions if allowed to mingle freely.  So, one primary usefulness of ist assertions (explicit, or implicit via the placement of assertions into "microtheories") is to give the inference engine an indication that ordinary reasoning should not be done across assertions in the different bundles.  Some mechanism of this type is quite essential in such a massive and varied ontology.

There is not Theory of Context in Cyc, for many of the reasons others have cited in this conversation. I don't think anyone working on the project believed that a uniform treatment would be possible in principle, never mind in practice. While the paper mentioned above named ten or so "dimensions", the number was disclaimed as arbitrary. Whether or not some way of qualifying assertions is formalized as a type of context was fundamentally a pragmatic question, resting on the need, usefulness, and feasibility of separating/bundling statements based on that qualification and/or reasoning across it.  There were continuous discussions back then, at least, about what ought to be handled as a context and what not. It was also understood even at the time of the paper's release that these aspects are not "dimensions" in the technical sense; some are independent, some are not.  Of course, there was plenty of disagreement over whether calling them "dimensions" in that case did more harm or good to ones intuitions about them, but so it goes.

Finally (I think), ist is a very general predicate that does the blocking/bundling work, but it has more specific sub-predicates that carry the weight of the logic, where such is feasible. That is, there is a hierarchy of predicates under ist. The specific ist-* predicate would indicate what, if any, reasoning could be performed across the dimension.   Temporal reasoning is a good example. When an assertion is qualified by temporal context, a more specific ist-Time predicate would be used to represent the contextualization. I can't recall the details, but actual temporal context of an assertion is stated using very specific  predicates along the lines of ist-SomeTimeDuring or ist-ContinuouslyThroughout (or something to that effect). These are specific enough to trigger appropriate reasoning modules, and a great deal of ontologist and inference engine developer effort went into both the identification of supportable context predicates and their implementation.  "Dimensions" that were identified and/or desired for specific applications can be implemented in varying degrees. The first is just the blocking-level, at which assertions that oughtn't mix are segregated from each other. Then come varying degrees of reasoning.  Many of the most intuitive "dimensions", some of which (belief contexts, competing theories, cultural contexts, for example) have been mentioned here, are without the kind of well studied and implementable candidate logics that time has. That makes the reasoning stages infeasible, so they are there as bundling-contexts only if there is benefit at that level. 

As I said, this could all be out of date with respect to Cyc; I offer it because it seems relevant to understanding the  usefulness/non-usefulness of context logics, regardless. 


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    (01)

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