[Top] [All Lists]

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

To: "Patrick Cassidy" <pat@xxxxxxxxx>
Cc: "[ontolog-forum] " <ontolog-forum@xxxxxxxxxxxxxxxx>
From: Pat Hayes <phayes@xxxxxxx>
Date: Thu, 3 Jan 2008 02:16:16 -0800
Message-id: <p06230902c3a25e5c4606@[]>
>   Concerning your comment:
>[Pat Hayes] >> If anyone can
>>  suggest a single axiom which relates propositions or sentences to
>>  contexts, and is true both when this means the proposition is true at
>>  or in that time, and also when it means the proposition is entailed
>>  by a set of beliefs, I would be delighted to be shown it. I have made
>>  this challenge repeatedly to proponents of 'contexts' for many years,
>>  and have yet to be given an answer.
>I suspect this isn't quite what you had in mind, but it seems relevant here:
>I use for my own purposes a very simple notion of "Context" reflecting a
>similar attitude to what you express:
>>  Thus, the term "context"
>>  has no single meaning: it is used simply to refer to anything (or
>>  sometimes everything) which can influence meaning but which is
>>  outside the scope of the current theory. It is always defined
>>  negatively
>Which, as I understand it is similar to the view describe by Graeme Hirst in
>his paper "Context as a Spurious Concept":    (01)

Thanks for this reference, which I hadn't seen. 
Im glad that Graeme and I agree :-)    (02)

>But contexts, whatever they are, have to be dealt with.    (03)

Evidently you entirely miss my point. There are 
no such things "as" contexts: or, if you prefer 
but amounts to the same thing, everything is a 
potential context. So what you are saying here 
is, things have to be dealt with. Which, as far 
as I can see, is so bland as to be close to 
meaningless.    (04)

>  With a general
>relation "isTrueInContext" and reified statements (here without the "that"
>operator, for simplicity), one can then use time intervals, events,
>situations, belief systems, fictional, hypothetical, counterfactual, or any
>other situation that can affect the truth of a statement and call it a
>"context"    (05)

What is gained by calling all these by one name? 
What is gained by assuming that all of these 
diverse categories of thing should be related to 
sentences in the same uniform way? NOTHING even 
remotely suggests that either of these are true. 
Calling this diverse collection all "contexts" is 
at best redundant verbiage, and at worst is 
misleading pseudo-science. It suggests that there 
is a theory of something here to be developed. 
But there isn't: this entire discussion of 
"contexts" is a waste of time. Discussing any one 
of these ideas in a little more depth would be a 
lot more use.    (06)

>     (isTrueInContext (livesIn SherlockHolmes London) SherlockHolmesStories)    (07)

Yes, this is standard McCarthy/Guha context logic.    (08)

>Such "ist" statements can be nested to any level:
>     (isTrueInContext (isTrueInContext (livesIn SherlockHolmes London)
>Year1887) SherlockHolmesStories)    (09)

Quite. But now, we have an issue. Since, on at 
least some people's view, logical terms can 
change their meaning in a context, does the inner 
'ist' mean the same as the outer one? Or does it 
mean what 'ist' means *in the outer context*? 
(Poll of answers: Guha and CYC say ist cannot 
change, Makarios says it can change, McCarthy 
isn't sure.)    (010)

>  . . .  or one can create an aggregate context (reified here for simplicity)    (011)

Which begs a host of difficult technical 
questions about how contexts are supposed to be 
'combined' and what logical rules apply here, if 
any.    (012)

>     (isTrueInContext (livesIn SherlockHolmes London)
>What that means is that anyone is free to assert any time interval, belief
>system, etc., or combination thereof as an instance of the Type "Context".    (013)

And what follows from such an assertion?    (014)

>Cyc defines 12 independent "dimensions" of context, and I can visualize
>other dimensions as well.    (015)

This all makes me feel SO much happier. BTW, the 
Cyc dimensions are not in fact fully independent.    (016)

>The problem is that in general, what is true in one context (thus defined)
>is not necessarily true in any other context.    (017)

No, that is not the problem. That observation has 
always been used to motivate context logics: but 
in fact in almost all cases, almost everything 
that is true in one context is also true in all 
the others. Context changes typically change 
nuances of meaning rather than entire slabs of 
basic meaning. There are no realistic cases where 
large amounts of meaning change between contexts. 
McCarthy-style context logics completely miss 
this basic point, thereby causing the same kind 
of problems in context reasoning that the 
situation calculus did in planning. If you know 
anything about AI you should be thinking of the 
frame problem at this point.    (018)

>  That makes it hard to use
>*any* assertions outside a context, if it is only explicitly specified to be
>true in that context.    (019)

Or inside the context, if it is asserted outside the context.    (020)

>A lattice of theories or contexts will, of course,
>allow any given context to inherit some assertions from an included context.    (021)

"Of course"? Can you give us a little detail of 
how this works? What determines which assertions 
are inherited and which are not? Ive never seen a 
convincing scheme for inheritance between 
contexts which fails to inherit everything.    (022)

>One special thing about time intervals as contexts    (023)

No, about time intervals AS TIME INTERVALS. 
Saying "as contexts" says (literally) nothing.    (024)

>  is that, assuming that
>one is dealing with the "real world" as the outermost context, what is true
>in a particular time interval has a definable probability of being true in
>subsequent time intervals.    (025)

Isnt the same true for states of rational belief?    (026)

>So if we have:
>   (isTrueInContext (hasLocation PodunkCentroid Lat40Lon80) Year 2000),
>. . . one can specify a probability (in this case fairly high) that the
>   (isTrueInContext (hasLocation PodunkCentroid Lat40Lon80) Year 2001)
>. . . is also true (i.e., if the town doesn't expand that year).    (027)

If the town doesn't expand, there is no need to talk of probability.    (028)

>The probability in this case can be derived from the historical observations
>on how often cities (perhaps finely classified) change their geographical
>boundaries.  Buildings may also move (large ones less often).  Mountains are
>a lot less likely to move.    (029)

Enemy forces, however, can move very rapidly. 
Electrical phenomena in the atmosphere can move 
at close to the speed of light. Obviously, this 
kind of 'slow change is normal' inference is not 
a logically valid principle. (BTW, the analogous 
approach to the FP, based on 'temporal inertia', 
also didn't work very well.)    (030)

>Each assertion for each different type of relation and object will have a
>different function    (031)

That is obviously not a practical suggestion, 
right? And in any case, this is too simple. Take 
an example of a cup of coffee sitting on a table, 
and an interval of, say, 5 minutes. What is the 
probability it will be in the same place? Its 
impossible to say. It depends on what happens. If 
nobody is around, quite high. Unless of course 
there is an earthquake or a nearby explosion. 
Unless the table is in an armoured bunker...    (032)

>describing the probability that it will remain true over
>time, but some regularities will be recognizable (e.g. the density of pure
>water at a certain temperature and pressure is likely to be the same
>forever).  People move around a lot, so the location of a person can only be
>relied on within regions that expand as fast as a car or plane can move -
>though more information or probabilities will reduce the likely rate of
>Finding some general way to reason with such a notion of context when more
>than one context is of interest is a problem for which I have no idea at
>all    (033)

Join the club. But this all about the ways, and 
rates, at which various kinds of thing or 
property can change. Some are rapid, others tend 
to be slow. Fair enough, an interesting topic for 
study. But nothing, it seems to me, is gained by 
using "context" talk here. What you are talking 
about here is, in fact, what I would call an 
aspect of naive physics, in this case that of 
change. It has nothing at all to do with 
contexts, and using context language to talk 
about it just confuses and muddles the issues.    (034)

>, other than (for time contexts) to specify in the ontological definition
>of each type of entity how rapidly specific assertions may lose their truth
>value over time for that type of thing.  This sounds like a lot of work, but
>it is in fact what people learn    (035)

Its part of it, I think rather a small part. And 
it has everything to do with time and change, and 
nothing to do with contexts. You can even 
formalize it better in a non-context logic than 
you can by using 'ist' (because its not the 
*sentences* that change, but the *things*).    (036)

>  as their understanding of the world grows,
>and it enables them to make "commonsense" predictions of what will be true
>later from what is true now.  I am not at all sure that we can avoid
>representing that kind of detail, unless we are using only a specific
>application that doesn't need it.    (037)

I agree with that. In fact, given your last 
qualification, I think its a tautology. But it 
still has nothing to do with contexts.    (038)

Pat Hayes    (039)

>Patrick Cassidy
>MICRA, Inc.
>cell: 908-565-4053
>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
>    (040)

IHMC            (850)434 8903 or (650)494 3973   home
40 South Alcaniz St.    (850)202 4416   office
Pensacola                       (850)202 4440   fax
FL 32502                        (850)291 0667    cell
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes    (041)

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    (042)

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