W3C home > Mailing lists > Public > www-rdf-interest@w3.org > December 2002

Re: context (comments on http://www.w3.org/TR/2002/WD-rdf-schema-20021112/)

From: pat hayes <phayes@ai.uwf.edu>
Date: Thu, 5 Dec 2002 16:51:55 -0600
Message-Id: <p05111b07ba157f57b651@[10.0.100.86]>
To: "Richard H. McCullough" <rhm@cdepot.net>
Cc: "RDF-Interest" <www-rdf-interest@w3.org>, "Richard S. Latimer" <latimer1@att.net>
>Note: Per the request of Brian McBride, this ongoing discussion of 
>context has been moved from rdf-comments (concerning current RDF 
>documents) to rdf-interest (concerning future RDF concepts).
>
>My comments are interspersed below, prefixed with +++++.
>============
>Dick McCullough

snip.

>>>I have consistently used the same definition of context 
>>>(<http://rhm.cdepot.net/doc/KEtutorial.txt>http://rhm.cdepot.net/doc/KEtutorial.txt)
>>>
>>     space = where action occurs
>>
>>+  time = when action occurs
>>
>>+  view = proposition list which captures prior knowledge
>>
>>
>>Well, OK, but that is a very odd collection of disparate kinds of 
>>thing,  seems to me, which have very little in common (and very 
>>little in common with the other uses of that word in other, er, 
>>contexts). What is the connection between actions and propositions, 
>>for example?
>>
>>##### Here's a proposition (in KR)
>>
>>#####    at space=s, time=t, view=v  { Joe do hit od the ball done }
>>
>>##### This proposition characterizes an action, "hit".  It 
>>specifies the subject, "Joe", the object, "the ball", and the 
>>context, s/t/v.
>>
>>##### The meaning of this proposition, the things it denotes in 
>>reality, is clearly dependent upon the context.  For example: 
>>compare space=the local sandlot and space=a major league baseball 
>>stadium; compare time=4 December 2002 and time=15 May 1941; compare 
>>view={Joe is Joe Doe} and view={Joe is Joe DiMaggio}.  (Of course, 
>>view should be a name, and might include many other propositions in 
>>addition to the identity/alias given here.)
>>
>
>Richard, I think you are confused between propositions and worlds. 
>What you are calling the context is just the choice of one 
>interpretation over another.
>+++++ I'm not confused.  I'm sorry that my comparison of contexts 
>misled you into thinking about "possible worlds".  In the 
>conventional sense, there are no "possible  worlds" and "possible 
>interpretations" in the KR language.

Then I do not think it can be called a formal language in any useful sense.

>  The is only one world -- the world we all live in -- and only one 
>interpretation -- the facts of reality in that one world.

Of course there IS only one world, but semantics has to be concerned 
not just with the way the world actually is, but with the ways that 
it COULD be given what we are able to say about it. It is impossible 
ever to say enough to pin down the actual world in full detail; and 
since a reasoner has no access to the actual world other than through 
sentences of the formal language, it has to as it were survey the 
possible ways that the world might be, given what is said about it. 
(You might find Tarski and Quine good authors to read on this topic, 
they both have the issues very clear. I speak of course as a disciple 
of theirs ;-))

If there is only one interpretation of KR, then all truths (ie 
sentences that are true in the actual world) must be logically true 
in KR; which means in turn that a complete KR reasoner must be 
omniscient. How do you implement such a thing?

>  "Possible worlds" only exist in the context of a person's mind, i.e.,
>     at view = Pat thinks { ... statements about possible worlds ... }

Whether or not possible worlds actually exist is beside the point. 
Think of the phrase as meaning 'possible way that the actual world 
might be'. Model theory is just a way of using this idea to formally 
define notions of consistency, contradiction and entailment between 
sentences.

>Look, the sentence you give here asserts that something - lets say 
>somebody - called Joe hit something called the ball. That is 
>obviously a highly incomplete assertion in that it does not fully 
>specify an entire world. You can fill it out with all kinds of other 
>information about Joe and the ball and where this happened and when 
>and whether or not the sun was shining and the phases of the moon at 
>the time, and whether or not Joe was hungry, and so on and on. There 
>is literally an infinite amount of information you could add to 
>this. All true, and all irrelevant. The fact remains that the 
>sentence itself asserts a proposition which might be true or false 
>in any particular interpretation. The truth of falsity of *other* 
>propositions is irrelevant to that. They are not somehow 'part' of 
>this proposition's 'context', somehow missing and needing to be 
>restored; they are just other propositions that one can put into a 
>representation or not, as one chooses. I can know that Joe hit the 
>ball without knowing anything else about Joe, and that might be all 
>I need or want to know about him.
>
>Now, it does make sense, I concede, to bundle some of these into 
>something one might want to call the 'context' of an action. But 
>your particular choice of where and when is only the tip of the 
>iceberg here. I know you can ask these questions about the 'context' 
>of an action, since actions always happen at a time and place (well, 
>strictly, that isnt true, but its a handy fiction for most purposes) 
>but you can ask others as well: actions take place in a manner, are 
>done by an agent, to an object, some of them in a direction, some of 
>them for a purpose, and so on. The study of these aspects of action 
>verbs is an entire branch of linguistics.
>+++++ KR expresses these other aspects of actions in the action statement
>     subject do action
>         with action characteristics    # e.g. purpose
>         out  products                          # output
>         od   direct objects                 # input
>         from initial characteristics    # e.g. location
>         to      final   characteristics    # e.g. location
>     done

OK, how would you handle examples like these:

Joe suddenly turned to face the Count.
As Susan went into the kitchen, Bill smiled enigmatically at Josephine.
It was clear that someone had been in the room recently.

>
>So why the two that you have chosen, particularly?
>+++++ I chose space & time because actions (e.g. walk) are 
>space-time-dependent & attributes (e.g. sex) are not.

Well, OK, but so what? That is, why the emphasis on space and time 
particularly? In a web context, time is important to be sure, but 
space seems largely irrelevant and in many cases hard to even define.

>+++++ (Don't bug me about sex-change operations.  KR can handle that.)
>+++++ Note that the same space-time context may apply to many 
>different statements, with the appropriate measurement.  For 
>example, space=New York City, time=1900.
>And I still think that your 'view' doesn't belong with the other 
>two, but is a separate kind of entity. For example, the view might 
>itself have information about the time and place.
>+++++ In general, view will have space-time information.  View names 
>a proposition list, and the propositions may include space-time 
>context.

Well then why do you single out those as separate aspects? Why not 
just have a 'view' that incorporates all the aspects of the action?

What do you mean by a proposition, here, BTW?

>
>>And in the first two cases, where are the spatiotemporal boundaries 
>>drawn? Take this email conversation that we are having and other 
>>people are maybe reading: what is the space where that action is 
>>occurring?
>>
>>##### at space=Pioneer California { I do write od this email done }
>>
>>##### at space=your office in Florida { you do read od this email done }
>>
>>##### If you want to consider the complex action/event of all the 
>>people reading this email, space=union of all their locations.
>>
>
>Unioning won't work for distributed things like email conversations. 
>That's why I asked the question.
>++++ I don't get your point.  I could have said space=Earth, but 
>that is overgeneralizing.

I meant that the union of the spaces will be disconnected; but a 
communication event is only physically possible across a connected 
piece of space+time. So the emailing space has to somehow involve the 
path taken by the various communication packets.

>Here's a nice example: four people sitting round a table, couples 
>opposite, and each couple is holding a conversation across the table 
>while ignoring the others (sound familiar?) . What is the spatial 
>envelope of each conversation?
>+++++ space=the table.  BTW, I don't know what you mean by "envelope".

Technical term: smallest enclosing space.

>
>>
>>>Depending of the context of the discussion, I sometimes emphasized 
>>>one of space/time/view,
>>>
>but my definition has not changed.
>
>
>
>
>I am fully aware that others do not agree on a definition of 
>context.  You and I attended the same
>
>Context Symposium at MIT in 1997.
>
>
>I recall. Things havn't gotten any better, you know.
>
>
>But more seriously, we can't be expected to use *your* definition of 
>context in a language intended for general Web use all over the 
>planet.
>
>##### I don't see why not.  It's based on sound principles.  It works.
>
>
>Im not convinced of either claim. Nothing personal, but this a big 
>topic and I don't trust any simple answers.
>+++++ Re principles, I have identified the essential characteristics 
>of context.

I don't agree that you have. Certainly the community as a whole is 
not, I believe, convinced that you have. You need to write a book 
about your ideas, or a few papers, at least.

>+++++ Re working, I have used KR/context very successfully for 6 
>years, in many different domains.

Used it for what?

>But in any case, there is a political aspect to this: even if you 
>were right, how would we get the rest of the world to agree?
>+++++ A serious problem to be sure.  We take it one step at a time, 
>beginning right here with W3C.

I was speaking rhetorically.

>
>>If we try to use a 'general' notion of context we will dissolve into smoke.
>>
>>##### You're being too pessimistic.
>>
>
>I don't think Im pessimistic at all. I rejoice in the elimination of context.
>+++++ Context can't be eliminated.  As you say below, we can make it explicit.

Then it isn't context.

>
>>And in any case, the purely functional requirements of distributed 
>>knowledge-processing architecture require that we reduce such 
>>contextuality as much as possible. Take your second example, where 
>>a sentence's truth depends on the 'context' of the document in 
>>which it occurs,
>>
>>##### There are two parts to the statement from your document.
>>
>>##### 1. Names denote things in the universe.
>>
>>##### 2. Sets of triples denote truth-values.
>>
>>##### We both agree that 1. is true.  Statement 2. is true in your document
>>
>
>No, statement 2 is true *of RDF*, everywhere, because my document is 
>part of the RDF spec. Its the authority of the W3C, backed up maybe 
>by the social assumptions made by the larger culture about what a 
>'standard' is, that makes it true, ultimately. Truth is not attached 
>or limited to documents, or even defined by documents.
>+++++ Truth is partly defined by documents, in the sense that the 
>documents give definitions of terms which are used to make 
>statements about reality.

I disagree. Truth is a relationship between documents and possible 
ways the world might be; or, more compactly, between documents and 
worlds. Documents can give definitions but they can also do many 
other things.

>
>>  because you define the denotation of sets of triples to be 
>>truth-values.  Statement 2. is false in my document because I 
>>define the denotation of sets of triples to be things in the 
>>universe (facts of reality).
>>
>
>If your document makes a claim about RDF, then your document is just 
>flat wrong. If on the other hand it makes a claim about some other 
>language, then the two documents are talking about different things. 
>Either way, there is no context involved.
>+++++ You have just identified a key component of context: RDF.

It seems like for you, anything whatever can be considered to be a 
component of a context. Do you consider languages to be part of the 
context?

>  In fact, my document makes no statements about RDF.
>My document talks about the KR language (and the English language). 
>I tried to bridge the gap between the RDF document and the KR 
>document by associating RDF "sets of triples" with KR "proposition 
>list" with "facts of reality".

I think that association might be mistaken, which may be giving us so 
much trouble. Certainly one should not identify RDF triples with 
'facts of reality'.

>
>>and ask yourself what happens when parts of such documents are 
>>being distributed across optical fiber, processed, inferences drawn 
>>from them, conclusions archived and then re-transmitted arbitrarily 
>>long times later, and used in other contexts far from their 
>>original source. If meaning depends on contexts which can vanish in 
>>microseconds, as they will in the SW, then meaning becomes 
>>meaningless.
>>
>>##### Meaning needs to be pinned down with good definitions.
>>
>
>Quite; and good definitions will pin down meanings in a 
>noncontextual way, as far as possible.
>+++++ Definitions, like all other statements, depend on context.

I disagree. Certainly for machine inference purposes they had better 
not, unless the machine has access to some representation of the 
context.

>Archived weather records don't say 'its raining', relying on the 
>context to indicate 'here, now'; they say explicitly where and when 
>it was raining. The added information makes the assertion simply 
>true or false, by encoding the 'context' explicitly in the record. 
>That's the way to make things clear and unambiguous, by getting rid 
>of the context as far as possible. This isn't rocket science, by the 
>way, it has been being done for centuries, probably for millenia. 
>Its likely that this is what writing was originally invented for, in 
>fact.
>+++++ You are not getting rid of the context, you are making it explicit.

Then it is no longer contextual. If I say 'it is raining' at 3 pm on 
5 Dec 2002, then the time of my utterance is part of its context. If 
I say 'it is raining at 3pm on 5 Dec 2002', then my utterance needs 
no temporal context, and in fact it would be meaningless to ask what 
its temporal context was, ie when it was true. As far as time is 
concerned, it has been removed from the context space: its just 
simply true or false. Making information explicit *removes* it from 
the context.

Pat
-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola              			(850)202 4440   fax
FL 32501           				(850)291 0667    cell
phayes@ai.uwf.edu	          http://www.coginst.uwf.edu/~phayes
s.pam@ai.uwf.edu   for spam
Received on Thursday, 5 December 2002 17:52:02 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:51:57 GMT