Re: Why? Re: rdf as a base for other languages

>Disclaimer:  I wrote this considerably after I had intended to leave 
>my office,
>and it is not well edited or concise.  I considered sending it to 
>Pat offline, but
>decided to mail it to the list so that it would be available for 
>anyone who cares
>to slog through it.  It is possible that I will be more articulate 
>or concise on
>Monday.

OK, Pat's responses are after he gets back from a several-beer TGIF, 
so similar sentiments apply. (But articulateness should take priority 
over concision, in any case.)

>pat hayes wrote:
>
> > > > In order to decide the truth of any statement "X" I cannot simply test
> > > > whether it is or is not a direct member of the set of 
>Statements. If it is
> > > > not a member, "X" might still be true, because it is inferred
> > >from the other
> > > > statements as you describe.
> > >
> > >This is essentially always true, unless you forbid inference and 
>close the set
> > >of sentences you're considering.  What's important is what we (minimally)
> > >require that you be able to tell/infer and what we (maximally) 
>permit you to
> > >infer and what we do about any discrepancy that (almost 
>inevitably) remains
> > >between the two.
> >
> > I don't see there being anything here to do with permissions or
> > requirements. How can one website emit such constraints on another
> > web agent's actions? There arent any guards here.
>
>This is social contract talk.  Let me try to put it into a more purely logical
>language.  (I do think, as I've said elsewhere, that part of the 
>misunderstanding
>going on here is between a logical understanding of these languages and an
>effective computation understanding of the same languages.  These are pretty
>radically different (and often entirely incompatible) views.
>
>In a more logical framework, if I say:
>
>(forall P. P(x) -> Q(x))

(I presume you mean, forall x...)

>what I mean -- what inferences are licensed, e.g., -- depends on the logic I'm
>using to interpret this statement.
>
>Since RDF doesn't come with its inference rules predefined, it's not 
>obvious which
>inferences are licensed (i.e., what logic RDF+rules is).

Hold on.  You have made a switch here from 'inferences licenced' to 
'inference rules predefined', and that is too much of a stretch. 
True, what any sentence means must depend on what logic you are using 
to interpret the statement, in a sense of "logic" where it refers to 
a well-defined syntax and a model theory. But that sense of logic 
does not come with any particular set of inference rules. Many 
alternative inference systems (sets of inference rules) may all be 
valid with respect to a given logic. A logic is not, in the sense I 
am using the word, to be identified with a particular set of 
inference rules, and still les with any particular kinds of 
*behavior*. There are no infernece rules yet defined for DAML, for 
example, but it has a reasonably precise langauge and a well-defined 
model theory. There could be a variety of different rule sets for 
DAML, all equally valid; or there could be none at all, and some 
completely different mechanism (such as a consistency checker which 
worked by constructing arithmetic models) could be devised. The logic 
itself - the syntax + model theory - does not determine the proof 
theory.
But in any case, seems to me that RDf does come with a syntax and a 
kind of semantics, so it is a logic, more or less.

>Now we *have* to switch into the effective computation world:

I don't see why.

>It is likely that
>different users of RDF will in fact choose to calculate different sets of
>consequences.

Fine, as long as they are all valid consequences according to the 
semantic specification. The semantic specification - the logic - 
doesn't say anything about that, nor does it need to.

>It is also likely that some minimal (perhaps empty) set of
>consequences will be deemed to be "required" in order to be compliant with the
>RDF+rules language and some larger set will be "consistent".  That 
>is, RDF+rules
>will be a family of logics with at least one middle category.  Think 
>of it as a
>modal system with a gap between necessary and possible.

No, I will not think of it that way!  People are far too free with 
this idea of modality. Modal logics are a very special class of 
logics with some very odd properties. If you really want to say that 
RDF or DAML are modal languages, then let us get that claim out into 
the open and debate it. I would be inclined to oppose it, on a 
variety of technical grounds, but I admit that many people in the 
ontology business do defend the use of modal notions.

> > > > So why say that statements are true just because they are 
>members of this
> > > > set? ...
> > >
> > > > To me, stating that each statement in Statements is a fact 
>does not gain me
> > > > anything (in ease of inferencing) and costs me alot (of 
>cumbersomeness and
> > > > confusion).
> > >
> > >Well, not "are true" and not "is a fact", but "is asserted by the
> > >document", at
> > >least, which may be the same as far as your concern goes.
> >
> > No, no. That is what "assert" MEANS.
>
>No, it's *not*.  Not in a speaker-relative context.  Not in a logic 
>of beliefs.
>Not in almost any modal context.  Assert here is a modality.  Sheesh, Pat, you
>know better than this.

Yes, I do know better than this. Assertion is NOT a modality. ("P 
asserts that..." might be a modality, but that is when one puts 
'asserts' into the formal language itself (and only when you use 
indirect quotation: if you are reifying my actual assertion, it isn't 
modal.). I was talking about the use of the language, not about its 
semantics. See my previous message for the distinction.) Think about 
it: if assertion is a modality, how is assertion done in a non-modal 
language? For example, most of mathematics consists of assertions, 
almost all of them in non-modal language. "The sun is shining" 
doesn't involve any epistemic modalities: it just talks about sun and 
shining, not about asserting.

Of course logics of belief are modal, but I wasnt aware that RDF was 
supposed to be a logic of belief. This particular exchange is the 
first time I have ever seen it even suggested that RDF is intended to 
have a modal interpretation of any kind. If that is what you are 
claiming, then please say it clearly and defend it. (What piece of 
RDF syntax indicates the modality? Does it have a Kripke semantics? 
What is the world-alternativeness relation? When the implicit 
existential quantifications project into a modal context, should they 
be interpreted de dicto or de re? And so on.)

>
> > Look, if I assert (note: *assert*) the following sentence: "The sun
> > is shining.", then while of course you are free to believe or
> > disbelieve me, or indeed to draw all sorts of conclusions about me,
> > it remains the case that what I said was that the sun was shining.
> > That is not a sentence, but a claim about the way the world is: it is
> > such as to make the sentence (that I asserted) true. You may observe
> > me and correctly say "Pat asserted: "The sun was shining" " , much in
> > the way that you might observe a frog and correctly say: "The frog
> > said: "Graaak" ", but that observation does not constitute an
> > exchange of content, and what you have asserted is not what I said.
>
>Absolutely.  But in RDF, if you said it, YOU asserted it.  Assert_PAT (Shining
>sun).  (Use the appropriately indexed modal.)

But that 'assert' isn't what the RDF says. (Is it??)  RDF (DAML, OIL, 
etc.) isn't ABOUT the business of information exchange between 
agents: it is the MEDIUM that the agents use to communicate 
information to one another.  (A fully-fledged logic of agent 
interchange would be a much more ambitious undertaking. The kind of 
stuff that Bradshaw is doing, for example, seems more in that line; 
it is all concerned with logics of permission and obligation, 
reasonings about policies, how to ensure that behaviors align with 
restrictions, that kind of thing. It seems completely different from 
this 'semantic web' world.)

Or so I was under the impression, at any rate. If I am in the wroing 
here, please as many people as possible tell me so (offline if 
necessary) since I will need to re-think the entire RDF enterprise.

> It's hard to imagine how a logic
>based on RDF as currently used and spec'd could treat this as 
>anything other than
>a requirement for the following interpretation: In the possible worlds
>corresponding to your perspectives,  (Shining sun) is true.

Wow. You amaze me. I have never even HEARD of this interpretation of 
RDF before. Where did you get it from? (Where in the entire 
literature on RDF is there any mention of 'perspectives'?)

> >
> > If we are having a conversation under normal rules of mutual
> > cooperative communication, then a basic act of communication has
> > taken place not when you make an observation about my utterance, but
> > when you interpret the *content* of that utterance and accept it, ie
> > when my asserting "the sun is shining" leads you to believe that the
> > sun is indeed shining. I presume that the basic idea of RDF (and DAML
> > and OIL, etc.) is that they are primarily intended to be used to
> > convey content in roughly this way, though of course greatly
> > simplified since it occurs between mechanical rather than human
> > agents. The wide world being what it is, no doubt issues of trust and
> > cynicism may well arise, but these surely are intended to be
> > concerned with the *content* of what the RDF (etc.) is encoding, not
> > about its *form*. If the primary goal of all this effort is just to
> > enable one agent to send some symbolic shapes to another to enable
> > the recipient to assert something of the form "A sent me the
> > following string:"kdjglafldgfla" " , then there seems to be little
> > point in the entire enterprise.
>
>Yes.  Agreed.
>
> > > Certainly the
> > >benefit (which is nonetheless present for some and in some 
>applications) comes
> > >at a cost (for those who buy into inference as a given).
> > >
> > >I am under the impression that assertion by the document of statements
> > >contained within the document is a fundamental tenet of RDF and there's no
> > >changing it without dropping RDF. Assertion is *not* the same as 
>truth of all
> > >statements or documents,
> >
> > It is the claim that they are true, however. The two are closely linked.
>
>Again, BY an agent.  You can't just drop the agents out.  Logics of 
>belief.  We've
>been through this all already, at least a decade ago.

OK, you and I are in different worlds. I have never thought of any of 
this semantic web activity as being in any way concerned with logics 
of belief. As for dropping the agents out, it seems to me that they 
are already dropped out of RDF. They might be involved in the RDF 
metatheory, but not in the language itself (except possibly in the 
URI naming conventions, but even then only in a rather weak, 
non-modal, sense.)

If RDF is really modal, then the problems just seem to get worse and 
worse. Now it has to be able to represent not just negation, 
disjunction and quantification, but epistemic modalities as well, all 
using simple ground triples. (What next: fuzzy truth values?) No 
wonder it is in such a confused state. For what its worth, Bradshaw 
has been trying to use DAML (not RDF) for agent policy reasoning, and 
the demands of that domain stretch the language close to its limits 
of expressiveness. RDF alone wouldnt have a snowball's chance.

Pat

---------------------------------------------------------------------
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

Received on Friday, 1 June 2001 21:23:21 UTC