Re: Semantic Layers (Was Interpretation of RDF reification)

First, the Semantic Web approach to building from the "bottom-up" using
minimalist models such as DL makes a whole lot more sense than trying to
build "top-down" from standard upper ontologies. Primarily because
people do not ever agree on standard upper ontologies, much less 
theories of pragmatics or semiotics - ever try to distinguish an
"Object" from a "Process"? How about what I meant by the utterance "Pan
Am Flight #73"? 

I think a distrust of standard upper ontologies is probably healthy,
although for certain applications they have their uses - since a lot of
work has gone into SUMO, DOLCE, etc.,  although it seems no-one will
ever agree on the one true ontology.Ditto for theories of semiotics and
pragmatics per se [1]. Now, maybe we can think of a theory of pragmatics
or semiotics for the "Web" or for "Certain well-defined standards" - but
even now that seems difficult, and there aren't too many good proposals
out there, although I'd be happy to read some - please send me links
off-list.

Therefore creating a flexible framework (be in Conceputal Graphs, KIF,
RDF, OWL-DL, Rules) that layers well with a well-defined formal
semantics is actually probably the right way to go, and I think history
has shown (re my reference to AI Winter) that this is correct. So,
perhaps the most productive way to address John's concerns would be to
figure out ways they could be layered onto the Semantic Web stack, or
precise ways which the stack should be reformulated to deal with
outstanding real-world issues that it deals with poorly or not is not
clear enough on (provenance and syntax come to mind, as other e-mails
lately have shown).

That's all - apologies to folks on the list who aren't concerned with
these rather esoteric topics. Perhaps we should start a
semantic-web-philosophy@w3.org listserv :)

                              -harry

[1] For the problems with attitude reports (modalities??), check out
situation theory. And note that their solution was more or less a
failure as well, since as soon as you define these sorts of things as
n-tuples, with any non-trival model  no-one has any idea what "n" is,
because it's *arbitrary* usually and defined by the modeller in attempt
to deal with whatever problem they are facing at the moment. In
linguistics people more or less gave up on KR and now do statistical
language processing, which has helped give us..well...Google. Which far
from perfect, probably gets more use than most KR systems unless you
count SQL as a KR system.

Azamat wrote:

> Harry,
>
> The whole matter is not thus complicated as you think. The talking
> point is that meanings have several basic aspects or dimensions or
> quantities: extensional, intensional, pragmatic and modal.  Intension
> is about a primary meaning or significance, basic definition and
> content and all the essential implications and relations involved;
> while extension relates to special, child classes or individual
> entities. Modality implies mental attitudes towards states, actions or
> changes usually indicated by lexical verbs. Pragmatics is about
> sentence utterances in the context of discourse, human or machine.
> Thus additionally to syntactic and semantic aspects, there is a
> pragmatical meaning involving an agent's intentions and communicative
> acts and understanding of communication. As John Sowa defines: 'a sign
> is an entity that indicates (represents) another entity to some agent
> (a human, animal or robot) for some purpose', in [Ontology, Metadata,
> and Semiotics]
>
> Being a special binary relation, the pragmatic (meaning) function P
> will designates all possible relationships between an agent and a
> proposition: P (a, p, t), like 'assume, believe, fear, hope, imagine,
> know, assert, doubt, mean, expect, regret, suggest, suppose, think,
> understand, want, and wish'. When making an utterance, expressing its
> thought, ideas and feelings, a cognitive agent (web software or human)
> both states some facts and performs some communicative acts. Returning
> to my example, ''Galileo knew that the earth moves around the sun''
> may be described as q = P (a, p, t). Alternatively, as a semiotic
> structure it will be written as:
> Pragmatic  function [of knowing] (a person [Galileo], a proposition
> [the earth move around
> the sun], time of utterance [ ], truth value [ ], modality [ ]).
>
> Thus there is the sentence meaning, the speaker expressive meaning (or
> hearer understanding) and modality, added with a truth value and
> temporality. Or, once we have a statement, we have to find out its all
> its meaning dimensions, intensional (content), extensional
> (reference), pragmatic, assign a truth value (not trust value) and
> establish its modality, all within one determinant ontological context.
>
> Thus the main issue is not a formal pragmatics, formal semantics or
> formal logic, but a unifying formal ontology like Standard Upper
> Ontology or Unified Framework Ontology or ONTAC making the largest
> meaning context (or the universe of discourse) as involving basic
> kinds of entities and relationships, also including pragmatic factors:
> psychological states, mental intentions, speech acts and contents,
> human actions, sociocultural environment.
> A formal theory of meaning, the foundation of semantic web, should be
> constructed as a machine-handled semiotics relied on the comprehensive
> ontological context (SUO or UFO), rather than as a formal semantics
> based on formal logic. For only the field of ontology-based semiotics
> may formally describe the relationships of signs, agents (web and
> human)  and extralinguistic entities, a necessary condition for the
> Semantic Web construction, although not sufficient.
>
>
> Azamat Abdoullaev
> http://www.eis.com.cy
>
>
> ----- Original Message ----- From: "Harry Halpin" <hhalpin@ibiblio.org>
> To: "John F. Sowa" <sowa@bestweb.net>
> Cc: <semantic-web@w3.org>; "Frank Manola" <fmanola@acm.org>; "Adrian
> Walker" <adrianw@snet.net>; "Peter F. Patel-Schneider"
> <pfps@research.bell-labs.com>; <psp@virtualTaos.net>; "ONTAC-WG
> General Discussion" <ontac-forum@colab.cim3.net>
> Sent: Wednesday, March 29, 2006 8:55 PM
> Subject: Re: Semantic Layers (Was Interpretation of RDF reification)
>
>
>>
>> The "formalization" is not even the problem with pragmatics - it's just
>> there's not even good "informal" theories of pragmatics. In linguistics,
>> there is so sort-of useful stuff in Searle's speech acts, Grosz's
>> attention-belief-intention model, and of course "relevance" theory (and
>> we all see how well similar ideas worked out with Cyc) and some rather
>> unreadable Habermas. But most of that stuff is only semi-useful to
>> computing, and always needs to be formalized - and often is formalized
>> in completely orthogonal ways by different implementations.
>>
>> So, there's lots of good pragmatic documents. Lots of RFCs, TAG work,
>> work from Apache etc. is says very straightforward and useful stuff
>> about the pragmatics of particular applications and domains and very
>> relevant to the Web. But as far as pragmatics as a distinct subject
>> matter and discipline goes, I would go as far to say that there aren't
>> even any good second-rate theories, much less first-rate theories. And
>> without an even informal theory, one can't formalize (or even vacuously
>> formalize), much less standardize in the domain independent way needed
>> by the SemWeb and other KR.
>>
>> That's a problem with KR in general, not just the SemWeb, and resolving
>> that problem lies in the hands of SemWeb application deployment, which
>> would vary from context to context. And that appears to be one of the
>> problems that led to AI winter.
>>
>> John F. Sowa wrote:
>>
>>>
>>> Harry,
>>>
>>> As much as I like logic, I admit that the most important
>>> questions of life (and of engineering, which is what we're
>>> talking about in this round of notes) have never been and
>>> probably never will be formalized.
>>>
>>> > There is, as far as I can tell, no good theories of pragmatics
>>> > that are capable of being formalized. "Pragmatics", at least
>>> > in linguistics where I come from, is usually a sort of fuzzy
>>> > "hand-waving" solution to any hard problem, much as the terms
>>> > "world-knowledge" and "common-sense" knowledge are. Whenever
>>> > I hear the word pragmatics I want to reach for my axe  :)
>>>
>>> Fuzzy hand-waving is generally bad, but it can be used to support
>>> any topic whatever.  Just because something is covered with a
>>> veneer of formalism doesn't mean it's good.  And just because
>>> some people have used a term while waving their hands doesn't
>>> mean its bad.  (By that criterion, the SemWeb would be bad.)
>>>
>>> For an example of what good common sense and an intuitive feeling
>>> for pragmatics can do, I suggest you compare the sales of Apple's
>>> iPod to anything comparable that has come from Sony.
>>>
>>> For an example of good pragmatics, I recommend John McCarthy's
>>> Elephant paper, which I believe should have been required reading
>>> for anybody working on the SemWeb:
>>>
>>>    http://www-formal.stanford.edu/jmc/elephant/elephant.html
>>>
>>> That paper was one of the inspirations for a paper I published in 2002:
>>>
>>>    http://www.jfsowa.com/pubs/arch.htm
>>>    Architectures for Intelligent Systems
>>>
>>> The Flexible Modular Framework (FMF), which is described in that paper
>>> has become the primary platform for developing and deploying everything
>>> we're doing in our VivoMind company.  Compared to that, everything
>>> I've seen from the SemWeb is legacy stuff that's trivial to deal with
>>> by importing it and converting it to usable formats.
>>>
>>> I'm perfectly happy to let the rest of the world suffer with RDF and
>>> OWL
>>> because they just kill off any competition we might encounter.
>>>
>>> John
>>>
>>>
>>
>>
>
>

Received on Wednesday, 29 March 2006 23:40:59 UTC