RE: A plea for peace. was: RE: DAML+OIL (March 2001) released: a correction

pat hayes wrote:

>
> Well, not a gun; but you have the right idea. I might put on record
> my negative reaction to RDF since day one, long before DAML was
> proposed. Not to the goals or ambitions of the RDF effort, to which I
> am committed, but to the formalism itself. I protested to members of
> the RDF working group when RDF-1 was first released that it was
> irresponsible to put out such an inadequate model as a proposed
> standard, and was told that things would be improved in version 2.

	XML 1.0 was similarly criticized (made no attempt at a model)-- getting
people to agree on things is a tricky business sometimes, and sometimes its
best to solve the easy problems (e.g. syntax) and leave the harder problems
for later. Sometimes you need to let people try and fail before they see
your way of doing things. The goal is to cause minimal long term damage. I
don't think RDF 1.0 has seen such widespread adoption that its failure to
have an acceptable underlying semantic model will cause any real long term
harm. Think Windows 1.0 -- didn't cause MS any long term harm did it?

> Instead, RDF-1 has become adopted as a de facto standard, essentially
> by inertia and political pressure.

	Adopted by whom? It serves the needs of RSS 1.0 perfectly well (IMHO) and
otherwise isn't a bad way to represent various data formats. Perhaps its not
an adequate platform for logical inferencing but 1.0 has stood in few
people's way.

>
> There are few things more harmful to the world than bad standards
> which take root and then have to be grandfathered for eternity. DOS
> is a memorable example.

DOS wasn't created via a standards process! Inarguably it was a brilliant
idea for those who profited by it. In the grand scheme of things I imagine
its helped rather than hurt (how much more powerful is the computer you are
typing on than the most powerful VAX 11/780 your university ever owned?)


> The way to fix RDF is to admit that logical content requires the use
> of some - maybe not many, but some - nontrivial syntactic
> constructions, in particular the use of nested expressions and
> quantifier scoping; to abandon the idea that syntax is the same as,
> or best coded using, reification; and to make a committment to
> precision in specifying logical meanings (which might include being
> precise about what one is NOT saying, by the way: precision doesnt
> entail being exhaustive.)

I guess I'd argue at the phrase "fundamentally flawed" which implies (in
english) that there is no hope to be fixed. I think RDF is fixable and worth
fixing. That said, banishing the term "reification" from the RDF lexicon
would at the very least reduce mailing list traffic by a considerable
amount. I for one would greatly welcome a definition that would be
unambiguous.

>
> >Indeed I
> >could easily state that _no model anywhere_ I've ever seen is
> anything other
> >than semantically flawed.
>
> In the sense that Peter and I are referring to, this is simply false.

That's my point. We can both be logically correct by operating under
different assumptions and speaking different languages.  This entire
excercise depends on the development of a common language of discourse. If a
German and Chinese scientist get together what language are they likely to
speak in? Probably English, certainly not the first choice of either.

> There are *many* models which do not have these very basic flaws.
> Prolog does not have these flaws. Most industrial-grade data models,
> such as EPISTLE, do not have them. CLASSIC, LOOM and OIL do not have
> them. KIF does not have them. Even FORTRAN doesn't have them.

The challenge, as I see it, is arriving at something that is usable by the
web, industry and logic communities. Certainly everyone can go off into
their own corners and do the same things they have been doing for years but
that's not as challenging nor as potentially useful.

>
> >Such discussions have no end, occur frequently,
> >and have little impact.
>
> I suspect that you have not understood what Peter and I were
> talking about.

I am not arguing against the issues you raise, only against the implication
that the problems are too deep rooted to be fixed. Accept the task of fixing
these probems and don't accept failure. And realize that the problems aren't
whether triples are or aren't a good way to do foo, or whether graphs are
better than matrices. The problems are loose definitions.

>
> Perhaps I should explain the source of my exasperation. These issues
> that we are referring to are not really very deep or subtle points,
> and they are not mere debating points, and they are not philosophical
> stances on the nature of true meaning. They are technical flaws which
> are vividly and immediately obvious to anyone with a basic technical
> education in the relevant subject-areas.

Right, and simple technical flaws are the easiest to correct. What is harder
is convincing someone else to accept your preconditions (i.e. language).

>
> Part of the problem, I suspect, is that many people seem to think
> that questions broadly concerned with meaning, semantics and
> representation are somehow off-limits for technical discussions, or
> can only be approached through literary criticism. Nothing could be
> further from the truth. The RDF community seems to be imbued with a
> whiff of Derridian postmodernism, which it needs to wash off
> thoroughly before we can make progress.

Firstly, don't take my opinions as speaking for the RDF community as a
whole, and I'd never shirk from discussing such terms, but neither would I
limit my discussions to one particular worldview -- e.g. a neuroscientist
might have a totally different viewpoint on terms such as "knowledge" or
"semantics" even "language" than a logician. Or for example the work of
Norbert Weiner.

>
> >Perhaps we should simply
> >agree not to use this word, nor the term equally ambiguous term
> "knowledge"
>
> I agree that this particular word has its dangers, though in the
> context of the expression "knowledge representation" it has a
> reasonably crisp technical meaning, and the philosophical hinterland
> which surrounds it has been extensively explored.

I accept your use as a term of art, but realize that those terms are among
those that a bunch of neuroscientists might sit around and have a good
chuckle at. My own opinion is that current usage of the term "knowledge
representation" is akin to Newtonian Mechanics at the beginning of the 20th
century. Entirely self consistent.

>
> >and define our shared goals in a more practical fashion e.g. "design an
> >agent
>
> "Agent" is a real horror-word.

I selected it on purpose ... perhaps because I am being afflicted with the
term "knowledge representation" ... I mean how many people do you know who
are both knowledgable and logical? (perhaps I am asking the wrong person
:-))

>
> I'm tempted to ask about those who DONT use RDF, but I guess this
> would be impolitic; so in the interests of collegiality I will
> refrain.
>
> Just for clarification, my diatribes against RDF are not in any way
> intended to be directed at XML.

Interesting that XML 1.0 has no hint of a semantic model (so I suppose
there's nothing wrong in it from your POV :-).

>
> > >[Jim Hendler:]
> > > That said, as long as we maintain syntactic compatibility, it's not
> > > clear to me that the higher levels of the diagram alluded to above
> > > can never have their own way of doing things.
> >
> >Exactly!
>
> Well, let me temper your enthusiasm here. As one of the original
> troublemakers on this thread (bundle?) , I also entirely agree with
> what Jim says here. But notice that he refers to *syntactic*
> compatibility, while Peter and I were grousing about RDF as a
> *semantic* model. Different issues.

I wholeheartedly agree with this. Of course in the XML world we've gotten
perfectly used to plunging ahead with shared syntax alone. I would favor
recasting RDF as an abstract syntax layered on top of the XML physical
syntax. I've made comments describing RDF as a simplified XML infoset which
are to this effect. I don't think doing so would cause a whit of problems
for current RDF applications.

>
> >.... The only way
> >this is all going to work is if it works on the Web, the non Web
> experiment
> >has already been tried.
>
> Could you expand on that interesting observation?
>
Somewhat in haste...

No one argues the utility of logic, yet logic systems have not achieved
widespread usage over the last several decades.

My view of the reason for this, partly based on personal experience in the
field, is that logic systems traditionally operate on a closed world model
(and realize that I am using english terms here). The real problems aren't
in the logic, rather in the assumptions made in modeling the real world.
Those assumptions cause subtle but real problems that propagate through the
system. We need to interact with the world as we find it, not as we model
it.

The non-Web experiment, refers to the use of systems such as Prolog etc as
they have been used for the last couple of decades. They have achieved
popularity only in very limited self contained settings. I could be totally
wrong but I don't hear developers commonly talking about these tools. They
appear to be a relatively mature technology that is at a steady state in
terms of usage and impact on society as a whole. The problem _isn't_
technology that doesn't work, rather difficulty integrating this technology
into the wild.

In contrast, the Web is not a closed world. URIs are the glue that connect
real world systems together. RDF is not at all interesting because it used
triples, rather because it is based upon triples of _URIs_. Now URIs are
very far from perfect, but in this day and age they are the sole link (sic)
between an abstract syntax and a real world event.

What interests me about this new experiment is that it will not be self
contained and while that will introduce new and perhaps difficult problems
it will also make things interesting.

-Jonathan

Received on Tuesday, 3 April 2001 23:09:16 UTC