Dear John,
Thank you for your detailed and kind reply. I always appreciate
your comments and your striking analyses.
I think
however that our general "ontological" positions are extremely distant,
and that they
can be very hard to reconcile. I could define your position as a
strongly "physicalist" one. When I think about "concepts" like "gold"
or "peanut butter", I am completely indifferent to their atomic or
sub-atomic nature - by the way, opinions in this domain have varied
considerably during the centuries, but certainly not the "notions" in
themselves - or to the fact they are related to grinding peanuts or
peas. Your remarks about the "fundamental physical issues" remind me of
the opening pages in the second Chapter of your 1984 book, "Conceptual
Structures".
In this, you seemed to equate perception to the reception of "physical"
icons somewhere in the brain, which irresistibly evoked to me the
"eidola" theory of pre-Socratic philosophy (I have
been always fascinated by the "hands" that go out of the eyes
to capture those little bodies, the eidola, which represent the
perceived objects).
With respect to the "user-defined relations", on the contrary, your
are perfectly right. The paper I mentioned has been written ten years
ago, before the re-implementation of NKRL as Java stuff. Now the
HClass part of the language - which, by the way, is certainly not the
most interesting one - follows an OKBC standard approach, with
properties derived from specialization of "prototype" (in Protege,
"template") slots, which renders the above mechanisms useless.
Eventually, you defined CycL as a "crappy notation" in a recent
mail, and I agree completely. My position about Cyc has been expressed
in depth in Chapter 3 of my book (with E. Bertino and B. Catania) on
"Intelligent Database Systems", Addison-Wesley/ACM Press, 2001.
With best regards,
Gian Piero ZARRI
John F. Sowa wrote:
Dear
Gian Piero,
Continuous quantities and measures have been handled in
many systems to a greater or lesser degree of success.
But it remains true that there is no "settled" standard
for relating liters, drops, cups, and gallons of stuff
to the quantifiers in logic and to the foundations of
ontology. In fact, there is no standard for ontology,
and two of the working documents used by the SUO group
(in the cc list above) are based on the Cyc approach,
which you and I have both criticized.
GPZ> I am very surprised to see the resurgence of the
> "substance" problem, that I considered as settled
> many years ago. You fill find here two excerpts from
> a paper of mine, whose final redaction goes back to 1997.
> I hope they can be correctly read; anyway, you will find
> the whole text of the paper at the address:
http://e-msha.msh-paris.fr/Agora/Tableaux%20de%20bord/Euforbia/Background%20Information/NKRL/
> I am also very happy to notice the evolution of the Cyclists:
> if I remember well, until few years ago, they took a substance
> like "wood" as the collection of all the piece of wood you
> can find in the whole universe...
That is the issue that started this thread: John Thompson
has been using the Cyc ontology, and he had found some
difficulties with that approach.
In any case, thank you for sending those references to the list.
Your papers present some interesting representations, many of
which are compatible with the representations used in conceptual
graphs, and the readers of CG list should find them very useful.
I did, however, have some comments about the two extracts from
those papers, which you included in your note:
> "Moreover, a fundamental assumption about the organisation of
> H_CLASS concerns the differentiation between the “notions which
> can be realised (instantiated) into enumerable specimens”,
> like ‘chair’ (a ‘physical object’) and “notions which cannot
> be instantiated into specimens”, like ‘gold’ (a ‘substance’) —
> please note that a notion like ‘white gold’ is a specialisation
> of gold, not an instance. The two high-level branches of H_CLASS
> stem, therefore, from two concepts that — adopting the terminology
> used, for instance, by Guarino, Carrara and Giaretta (1994) —
> we can label as sortal_concepts and non_sortal_concepts. The
> specialisations (subsets) of the former, like chair_, city_ or
> european_city, can have direct (immediate) instances (chair_27,
> paris_), whereas the specialisations of the latter, like gold_,
> or colour_, can admit further specialisations (subsets), see
> white_gold or red_, but do not have direct instances (see,
> again, section 3.2)."
This is an interesting approach, but the idea of introducing a
distinction between sortal and nonsortal concepts is itself a
controversial issue, which creates more problems than it solves.
> "With respect to substance_, and with reference to the Lenat
> and Guha’s example of ‘peanut butter’ (Guha and Lenat 1990),
> we will have in NKRL: (i) butter_ is, like gold_ or wood_,
> a substance_; (ii) peanut_butter is a specialisation (not an
> instance) of butter_ (and, a fortiori, of substance_) and it
> is still, therefore, a nonsortal concept without immediate
> instances; (iii) piece_of_peanut_butter is not an instance
> of peanut_butter, nor a specialisation of this last concept,
> but a totally new concept deriving (through, e.g., the
> foodstuff_ subtree of physical_entity) from sortal_concepts
> — it is, in fact, possible to buy a particular instance
> (a pat of butter) of piece_of_peanut_butter (or of loaf_of
> _bread) in a supermarket.
This approach obscures some fundamental physical issues:
first, every quasi-continuous substance is actually made up
of other things, which have a granularity that is being
ignored for the purpose of the application at hand. Second,
this nonsortal stuff, such as peanut butter, is generated
by grinding sortal things, such as peanuts. That suggests
some kind of miraculous transformation whereby sortal things
disappear and nonsortal stuff is created.
One might object that a "commonsense ontology" need not be
concerned with atoms and molecules, which are scientific
entities. But that position admits defeat right at the start.
Furthermore, the commonsense terminology of stuff is applied to
rice, which every cook and consumer knows very well is made up
of individual, clearly visible grains.
Another example is the English word _pea_, which is actually
a back formation from the older word _pease_ for the stuff.
Since the individual grains of pease are larger than grains
of rice, the word was misinterpreted as a plural noun instead
of a singular word denoting stuff.
The fundamental point is that granularity is necessary for
both common sense and the most advanced sciences. It is
inherent in natural languages, and it must be recognized
in our ontologies and knowledge representations.
Once you recognize granularity as fundamental, then you can
get rid of the artificial distinction of sortal and nonsortal
concepts. All stuff is a composite made up of "atoms" or
"grains". Most grains and atoms, even in physics, are not
truly atomic, since they can be subdivided, but their parts
are of a different type from the whole.
> There is, of course, a relationship between the sortal
> concept piece_of_peanut_butter (piece_of_gold), which admits
> now direct instances, and the nonsortal concept peanut_butter
> (gold_). Some ‘intrinsic’ properties of peanut_butter (gold_),
> i.e. some slots of the 'attribute’ type (see below), like
> ColourOf and MeltingPoint, and their values, must also be
> associated with piece_of_… To this end, NKRL makes use of
> specific ‘user-defined relations’, such as GetIntrinsicProperties,
> where the inheritance semantics (i.e. the information passing
> characteristics that indicate which slots and values can be
> inherited over that relation) must be explicitly specified.
> This way of dealing with the problem of representing ‘chunk
> of matters’ allows us to, for example, avoid some of the
> intricacies of the ontology of CYC."
I agree with you about the desirability of getting rid of the
"intricacies of the ontology of Cyc". However, I am bothered
by the "user-defined relations", which must be introduced as
new primitives. If you treat stuff as a composite of grains
(which may or may not be identical, as in various kinds of
mixtures), then you can define those relations in terms of
more basic operations, which are equally applicable to both
science and common sense.
For example, the purity of gold is simply the proportion of
gold atoms to other atoms in the alloy. The operation of
making peanut butter consists of breaking down the composite
"peanut stuff" and rearranging the more basic grains and
peanut oil to form peanut butter, which can be subdivided
into lumps and spreads of various sizes. You also have the
option of talking about the proportion of oil and dry "stuff"
in the result. The theory and the practice are physically
more realistic and intuitively easier to understand.
John Sowa
========================
To post a message, send mail to cg@cs.uah.edu.
To unsubscribe, send mail to majordomo@cs.uah.edu with the
command
'unsubscribe cg' in the message body.
See http://news.gmane.org/gmane.comp.ai.conceptual-graphs
for the
mailing list archive.
See http://www.cs.uah.edu/~delugach/CG
for the Conceptual Graph Home
Page.
For help or administrative assistance, mail to owner-cg@cs.uah.edu
|