Re: Roles, again
phayes@cs.uiuc.edu (Pat Hayes)
Message-id: <199509112104.QAA18937@eris.ai.uiuc.edu>
X-Sender: phayes@tubman.ai.uiuc.edu
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Date: Mon, 11 Sep 1995 16:05:33 -0600
To: "Nicola Guarino" <guarino@ladseb.pd.cnr.it>, cg@cs.umn.edu,
srkb@cs.umbc.edu
From: phayes@cs.uiuc.edu (Pat Hayes)
Subject: Re: Roles, again
Cc: "Pierdaniele Giaretta" <giaretta@ipdunidx.unipd.it>,
"Massimiliano Carrara" <carrara@cs.umbc.edu>
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk
At 5:30 PM 9/11/95 +0100, Nicola Guarino wrote:
.....
Part of my "job at hand" is to make explicit, by means of ordinary logic,
>the ontological commitments usually hidden within "ad-hoc" (task-dependent)
>logical theories in order to make them understandable (if not reusable) by
>people different from those who originally wrote them. To understand such
>commitments, something different from ordinary logic is necessary: logic can
>be (with some difficulties) a good instrument to describe reality, but
>reality itself needs to be studied as such, at least in its very general,
>"qualitative" aspects. That's why ontology (as a philosophyical discipline)
>is important, and why I find stimulating the work by those people (including
>Aristotle, Husserl, and Peirce) who have addressed that task.
If we are to study reality AS SUCH then we should turn to science, probably
physics. But even then, any discussion of reality can only be done by using
a language, and most of this debate, it seems to me, is driven more by
intuitions arising from our natural language than from anything in
'reality'. The idea of a 'role' is essentially a linguistic one.
As you say, these are philosophical issues, not AI issues. (Perhaps there
is a confusion here over 'ontology', which changed its meaning when Tom
Gruber got hold of it.) God knows, it is hard enough already to formalise
our knowledge of the everyday world, without us having to also delve into
philosophy. I really don't think that philosophers have said anything much
of use in building practical (Gruber-)ontological frameworks. (How about
the word 'grontology'?)
........
>
>In fact we must be *very* careful: there are many forms of dependence, and
>their formalization still presents various technical difficulties. One major
>problem is exactly what you have addressed: every concept is bound to a
>network of relations,
This is not a 'problem', unless you insist on looking for ghosts in the network.
but the point to be captured is that these relations
>are not all the same (as you seem to suggest): some of them are - I would
>say - "internal" to the concept, while others are "external",
There are no insides and outsides. To make an analogy, I think that you are
like someone looking at a rainbow and worrying about where the edges of
'red' are. The answer is that there arent any edges, and moreover that this
is a perfectly fine state of affairs. When you realise this, a load is
lifted from your mind.
in the sense
>that they bound together different concepts. For instance, database people
>distinguish between attributes and relations, while from the logical point
>of view both of them are just binary relations.
I dont know this literature well, but I thought that an attribute was more
like a function (which is a special case of a relation, but one worth
singling out for its logical properties.)
>Now, the idea of conceptual dependency should somehow account for this:
>internal relations (like the one between a person and his time of birth)
>don't contribute to dependency, while external relations (like the one
>between a mother and her child) do.
So what IS dependency, then? You keep insisting on a categorical
distinction which is invisible to me, and worry about the difficulty of
keeping track of it. Try pretending that it isnt there.
>.... However, the fact that database (and object-oriented) people use such
>a distinction (whithout any formalization) should suggest us that it may be
>useful...
It arises simply from a basic flaw in OO notations. If I had to give a name
for a potential relation between two things one of which hasnt been
specified yet, I might get stuck in similar knots.
>>Exercise: consider, if you can, an orange which is shiny,
>>purple, oval, used to make savory dinners, and tastes like an eggplant
.....
>The only way to recognize such a mistake is by detecting an inconsistency
>with the "meaning postulates" which characterize the predicate "Orange".
You still havnt got it. There aren't any 'meaning postulates' which
'characterize' anything. There is just the network of assertions (beliefs,
whatever). All of these problems that worry you so much come from your
looking for something that isnt there!
One is willing to put up with a certain amount of conceptual strain - a red
orange is OK, for example, and even a green orange if you know a little
about fruit - but a purple cubic orange is getting difficult (it has all
sorts of implausible consequences; you maybe could grow a cubic orange,
with difficulty, but was it painted purple?), and (say) a purple cubic
orange made of brass is just too hard to accomodate (made of brass it would
have to be an imitation orange, not a fruit, and then it can't be visually
unlike an orange. Even there I can imagine a very peculiar story which
would result in that description being just about acceptable: it involves a
collision between a truck carrying paint and a rustproof store sign for a
fruit grocery.) But I dont think that there are any hard lines, just
increasing difficulty at finding a possible interpretation.
>This is not related to the fact that "Orange" is dependent or not. Knowing
>that a predicate is dependent, on the other hand, can help for instance in
>maintaining and organization purposes (if X depends on Y then I should
>consider to delete X when deleting Y; X and Y may be allocated in contiguous
>locations, and so on...)
These are more pragmatic issues that can be addressed directly. But lets
not get lost in old philosophy looking for general answers that probably
arent there.
......
<in reply to Dobson>
.....
>Even if we don't place the role in the taxonomy, however, it's important to
>"mark" the corresponding unary predicate as a role, just to distinguish it
>from a property (like "Red"), whose corresponding unary predicate is usually
>kept out of the taxonomy, too.
>
Why? The concept of 'role' seems only to be there in order to make work. It
is important to be able to keep track of the taxonomies, because they play
a special role in the inference process; but do these other ideas?.
Pat Hayes
------------------------------------------------------------------------------
Beckman Institute (217)244 1616 office
405 North Mathews Avenue (415)855 9043 or (217)328 3947 home
Urbana, Il. 61801 (217)244 8371 fax
Phayes@cs.uiuc.edu