On 10/6/2012 4:19 PM, Gian Piero Zarri wrote:
> "John" must be modelled a priori making use of a SPECIFIC AND UNIQUE
> instance of a concept like, e.g., "individual_person". (01)
What do you mean "a priori"? Suppose that you get the information
that John is a student. Later you discover that John is a beagle
who attended obedience training at a dog school. Why do you have
to make an immediate decision? You can just state (02)
(Ex)(student(x) & nameOf(x,'John')). (03)
The existential quantifier is sufficient to instantiate x.
If you have a sorted or typed logic, you can write (04)
(Ex:Student)nameOf(x,'John') (05)
Every type (or sort) is defined by a monadic relation. You can
choose to put it in the quantifier position, or you can put it
in relational position. Logically, there is no difference.
Computationally, it's your choice of algorithms that will
determine what is more efficient. (06)
> "Student" and "customer" are generic properties, totally independent
> in principle from John and from any other possible character. They must,
> of course, be ASSOCIATED with John within a specific, transient spatio-
> temporal environment but, being properties, cannot be INSTANTIATED
> giving rise, among other things, to (multiple) new instances of John. (07)
Whenever you have an existential quantifier, you can try to link it
to other named entities in your catalog of what exists in the context.
If you can't, you can create a new Skolem constant for the entity,
and link the known information to it. Then you can assign a local
identifier (i.e., GENSYM) to keep track of it. (08)
If you have multiple instances of entities named "John" (which is
very likely), the type label Human won't help much. When you get
new information, you can compare the entries in the catalog and put
equal signs or not-equal signs between the identifiers. (09)
There is a controversy of whether the natural types and role types
should be organized in a single hierarchy or separate hierarchies.
But there is no clear distinction for many kinds of concepts: (010)
1. In biology, many so-called species can interbreed. Dogs,
for example, evolved from wolves, but they can interbreed,
and the mixtures are fertile. (011)
2. Some species, such as the mustard-cabbage group, have been
cultivated in so many wildly different varieties -- ranging
from kale to cauliflower -- that it is reasonable to call
the different varieties natural types. (012)
3. With genetic engineering, genes can be transferred from
different species to create viable combinations that blur
the "naturalness" of natural types. (013)
4. There are types, such as Tree and Berry, which characterize
plants according to their appearance rather than the roles
they play. That makes them seem to be closer to natural
types than role types. But that classification groups them
in ways that conflict with the grouping by DNA. (014)
5. Rock formations are classified according to appearance and
their chemical composition. But the continuous range of
variation makes it hard to find natural types for rocks.
But they can have names, such as the Blarney Stone or Uluru. (015)
6. Any type hierarchy has to deal with the huge number of types
of artifacts. In a sense, all artifacts have roles, which can
be classified according to their function. But then it's hard
to say what, if anything, the natural types would be. (016)
7. Finally, there are huge numbers of "named entities" that must
be "instantiated" and classified in any ontology, but they lie
on a continuum with phenomena that are transient and almost
imperceptible. Examples include the named hurricanes,
which persist for days or weeks and are recorded in history.
Another example is the Great Red Spot in the atmosphere of
Jupiter, which has been named and observed for centuries. (017)
These problematical issues could be multiplied indefinitely.
The real world is much more complex and messy than any neat
classification. (018)
> These last would be, incidentally, a little bit strange, similar
> to instances in the style of "John is an instance of red" or
> "John is an instance of windy". (019)
As for instances of events and colors, professionals in any field
refer to and relate those things that are significant to them.
Musicians routinely refer to and talk about individual notes and
passages in a symphony. Painters (both artists and house painters)
refer to individual patches of redness, compare them, and discuss
ways of changing them. (020)
> All this is taught in one of the first lessons of whatever course
> in artificial intelligence and knowledge representation. (021)
What courses are you talking about? There are introductory courses
where the instructor teaches a particular version of some KR language
and illustrates it with toy problems that fit the logic. But students
who work on real applications quickly learn that the real world is
much more complex. (022)
Look at real applications in business, law, biology, weather
forecasting, mining, manufacturing, finance, sales... (023)
John (024)
PS: The name 'Sowa' is less ambiguous than 'John', but I get junk mail
for Polish Americans and Japanese Americans. And I once got a phone
call asking if I was from Ghana. On the island of Vanuatu, there is
a language called Sowa, which had 37 speakers about 40 years ago.
Basic principle: Names of people aren't rigid identifiers. (025)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J (026)
|