Sean, (01)
You raised some important points in all three of your recent notes,
but in three different threads. Since those three notes touch on
related themes, I'd like to comment on all three of them together. (02)
SB> I get very uncomfortable when I see discussions of words having
> "senses" and "semantic primitives"... (03)
I don't blame you. I prefer Wittgenstein's notion of language games
and Alan Cruse's term 'microsense' for the open-ended number of
variations that are possible with each use of a word in a different
context or language game. (04)
> To use a different analogy, I would represent semantics as a
> continuum, like a power cable, and that words are the pylons holding
> up the cable... (05)
That is a good analogy. Saussure made a similar point that the
meaning of every word in each language is determined by its pattern
of relationships with all the other words. Whenever changes in the
world cause the referent of one word to change or when a new word is
coined, the entire pattern shifts to accommodate it. Furthermore, no
two languages or even two individuals have exactly the same patterns. (06)
> ... therefore choosing the terms of an ontology involves an
> a priori categorization of the forms of life that the ontology users
> need to distinguish. Interoperability between businesses is tricky,
> because they have evolved local forms of life, and therefore each
> use a vocabulary with a different set of boundaries between terms
> to those of any other business. (07)
Certainly. That is a Wittgensteinian approach that many people
who develop formal ontologies try to ignore -- largely because it
doesn't fit their paradigm. I believe that Wittgenstein's position
is compatible with logic, but in a more dynamic way than most
formal ontologies support. See the abstract and URL below. (08)
> The role of knowledge and the different levels of awareness
> (perception, comprehension, etc) is another, and I suspect that
> is where discussions of context should sit. (09)
I agree. (010)
> What we are looking for are:
>
> * Descriptions of how perception level terms are used to derive
> comprehension level terms;
>
> * Methods for inferring perception level terms from comprehension
> level terms;
>
> * Method for identifying what perception information is lost when
> decomposing a comprehension level term, so that failure to find
> a match is not taken as confirmation of absence. (011)
We certainly need better methodologies for developing and using
metadata (including ontologies). Those points are important, but
more could be added, including methods for generating the tags
automatically. The following article illustrates some of the kinds
of technology that are currently being developed: (012)
http://www.jair.org/media/2669/live-2669-4346-jair.pdf (013)
> IMHO what is needed is not just a clear statement of the theory
> so far, but a paradigm shift. (014)
I agree. Although I have always been a strong proponent of logic,
I don't believe that the traditional AI methodology (e.g., Cyc and
the current version of the SemWeb) is adequate. One reason why I
have been so critical about many of the proposals in this forum
is that many of them are just proposing to reuse the same paradigm
that even Lenat is trying to change in Cyc. (015)
> The shift I suspect is needed is from thinking about how symbols
> represent knowledge to thinking about how knowledge fans out
> from symbols. That is, how the use of a symbol invokes a set
> of knowledge processes in the recipient - these may be simple,
> as when we say "the number on the dial is six", or complex, such
> as "Amy Winehouse is the apotheosis and nadir of post-modern
> femininity". Deciding the truth of the former is a very much
> simpler process than deciding the truth of the latter. (016)
I agree. (017)
> And in both cases we are still using the same definition of symbol
> as C. S. Peirce, which John S. has referred to on more than one
> occasion. (018)
Unfortunately, Peirce's definitions are *not* the current mainstream.
The mainstream of formal ontology (including Cyc and the SemWeb)
follows the 20th century analytic philosophers, who were most strongly
influenced by Frege, Russell, Carnap, and the early Wittgenstein.
Peirce, Whitehead, and the later Wittgenstein were largely ignored
because their ideas were much closer to what you advocate above.
That approach is much harder to implement. For a discussion of
how Peirce's theories relate to post-analytic philosophy, see (019)
http://www.jfsowa.com/pubs/csp21st.pdf
Peirce's Contributions to the 21st Century (020)
There are logicians who recognize that the current mainstream
has dried up. Among them is Alan Bundy, who has been active in
research on theorem proving and problem solving for about 40 years.
But he recognizes that a major paradigm shift is essential. See
the excerpt and URL immediately after mine. (021)
And by way, Bundy and I both gave talks at an AI conference two
years ago, where we found ourselves in violent agreement about
the need to change the current paradigms in logic and ontology.
The two papers cited below cover approximately the same material
that each of us presented in our talks at that conference. (022)
John
__________________________________________________________________ (023)
Source: http://www.jfsowa.com/pubs/lgsema.pdf (024)
Language Games, A Foundation for Semantics and Ontology (025)
John F. Sowa (026)
The issues raised by Wittgenstein’s language games are fundamental to
any theory of semantics, formal or informal. Montague’s view of natural
language as a version of formal logic is at best an approximation to a
single language game or a family of closely related games. But it is
not unusual for a short phrase or sentence to introduce, comment on, or
combine aspects of multiple language games. The option of dynamically
switching from one game to another enables natural languages to adapt to
any possible subject from any perspective for any humanly conceivable
purpose. But the option of staying within one precisely defined game
enables natural languages to attain the kind of precision that is
achieved in a mathematical formalism. To support the flexibility of
natural languages and the precision of formal languages within a common
framework, this article drops the assumption of a fixed logic.
Instead, it proposes a dynamic framework of logics and ontologies that
can accommodate the shifting points of view and methods of argumentation
and negotiation that are common during discourse. Such a system is
necessary to characterize the open-ended variety of language use in
different applications at different stages of life -- everything from an
infant learning a first language to the most sophisticated adult
language in science and engineering. (027)
__________________________________________________________________ (028)
Source: http://www.inf.ed.ac.uk/publications/online/0836.pdf (029)
Representation as a Fluent: An AI Challenge for the Next Half Century (030)
Alan Bundy and Fiona McNeill (031)
Excerpt: (032)
We will argue that attempting to build a general-purpose representation
is chasing rainbows. The world is infinitely complex, so there is no
end to the qualifications, ramifications and richness of detail that
one could incorporate and that you might need to incorporate for a
particular application. This is not just a question of adding some
additional facts or rules; it may also be necessary to modify the
underlying representational language: its syntax, its representational
power or even its semantics or logic. For a narrow application, it is
often sufficient to hand-craft a representation that hits the desired
sweet spot. But this will not be sufficient for the deeper and wider
ranging applications that are the ultimate goal of artificial
intelligence, e.g., autonomous agents able to solve multiple and
evolving goals in a complex and messy environment in collaboration
with some other agents and in conflict with others. For these, more
ambitious, applications, the representation must be a fluent, i.e.,
it must evolve under machine control. This proposal goes beyond
conventional machine learning or belief revision, because these both
deal with content changes within a fixed representation. The
representation itself needs to be manipulated automatically.
We believe that automatic representation development, evolution
and repair must be a major goal of artificial intelligence research
over the next 50 years. (033)
_________________________________________________________________
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
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (034)
|