Sean, Rob, and Dick, (01)
SB> Following up on my suggestion last week of a paradigm shift...
> ... my suggestion revolved round the question of what knowledge
> one brings (and might be expected to bring) to understanding a
> statement. (02)
I'm all in favor of Kuhnian-style paradigm shifts, but that idea
is one of the oldest in the business, it has been analyzed and
treated in many different ways, and the literature on the subject
is immense. That doesn't mean that the problem is solved, because
every partial solution opens up a vast number of new problems. (03)
SB> For a worked example, I return to "Amy Winehouse is the apotheosis
> and nadir of post-modern femininity"... (04)
Your analysis is fine, and your suggestions are quite reasonable.
But they have been part of the mainstream of AI and computational
linguistics for half a century. (05)
In 1976, Jerry Hobbs wrote "a long and unreadably detailed account
of what a computer interpreter would do with one paragraph from
_Newsweek_. The specifications of the underlying knowledge base
took 43 pages, and the account of what the interpretation procedure
would do with the text and the knowledge base ran to 58 pages." (06)
(Quoted from p. 26 of _Literature and Cognition_, by Jerry R. Hobbs,
CSLI, Stanford, CA, 1990.) (07)
A typical paragraph from a newspaper is much easier to understand
than the Amy Winehouse sentence, which would probably require a
lot more background knowledge -- including, as you pointed out,
metalevel knowledge about the knowledge of the author and intended
readers of that sentence. A complete specification would probably
take much more than 43 pages, and a step-by-step explanation of
how the computer would use that knowledge would require much more
than 58 pages. (08)
SB> Having this distinction between ontologies and natural language
> clearly expressed would help enormously in set the right expectations
> about what can be delivered by the semantic web in general, and
> ontologies in particular. (09)
I agree. Following are reviews of books by Margaret Masterman and
M. A. K. Halliday, who were two founders of the Cambridge Language
Research Unit (CLRU) in the 1950s: (010)
http://www.jfsowa.com/pubs/mmb_rev.htm
Review of _Language, Cohesion and Form_ (011)
http://www.jfsowa.com/pubs/halliday.pdf
Review of _Construing Experience through Meaning_ (012)
They made those distinctions and many others a half century ago.
But the SemWebbers haven't yet noticed. (013)
Note, by the way, Halliday uses the term 'meaning base' instead
of 'ontology'. The uppermost category of a meaning tree is
labeled Phenomenon instead of Entity or Thing. Halliday begins
with the phenomena of experience, *not* the existing things or
entities that generate those phenomena. That is a significant
paradigm shift, which, I believe, has a great deal of merit for
the purpose of understanding cognition. (014)
RA> But what seems a key issue is the boundary between the text/
> manuscript in which it appears and the greater sphere of all of
> one's knowledge. As a default, I think it is reasonable for most
> readers to expect authors to do much of the heavy lifting in
> this regard and supply as much context as possible (thereby
> making it more clear what they are really trying to say) and
> reduce the 'exercise for the reader' to a modest amount. (015)
That depends on the intended audience. An introductory textbook
on any subject will present a great deal of background knowledge
about the subject. But most of that would be repetitious or boring
to somebody who had taken a course on the subject. (016)
But I agree that there is a need for good editors, who can help
an author provide suitable background material for the intended
audience. (The WWW has made many print publishers obsolete, but
there is still a need for good editors.) (017)
RA> When the gap becomes a chasm (trying to use all of society's
> collected knowledge to interpret single sentences, the problem
> becomes too intractable (I submit). (018)
RHM> It may require a lot of knowledge, but surely not all knowledge. (019)
Unfortunately, there is no way to know, in advance, what knowledge
will be needed for any particular sentence. That is why it's always
better to have a dictionary or encyclopedia that's too big rather
than too small. (020)
If you have indexing methods, you can find any item in logarithmic
time. That means it would take just twice as long to find an item
in a petabyte (10^12) as in a megabyte (10^6). (021)
RHM> You need to identify and name the knowledge. (022)
That option is important for some applications. But more often
than not, an agent (human or computer) that needs some knowledge
does not know the name assigned to it by some agent that first
discovered or recorded the knowledge. (023)
John (024)
_________________________________________________________________
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 (025)
|