Pat, (01)
That is the source of the confusion: (02)
JFS>> A static combination in "word experts" would imply that
>> domain knowledge encoded in an English word expert would
>> have to be rewritten for Russian or French. (03)
PC> That is not the way I would view the structure or purpose
> of a Word Expert. All language-independent knowledge should
> be in the ontology. (04)
The term 'word expert' was originally used by Steven Small for
an incredibly non-modular approach that died for two reasons
(1) very large time and space requirements in the computer,
(2) a very large investment in human space and time in writing
the word experts. (05)
Following is a paper by Small that describes the approach: (06)
http://acl.ldc.upenn.edu/P/P79/P79-1003.pdf
Word Expert Parsing (07)
Note the example on p. 12 for the word 'heavy'. That is a very
"heavy handed" method of representing information that most
current parsers handle in a much simpler and more general way. (08)
Following is an abstract of an article on the Parallel Expert
Parser (PEP) as a replacement for (WEP): (09)
http://portal.acm.org/citation.cfm?id=991635.991665
The Parallel Expert Parser (PEP) (010)
Abstract: In this paper we present PEP (the Parallel Expert
Parser, Devos 1987), a radically revised descendant of WEP
(the word Expert Parser, Small 1980). WEP's idea of linguistic
entities as interacting processes has been retained, but its
adherence to the word as the only entity has been rejected.
Experts exist at different levels, communicate through rigidly
defined protocols and are now fully designed to run in parallel.
A prototype of PEP is implemented in Flat Concurrent Prolog
and runs in a Logix environment. (011)
When you used the term 'topic expert', I said that was a much
better term. PEP is even better. But I would take the next
step of dropping the second letter P for parser, since I believe
that a system of parallel agents (somewhat along the lines of
Minsky's _Society of Mind_) is appropriate for modeling *all*
mental activities, not just parsing. See the following paper: (012)
http://www.jfsowa.com/pubs/arch.htm
Architectures for Intelligent Systems (013)
Suggestion: Please drop the term 'word expert' because it is
extremely misleading. It suggests Steven Small's original method,
which was an interesting idea in 1980, but not an approach that
should be followed today. The terms 'topic expert' or 'parallel
expert' are much better. But I would suggest that the best idea
would be to develop ontologies without recommending *any* method
for integrating them into an intelligent system, for one very
important reason: (014)
There is an open-ended number of promising methods today, none
of which are fully successful in doing language understanding.
So the jury is still out in which, if any, of today's methods
should be recommended for the future. (015)
I would also suggest that you qualify the following statement: (016)
> All language-independent knowledge should be in the ontology. (017)
Episodic knowledge about specific people, places, things, and
events certainly does not belong in the ontology, but it may
be significant for disambiguating sentences. And there are
many kinds of theories, hypotheses, and guesses that are not
yet established or may even have been discredited. Yet they
may be important for understanding a particular text. (018)
For such reasons, I prefer a very lightweight ontology, which
is more like a system of cross-references between language
dependent knowledge of grammar and vocabulary and language
independent knowledge (or beliefs) of any kind. The really
detailed axiomatization for deep reasoning belongs in those
domain-specific modules, which may contain a great deal of
encyclopedic information that does not belong in the ontology.
(Just look at the size of Wikipedia, for example. We certainly
don't want all of that in the ontology -- or in "word experts".) (019)
And I would recommend that the ontology be designed in a way
that it could be incorporated into any kind of system for
language processing or reasoning. We certainly do not yet
have systems that are truly intelligent, and it is essential
to keep all options (or at least as many as possible) open. (020)
John (021)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: 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 Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (022)
|