Ferenc, (01)
Since I did not work on Watson, all I know about it is what
I have read about UMIA, the IBM technical report from 2009,
the AI magazine article from 2010, and the various lectures
and other material on the IBM web site. (02)
What I'll say in the following is based on my interpretation
of those reports, what I know about some of the people who
worked on Watson (such as Michael McCord, whose system
produced the "deep parses" for Watson), and from some
"reading between the lines" of the published material. (03)
> 1. Has Watson integrated domain ontologies through an upper ontology? (04)
There is some discussion in the UMIA reports about a very simple
ontology, which I suspect is not much more than a hierarchy of
terms with very little axiomatization. They may have added more
for Watson, but they don't say so. (05)
The other material indicates that they used WordNet and other lexical
resources. Those things are not usually considered formal ontologies,
and they don't have any detailed axioms. (06)
> 2. If it has, would that upper ontology be shared among domain ontologies to
> restructure their own ontologies in order to have more efficient search time
>and
> and storage strategy? (07)
They may have used their hierarchy of terms to organize the search.
But IBM does have some optimized search tools, which they use in
conjunction with UMIA and their other hardware and software. They
also mention more work that was done specifically for Watson. (08)
From what I see in their reports and lectures, I gather that they
are using a variety of different processing modules that interact
by passing messages. They also mention the use of a "blackboard"
for posting messages. (09)
That idea has some similarity to a paper I published in 2002 as
an invited paper for a special issue on AI of the IBM Systems
Journal. Following is a copy: (010)
http://www.jfsowa.com/pubs/arch.htm
Architectures for Intelligent Systems (011)
None of their reports cite my paper, but since it was published
in an IBM journal, it is conceivable that somebody at IBM might
have read it. But whether they adopted my idea or re-invented
something similar, I don't know. (012)
> 3. Have they realised why high quality translating NLs is not possible with
> statistical and rule based models? Why the concept of "mechanical translation"
> was wrong from the very start? (013)
I don't see any indication that anybody on the project has thought
about theoretical questions of the feasibility of machine understanding
or high-quality machine translation. (014)
For anybody who may be interested in issues about how such systems
may or may not relate to human cognition, I recommend the following
publications by Lawrence Barsalou on topics of "grounded cognition"
and "situated simulation": (015)
http://psychology.emory.edu/cognition/barsalou/onlinepapers.html (016)
John (017)
_________________________________________________________________
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 (018)
|