Solution Discovery!! (Re: Extension to general Semantic Neighbourhood )
Yuva-On-Net
(I am forced to create a new topic since I cannot reply on my previous
post dated 31 Dec 2004. Why??)
I read an interesting article at
<http://www.ercim.org/publication/Ercim_News/enw61/vitanyi.html> that
gives interesting answers for the question I asked : how can membership
definition be deduced from members?. I answered it that probably people
can help through a "voting procedure".
However, the authors in the article explain how normalized Google
distance or NGD can be used to deduce semantics. The conclusion they
come to is interesting...
"By calculating the NGD's, networks of semantic distances between words
can be generated, with which a computer can internalize the meaning of
these words by their semantic interrelations. In several tests, we
demonstrated that the method can distinguish between colors and
numbers, between prime numbers and composite numbers, and can
distinguish between 17th century Dutch painters. It also can understand
the distinction between electrical terms and nonelectrical terms, and
perform similar feats for religious terms, and emergency incidents
terms. Furthermore, we conducted a massive experiment in understanding
100 randomly selected WordNet categories. There, we obtained an 87.5
percent mean agreement in accuracy of classifying unknown terms with
the PhD-expert-entered semantic knowledge in the WordNet database. The
method also exhibited the ability to do a simple automatic
English-Spanish translation. "
Brilliant! Google sets must start using NGD for discovering the
semantic constructs that are implicit in sets.