Next: The proposal of
Up: The problem statement
Previous: Language
Current indexing and keyword search
techniques provide only a limited performance in accessing
heterogeneous knowledge sources. A more advanced approach is to
annotate the opaque information by simple sentences and fetch the
piece of data whose annotation matches a query. This is still fairly
limited in the following sense: Consider a picture in a library of
news photographs, annotated ``A soldier holding a gun to a woman's
head.'' It is conceivable to retrieve this photograph by a query such
as: ``Show me the picture of someone holding a gun.'' But to fetch
the same picture based on queries like: ``Show me a picture of a
frightened person'', or ``a man threatening a woman,'' requires
knowing about the meaning of the symbols.
What do the symbols of our current programs mean? Think about the link between
the symbol ``fly'' and the action of flying in the real world. The only
connection seems to be the human who is using or writing the program. He is
responsible from coming up with this concept in the first place, and then
learning the assignment of this particular sign to it. In other words, the
symbols in systems such as ZOOKEEPER take their meanings in the eye of
the observer.
Cyc's goal is to provide a solution to exactly this problem. It is
assumed that if enough background knowledge is explicitly collected,
programs can use this source to answer trivial questions and make
obvious inferences about their own symbols.
Deniz Yuret
Tue Apr 1 21:26:01 EST 1997