>Where would "Theorem 3.2" come from?
I'm befuddled at this point. "Theorem 3.2" seems to be a proper name for a
particular structural element, rather than a lexical relationship.
It occurred to me last night that "is a" is perhaps the worst example to
use, since it doesn't add much more lexical information than any tagged
element. For example, when you surround some text with blockquote tags,
that's implicitly an "is a" lexical relationship.
Things get interesting and more useful when the relationship isn't as
direct, as in expressing the idea that a Cadillac is a type of General
Motors automobile. Our experience also suggests that fuzzier
relationships, such as "Bill Gates is closely related to Microsoft" and
"multimedia is somewhat related to networking," are easier to build and
maintain. (Our knowledgebases quantify those relationships, by the way;
Gates and Microsoft might be a 0.9; multimedia and networking might be a
0.2.)
Nick