[Corpora-List] Syntactic zeros in a corpus

John F. Sowa sowa at bestweb.net
Fri Aug 25 13:14:45 UTC 2006


Mikhail,

What you are asking for is the Holy Grail of NLP.

> ellipsis like “[Where are you going?] – *To town*”

You might be able to use a special-case heuristic for
some examples.  For example, the question word "where"
expects a place, and the prepositional phrase supplies
one.  But in the general case, you would need a full
analysis of syntax, semantics, and pragmatics in order
to fill the ellipses in all possible cases.

> gapping like “I read a book, and he -- a newspaper”

Some of these cases might be slightly simpler, since you
can often solve them with just syntax and semantics
without full pragmatics.

> The problem might be relevant for any annotation scheme...

I agree.  Fully automatic annotation requires a full
natural language understanding system.  And if you have
that, you don't really need annotation.

John Sowa



More information about the Corpora mailing list