Open NLP Problems

NLP is a popular field.
It is used for machine translation, text extraction, text mining and of course conversational systems.
I do some work on NL Parsing. The best NL Parsing systems only get around 90% of proper syntactic bracketings.
People get around 100%.
The best machines perform well below people on even this simple subset of NLP.
One problem is PP ambiguity. It needs an understanding of semantics to resolve the ambiguity.
I saw the girl with the telescope.
Machines still do not have a good understanding of semantics. (Nor do philosophers for that matter.)
NL Generation has fewer researchers working on it, but it also has its problems.