Results of Plink

Plink has been used for MUC-5 and 7. It's been used in Arabic text extraction. It's the basis of a few other systems. (It works in GATE.)
Training PTB Test PTB Susanne

Plink 99.0/94.9 70.8/74.1 61.5/66.4

Collins 88.1/88.6

Lin 71.8/79.9
The measurements are recall/precision.
The PTB is the standard Penn Tree Bank measurement, and the Susanne measurement is a dependency based analysis via Lin.
These tests really don't test ungrammaticality.
Plink's value on the tests show it is reasonable but not state of the art.
Plink is largely hand-coded.
The training set shows the potential of Plink, but some automated learning is needed to bring Plink to the state of the art.
This leads us to the semantic question. Should we use *Dog* or other symbols? Should we use Latent Semantic Analysis? How about XI? Maybe, we should actually ground symbols.