You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This implementation is based on the cnn library for this software to function. The paper is "Shift-Reduce Constituent Parsing with Neural Lookahead Features".
Building
mkdir build
cd build
cmake .. -DEIGEN3_INCLUDE_DIR=/path/to/eigen
make
Preprocessing
You can get the constituent hierarchy by scripts/conhier_s.py for s-type constituent hierarchy and scripts/conhier_e.py for e-type constituent hierachy
./scripts/conhier_s.py [training data in bracketed format] > [s-type training data]
./scripts/conhier_e.py [training data in bracketed format] > [e-type training data]
./scripts/conhier_s.py [development data in bracketed format] > [s-type development data]
./scripts/conhier_e.py [development data in bracketed format] > [e-type development data]
./scripts/conhier_s.py [test data in bracketed format] > [s-type test data]
./scripts/conhier_e.py [test data in bracketed format] > [e-type test data]
It will automatically generate the output file test.sOUT and test.eOUT, respectively, which then can be used as extra features on the lookahead implementation of ZPar