deep-simplification

Text simplification using RNNs. Our deep-simplification model achieves a BLEU score of 61.14

Training

Navigate to the simplify directory and run

THEANO_FLAGS=device=gpu,floatX=float32 python train_nmt.py 

Sampling

Navigate to the simplify directory and run

THEANO_FLAGS=device=cpu,floatX=float32 python translate.py -n -p 1 model_hal.npz \
    ../data/dictionaries/simple.wiki.tok.pkl \
    ../data/dictionaries/simple.wiki.tok.pkl \
    ../data/testing/testing.normal.aligned.tok \
    ../data/testing/testing.simple.aligned.out.tok

Testing

Navigate to the simplify directory and run

perl multi-bleu.perl ../data/testing/testing.simple.tok < \
    ../data/testing/testing.simple.aligned.out.tok