Skip to content

Latest commit

 

History

History
38 lines (24 loc) · 1.36 KB

README.md

File metadata and controls

38 lines (24 loc) · 1.36 KB
Raw

EncDecDRSparsing

The codes to the paper "Discourse Representation Structure Parsing" ACL 2018.

Requirements

python 2.7
pytorch 0.3.0.post4

Experiments

data

The data used in the experiments are stored in folder data, the pretrained word embeddings could be got in https://drive.google.com/open?id=1ICyISR-0PhuQYxIsqE5P7_r-OCsETIEU, and then put it into the folder data.

train and test

Currently, we do test for each epoch, because the evaluation is carried by external components

cd EncDecDRSparsing
mkdir output_dev # storing development outputs
mkdir output_tst # storing test outputs
mkdir output_model # storing models
python encdec.py

Evaluation

Tree-like structure should be converted into Discourse Representation Graph (DRG) for evaluation by drs2tuple.py. Take output_tst/1.drs for example.

python drs2tuple.py data/test.drs > data/test.tuple
python drs2tuple.py output_tst/1.drs > output_tst/1.tuple
python D-match/d-match.py -f1 data/test.tuple output_tst/1.tuple -pr -r 100 -p 10

Note that D-match is implemented in https://github.com/RikVN

Demo

The trained model can be got in https://drive.google.com/open?id=1vkhkYt3_Hmtz0x2GuynPT3PZLLu68day (GPU), https://drive.google.com/open?id=1gSkq2KDEtD5dMDYpRtA-B0Iatbc2gPrv (CPU), and then put it into the folder data.

python sent2drs.py