Extracting Atomic Sentences from AMRs

Environment

The easiest way to proceed is to create a conda environment:

conda create -n atomicamr python=3.8

Further, install PyTorch and dependencies:

conda activate atomicamr
conda install pytorch==1.7.0 torchvision cudatoolkit=10.2 -c pytorch
pip install -r requirements.txt
python -m spacy download en_core_web_sm
git clone https://github.com//ablodge/amr-utils
pip install ./amr-utils
pip install ./amrlib

Generate sentences from subgraphs extracted from AMRs:

python atomic_amr.py

Run intrinsic evaluation:

python intrinsic_evaluation.py

Install dependencies for Interface:

(cd eval_interface)
yarn install

Start Interface:

(cd eval_interface)
yarn start

Run extrinsic evaluation:

python extrinsic_evaluation.py