The implementation for Findings of EMNLP2023 paper "Guiding AMR Parsing with Reverse Graph Linearization". Paper Link (Arxiv).
We thank authors of AMRBART for releasing their code. Our implementation is based on their repository.
You may download the AMR corpora at LDC.
Please follow this respository to preprocess AMR graphs. For the reverse linearized data, we will release the code soon.
We release some examples for preprocessed data at AMR_reverse_graph_linearization/data. If you have the license, feel free to contact us for getting the preprocessed data.
After After configuring the path of the scripts, run
cd fine-tune
bash train_rgl.sh
cd fine-tune/evaluation
bash eval_smatch.sh
For better results, you can postprocess the predicted AMRs using the BLINK tool following SPRING.