Add standardized smatch++ scores with ILP optimal graph alignment

goodbai-nlp / AMRBART

Code for our paper "Graph Pre-training for AMR Parsing and Generation" in ACL2022

MIT License

92 stars 28 forks source link

Add standardized smatch++ scores with ILP optimal graph alignment #28

Closed flipz357 closed 2 months ago

flipz357 commented 2 months ago

Smatch++ does standardized and optimal AMR evaluation. CF https://aclanthology.org/2023.findings-eacl.118/

Further information that I did not include in the table but may be of interest: The confidence intervals are [84.76,85.96] for AMR2 and [83.68,84.81] for AMR3.