AMR models - Sentence to Graph (STOG) trained on LDC AMR 3.0 + augmented data

Three different models for AMR graph generation for English language sentences, trained on LDC AMR 3.0 plus additional data:

- LDC+QU - entities with quotes in surface form (SMATCH F-score: 81,9)

- LDC+LC - lowercased entity surface forms (SMATCH F-score: 87,7)

- LDC+LC+QU - combination of training datasets above (SMATCH F-score: 87,7)

All models als trained using BART Large, Batch Size 16, 32 Epochs.