Data are available at /data folder, including:
✅ pmb-5.0.0
✅ pmb-5.0.1
✅ pmb-5.1.0
❌ pmb-4.0.0
👉 Recommendation: please use the latest version 5.1.0.
❗ In this GitHub repository, we only provide the seq2seq data. For more detailed data, please refer to PMB's official website.
Training a DRS (SBN) parser using PMB data involves utilizing gold, silver, bronze, and copper data. The statistics of pmb-5.1.0 are as follows:
| Gold-Train🥇 | Gold-Dev🥇 | Gold-Test🥇 | Silver🥈 | Bronze🥉 | Copper🥉 | |
|---|---|---|---|---|---|---|
| English🇬🇧 | 9560 | 1195 | 1195 | 146718 | 141435 | - |
| German🇩🇪 | 1256 | 936 | 936 | 6946 | - | 155974 |
| Dutch🇳🇱 | 600 | 447 | 447 | 1660 | - | 29116 |
| Italian🇮🇹 | 776 | 576 | 576 | 4336 | - | 94648 |
🤖 To change the hyperparameters, please go to src/parsing/model.py
python3 src/parsing/run.py -l en -m google/byt5-base -ip -pt data/pmb-5.1.0/seq2seq/en/train/gold_silver.sbn -t data/pmb-5.1.0/seq2seq/en/train/gold.sbn -d data/pmb-5.1.0/seq2seq/en/dev/standard.sbn -e data/pmb-5.1.0/seq2seq/en/test/standard.sbn -c data/pmb-5.1.0/seq2seq/en/test/long.sbn -s results/parsing/ -epoch 50 -lr 1e-4🐄 Xiao Zhang, Chunliu Wang, Rik van Noord, Johan Bos
📧 [email protected]
Xiao Zhang, Chunliu Wang, Rik van Noord, and Johan Bos. 2024. Gaining More Insight into Neural Semantic Parsing with Challenging Benchmarks. In Proceedings of the Fifth International Workshop on Designing Meaning Representations @ LREC-COLING 2024, pages 162–175, Torino, Italia. ELRA and ICCL.