Transforming Multi-Conditioned Generation from Meaning Representation (RANLP 2021)

• https://arxiv.org/pdf/2101.04257.pdf

Requirements

1. Pytorch 1.2+
2. Python 3.5+
3. Huggingface Transformer
4. nltk library

Datasets & Evaluation

1. E2E Dataset
2. e2e-metrics
3. BERTScore

Train

For All dataset

python3 train.py

For sampling dataset

python3 train_sampling.py

Inference

Run inference.py to generate pred.txt with the trained model.

Evaluation

Refer to evaluation.ipynb file for e2e-metrics (BLEU, NIST, METEOR, ROUGE_L, CIDEr)

./e2e-metrics/measure_scores.py ./dataset/f_test.txt {prediction.txt}

For BERT score, edit the prediction file in the eval_BERTscore.py and

python3 eval_BERTscore.py

Citation

@inproceedings{lee-2021-transforming,
    title = "Transforming Multi-Conditioned Generation from Meaning Representation",
    author = "Lee, Joosung",
    booktitle = "Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021)",
    month = sep,
    year = "2021",
    address = "Held Online",
    publisher = "INCOMA Ltd.",
    url = "https://aclanthology.org/2021.ranlp-1.92",
    pages = "805--813"
}