metadata
language:
- fr
- en
license: apache-2.0
datasets:
- SLPG/Biomedical_EN_FA_Corpus
metrics:
- bleu
library_name: fairseq
pipeline_tag: translation
Biomedical Domain French to English Machine Translation
French to English translation model is a Transformer model trained on in-domain web crawled corpus from Wikipedia. This model is produced during the experimentation related to building domain-adapted NMT models for specialized domains. The evaluation is done on Medline 19-20 standard development and test sets.
source group: French
target group: English
Model: Transformer
Test Set: Medline-20
Pre-processing: Moses Tokenizer
Dataset Details: (https://huggingface.co/datasets/SLPG/Biomedical_EN_FA_Corpus)
For a more in-depth exploration of our work, please refer to our paper:
Benchmarks
| testset | BLEU |
|---|---|
| Medline20 | 21.11 |
How to use model?
- This model can be accessed via git clone:
git clone https://huggingface.co/SLPG/Biomedical_MT_EN-FR - You can use Fairseq library to access the model for translations:
from fairseq.models.transformer import TransformerModel
Load the model
model = TransformerModel.from_pretrained('path/to/model')
Set the model to evaluation mode
model.eval()
Perform inference
input_text = 'Saisir du texte'
output_text = model.translate(input_text)
print(output_text)
Citation
If you use our model, kindly cite our paper:
@inproceedings{firdous-rauf-2023-biomedical,
title = "Biomedical Parallel Sentence Retrieval Using Large Language Models",
author = "Firdous, Sheema and
Rauf, Sadaf Abdul",
editor = "Koehn, Philipp and
Haddow, Barry and
Kocmi, Tom and
Monz, Christof",
booktitle = "Proceedings of the Eighth Conference on Machine Translation",
month = dec,
year = "2023",
address = "Singapore",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.wmt-1.26",
pages = "263--270",
abstract = "We have explored the effect of in domain knowledge during parallel sentence filtering from in domain corpora. Models built with sentences mined from in domain corpora without domain knowledge performed poorly, whereas model performance improved by more than 2.3 BLEU points on average with further domain centric filtering. We have used Large Language Models for selecting similar and domain aligned sentences. Our experiments show the importance of inclusion of domain knowledge in sentence selection methodologies even if the initial comparable corpora are in domain.",
}