English-Hungarian Qualtiy Estimation with finetuned XLM-RoBERTa model

For further models, scripts and details, see our demo site.

Pretrained model used: XLM-RoBERTa base
Finetuned on HuQ corpus
Labels: regression model
Separator: <sep>

Limitations

max_seq_length = 256
input format: {src_en_sentence} <sep> {tgt_hu_sentence}

Results

10-fold cross validation on HuQ corpus

Model	Correlation	MAE	RMSE
baseline	0.6100	0.7459	0.9775
XLM-R	0.7948	0.6451	0.8898

Citation

If you use this model, please cite the following paper:

@article{yang-rl,
    title = {Enhancing Machine Translation with Quality Estimation and Reinforcement Learning},
    journal = {Annales Mathematicae et Informaticae},
    year = {2023},
    author = {Yang, Zijian Győző and Laki, László János},
    pages = {Accepted}
}