Data Selection Curriculum for Neural Machine Translation
This work addresses training efficiency and quality for machine translation practitioners, offering incremental improvements over standard curriculum methods.
The paper tackled the problem of inefficient training in Neural Machine Translation by introducing a two-stage curriculum training framework that selects data subsets using deterministic and online scoring, resulting in up to +2.2 BLEU improvement and approximately 50% faster convergence across six language pairs.
Neural Machine Translation (NMT) models are typically trained on heterogeneous data that are concatenated and randomly shuffled. However, not all of the training data are equally useful to the model. Curriculum training aims to present the data to the NMT models in a meaningful order. In this work, we introduce a two-stage curriculum training framework for NMT where we fine-tune a base NMT model on subsets of data, selected by both deterministic scoring using pre-trained methods and online scoring that considers prediction scores of the emerging NMT model. Through comprehensive experiments on six language pairs comprising low- and high-resource languages from WMT'21, we have shown that our curriculum strategies consistently demonstrate better quality (up to +2.2 BLEU improvement) and faster convergence (approximately 50% fewer updates).