You Can Have Your Data and Balance It Too: Towards Balanced and Efficient Multilingual Models
This addresses the issue of poor NLP system performance for low-resource languages, offering an incremental improvement in multilingual model training.
The paper tackles the problem of low-resource language underrepresentation in multilingual models by proposing a teacher-student knowledge distillation technique with balanced data, achieving improved performance on low-resource languages while maintaining performance on high-resource languages using the same data amount.
Multilingual models have been widely used for cross-lingual transfer to low-resource languages. However, the performance on these languages is hindered by their underrepresentation in the pretraining data. To alleviate this problem, we propose a novel multilingual training technique based on teacher-student knowledge distillation. In this setting, we utilize monolingual teacher models optimized for their language. We use those teachers along with balanced (sub-sampled) data to distill the teachers' knowledge into a single multilingual student. Our method outperforms standard training methods in low-resource languages and retrains performance on high-resource languages while using the same amount of data. If applied widely, our approach can increase the representation of low-resource languages in NLP systems.