Bitext Mining Using Distilled Sentence Representations for Low-Resource Languages
This work addresses the problem of insufficient multilingual models for low-resource languages, enabling bitext mining and NMT validation, though it is incremental in improving existing methods.
The paper tackled the challenge of scaling multilingual representation learning to low-resource languages by training language-specific encoders in a shared space using a teacher-student scheme, resulting in significant outperformance over the LASER encoder and enabling bitext mining for 50 African languages.
Scaling multilingual representation learning beyond the hundred most frequent languages is challenging, in particular to cover the long tail of low-resource languages. A promising approach has been to train one-for-all multilingual models capable of cross-lingual transfer, but these models often suffer from insufficient capacity and interference between unrelated languages. Instead, we move away from this approach and focus on training multiple language (family) specific representations, but most prominently enable all languages to still be encoded in the same representational space. To achieve this, we focus on teacher-student training, allowing all encoders to be mutually compatible for bitext mining, and enabling fast learning of new languages. We introduce a new teacher-student training scheme which combines supervised and self-supervised training, allowing encoders to take advantage of monolingual training data, which is valuable in the low-resource setting. Our approach significantly outperforms the original LASER encoder. We study very low-resource languages and handle 50 African languages, many of which are not covered by any other model. For these languages, we train sentence encoders, mine bitexts, and validate the bitexts by training NMT systems.