Lifting the Curse of Multilinguality by Pre-training Modular Transformers
This addresses the problem of performance degradation in multilingual models for NLP practitioners and researchers, offering a modular approach that is incremental but impactful.
The paper tackles the curse of multilinguality in pre-trained models by introducing language-specific modules during pre-training, which mitigates negative interference and enables positive transfer, resulting in improved monolingual and cross-lingual performance on tasks like natural language inference, named entity recognition, and question answering, with the ability to add languages post-hoc without performance drop.
Multilingual pre-trained models are known to suffer from the curse of multilinguality, which causes per-language performance to drop as they cover more languages. We address this issue by introducing language-specific modules, which allows us to grow the total capacity of the model, while keeping the total number of trainable parameters per language constant. In contrast with prior work that learns language-specific components post-hoc, we pre-train the modules of our Cross-lingual Modular (X-Mod) models from the start. Our experiments on natural language inference, named entity recognition and question answering show that our approach not only mitigates the negative interference between languages, but also enables positive transfer, resulting in improved monolingual and cross-lingual performance. Furthermore, our approach enables adding languages post-hoc with no measurable drop in performance, no longer limiting the model usage to the set of pre-trained languages.