HerBERT: Efficiently Pretrained Transformer-based Language Model for Polish
This work addresses the problem of adapting BERT training to typologically different languages for the NLP community, though it is incremental as it applies known methods to a new language context.
The paper tackles the lack of efficient pretraining procedures for BERT-based models in non-English languages, specifically Polish, by conducting the first ablation study for Polish and designing a knowledge transfer method from multilingual to monolingual models, resulting in the HerBERT model that achieves state-of-the-art results on multiple downstream tasks.
BERT-based models are currently used for solving nearly all Natural Language Processing (NLP) tasks and most often achieve state-of-the-art results. Therefore, the NLP community conducts extensive research on understanding these models, but above all on designing effective and efficient training procedures. Several ablation studies investigating how to train BERT-like models have been carried out, but the vast majority of them concerned only the English language. A training procedure designed for English does not have to be universal and applicable to other especially typologically different languages. Therefore, this paper presents the first ablation study focused on Polish, which, unlike the isolating English language, is a fusional language. We design and thoroughly evaluate a pretraining procedure of transferring knowledge from multilingual to monolingual BERT-based models. In addition to multilingual model initialization, other factors that possibly influence pretraining are also explored, i.e. training objective, corpus size, BPE-Dropout, and pretraining length. Based on the proposed procedure, a Polish BERT-based language model -- HerBERT -- is trained. This model achieves state-of-the-art results on multiple downstream tasks.