Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

HerBERT: Efficiently Pretrained Transformer-based Language Model for Polish

About

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.

Robert Mroczkowski, Piotr Rybak, Alina Wr\'oblewska, Ireneusz Gawlik• 2021

Related benchmarks

TaskDatasetResultRank
General Language UnderstandingAll tasks (25 tasks) (val)
Overall Accuracy84.88
13
Language UnderstandingKLEJ 9 tasks (val)
KLEJ Score87.83
13
Language UnderstandingOther tasks (9 tasks) (val)
Other Tasks Score82.94
13
Financial Language UnderstandingFinBench 7 tasks (val)
FinBench Score83.6
13
Long-context Language UnderstandingLong tasks 4 tasks (val)
Long Tasks Score81.67
13
Binary ClassificationIMDB
Accuracy93.55
9
Polish Language UnderstandingKLEJ
NKJP-NER Accuracy96.07
3
Financial NLPFinBench
Banking-Short Accuracy81.8
3
Multi-Label ClassificationEURLEX
Weighted F179.68
3
Single-label ClassificationPPC
Accuracy89.78
3
Showing 10 of 22 rows

Other info

Follow for update