Discover The Most Popular Algorithms

Multilingual vs. Monolingual BERT

Devlin et al. (2019) produced 2 BERT models, for English and Chinese. To support other languages, they trained a multilingual BERT (mBERT) model combining texts covering over 100 languages, in the hoped to benefit low-resource languages with the linguistic information obtained from languages with larger datasets.

In reality, however, mBERT performance on specific languages has not been as successful as English. Consequently, several research efforts focused on building monolingual BERT models as well as providing languagespecific evaluation benchmarks.

Liu et al. (2019) trained CamemBERT, a French BERT model evaluated on syntactic and semantic tasks in addition to natural language inference tasks.

Rybak et al. (2020) trained HerBERT, a BERT PLM for Polish. They evaluated it on a diverse set of existing NLU benchmarks as well as a new dataset for sentiment analysis for the e-commerce domain.

Polignano et al. (2019) created Alberto, a BERT model for Italian, using a massive tweet collection. They tested it on several NLU tasks — subjectivity, polarity (sentiment) and irony detection in tweets.

In order to obtain a large enough training corpus in low-resources languages, such as Finnish (Virtanen et al., 2019) and Persian (Farahani et al., 2020), a great deal of effort went into filtering and cleaning text samples obtained from web crawls.


  • Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171–4186, Minneapolis, Minnesota. Association for Computational Linguistics.
  • Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. RoBERTa: A Robustly Optimized BERT Pretraining Approach.
  • Piotr Rybak, Robert Mroczkowski, Janusz Tracz, and Ireneusz Gawlik. 2020. KLEJ: Comprehensive benchmark for Polish language understanding. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1191–1201, Online. Association for Computational Linguistics.
  • Marco Polignano, Pierpaolo Basile, Marco de Gemmis, Giovanni Semeraro, and Valerio Basile. 2019. Alberto: Italian bert language understanding model for nlp challenging tasks based on tweets.
  • Antti Virtanen, Jenna Kanerva, Rami Ilo, Jouni Luoma, Juhani Luotolahti, Tapio Salakoski, Filip Ginter, and Sampo Pyysalo. 2019. Multilingual is not enough: Bert for finnish.
  • Mehrdad Farahani, Mohammad Gharachorloo, Marzieh Farahani, and Mohammad Manthouri. 2020. Parsbert: Transformer-based model for persian language understanding.