How Multilingual is Multilingual BERT?
Open accessThe DOI matches its Crossref record; 19 of the 19 references it cites were found in a registry or an index.
Abstract
In this paper, we show that Multilingual BERT (M-BERT), released by Devlin et al. (2019) as a single language model pre-trained from monolingual corpora in 104 languages, is surprisingly good at zero-shot cross-lingual model transfer, in which task-specific annotations in one language are used to fine-tune the model for evaluation in another language.To understand why, we present a large number of probing experiments, showing that transfer is possible even to languages in different scripts, that transfer works best between typologically similar languages, that monolingual corpora can train models for code-switching, and that the model can find translation pairs.From these results, we can conclude that M-BERT does create multilingual representations, but that these representations exhibit systematic deficiencies affecting certain language pairs.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Figures



Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Reif et al. [128] explored the geometry of internal representations in BERT and find some evidence: 1) linguistic features seem to be represented in separate semantic and syntactic subspaces;”
From Pre-trained models for natural language processing: A survey (2020) · cites this paper as [128]
In the citing paper
“2) attention matrices contain grammatical representations;”
From Pre-trained models for natural language processing: A survey (2020) · cites this paper as [128]
In the citing paper
“3) BERT distinguishes word senses at a very fine level.”
From Pre-trained models for natural language processing: A survey (2020) · cites this paper as [128]
In the citing paper
“Experiments using and analyzing multilingual foundation models have shown that there is indeed a surprising amount of transfer between and parallel encoding of the different languages in multilingual foundation models [Wu and Dredze 2019; Choenni and Shutova 2020; Pires et al. 2019; Libovickỳ et al. 2019; Chi et al. 2020; Papadimitriou et al. 2021; Cao et al. 2019].”
From On the Opportunities and Risks of Foundation Models (2021) · cites this paper as Pires et al. 2019
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173 · cited 6×
“Deep, contextualized language models provide powerful, general-purpose linguistic representations that have enabled significant advances among a wide range of natural language processing tasks (Peters et al., 2018b; Devlin et al., 2019).”
BackgroundMethodInfluentialBERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · 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) · doi:10.18653/v1/n19-1423 · Cited by 54,325 · cited 6×
“Deep, contextualized language models provide powerful, general-purpose linguistic representations that have enabled significant advances among a wide range of natural language processing tasks (Peters et al., 2018b; Devlin et al., 2019).”
BackgroundMethodInfluentialDeep Contextualized Word Representations
Peters, M., Neumann, M., Iyyer, M., Gardner, M., Clark, C., Lee, K., & Zettlemoyer, L. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers) · doi:10.18653/v1/n18-1202 · Cited by 11,983 · cited 13×
“However, cross-script transfer is less accurate for other pairs, such as English and Japanese, indicating that M-B ERT ’s multilingual representation is not able to generalize equally well in all cases.”
BackgroundInfluentialNeural Architectures for Named Entity Recognition
Lample, G., Ballesteros, M., Subramanian, S., Kawakami, K., & Dyer, C. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/n16-1030 · Cited by 4,401
Introduction to the CoNLL-2003 Shared Task: Language-Independent Named Entity Recognition
Tjong Kim Sang, E. F., & De Meulder, F. (2003) · Proceedings of the seventh conference on Natural language learning at HLT-NAACL 2003 - · doi:10.3115/1119176.1119195 · Cited by 2,788
Cross-lingual Language Model Pretraining
Lample, G., & Conneau, A. (2019) · arXiv · doi:10.48550/arxiv.1901.07291 · Cited by 1,586
Introduction to the CoNLL-2002 Shared Task: Language-Independent Named Entity Recognition
Sang, E. T. K. (2002) · arXiv (Cornell University) · doi:10.48550/arxiv.cs/0209010 · Cited by 1,560 · cited 1×
“We perform NER experiments on two datasets: the publicly available CoNLL-2002 and -2003 sets, containing Dutch, Spanish, English, and German (Tjong Kim Sang, 2002; Sang and Meulder, 2003); and an in-house dataset with 16 languages, 2 using the same CoNLL categories.”
MethodBERT Rediscovers the Classical NLP Pipeline
Tenney, I., Das, D., & Pavlick, E. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1452 · Cited by 1,307 · cited 1×
“Previous work on model probing has shown that these representations are able to encode, among other things, syntactic and named entity information, but they have heretofore focused on what models trained on English capture about English (Peters et al., 2018a; Tenney et al., 2019b,a).”
BackgroundUniversal Dependencies v1: A Multilingual Treebank Collection
Nivre, J., de Marneffe, M.-C., Ginter, F., Goldberg, Y., Hajič, J., Manning, C. D., McDonald, R., Petrov, S., Pyysalo, S., Silveira, N., et al. (2016) · Language Resources and Evaluation Conference (LREC) · doi:10.63317/2hnj7od8afy8 · Cited by 1,028 · cited 1×
“We perform POS experiments using Universal De-pendencies (UD) (Nivre et al., 2016) data for 41 languages.”
MethodMassively Multilingual Sentence Embeddings for Zero-Shot Cross-Lingual Transfer and Beyond
Artetxe, M., & Schwenk, H. (2019) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00288 · Cited by 791 · cited 1×
“will likely require the model to incorporate an explicit multilingual training objective, such as that used by Lample and Conneau (2019) or Artetxe and Schwenk (2018).”
MethodTenney, I., Xia, P., Chen, B., Wang, A., Poliak, A., McCoy, R. T., Kim, N., Van Durme, B., Bowman, S. R., Das, D., et al. (2019) · arXiv · doi:10.48550/arxiv.1905.06316 · Cited by 698 · cited 1×
“Previous work on model probing has shown that these representations are able to encode, among other things, syntactic and named entity information, but they have heretofore focused on what models trained on English capture about English (Peters et al., 2018a; Tenney et al., 2019b,a).”
BackgroundIntroduction to the CoNLL-2002 shared task: language-independent named entity recognition
Tjong Kim Sang, E. F. (2002) · proceeding of the 6th conference on Natural language learning - COLING-02 · doi:10.3115/1118853.1118877 · Cited by 519
Dissecting Contextual Word Embeddings: Architecture and Representation
Peters, M., Neumann, M., Zettlemoyer, L., & Yih, W.-T. (2018) · Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d18-1179 · Cited by 434 · cited 12×
“However, cross-script transfer is less accurate for other pairs, such as English and Japanese, indicating that M-B ERT ’s multilingual representation is not able to generalize equally well in all cases.”
BackgroundInfluentialCoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies
Zeman, D., Popel, M., Straka, M., Hajic, J., Nivre, J., Ginter, F., Luotolahti, J., Pyysalo, S., Petrov, S., Potthast, M., et al. (2017) · Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies · doi:10.18653/v1/k17-3001 · Cited by 367
Selective Sharing for Multilingual Dependency Parsing
Naseem, T., Barzilay, R., & Globerson, A. (2012) · DSpace@MIT (Massachusetts Institute of Technology) · Cited by 138
Results of the WMT17 Metrics Shared Task
Bojar, O., Graham, Y., & Kamran, A. (2017) · Proceedings of the Second Conference on Machine Translation · doi:10.18653/v1/w17-4755 · Cited by 118
Results of the WMT16 Metrics Shared Task
Bojar, O., Graham, Y., Kamran, A., & Stanojević, M. (2016) · Proceedings of the First Conference on Machine Translation: Volume 2, Shared Task Papers · doi:10.18653/v1/w16-2302 · Cited by 103 · cited 2×
“We sample 5000 pairs of sentences from WMT16 (Bojar et al., 2016) and feed each sentence (separately) to M-B ERT with no fine-tuning.”
MethodUniversal Dependency Parsing for Hindi-English Code-Switching
Bhat, I. A., Bhat, R. A., Shrivastava, M., & Sharma, D. M. (2018) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/n18-1090 · Cited by 70 · cited 4×
“We test M-B ERT on the CS Hindi/English UD corpus from Bhat et al. (2018), which provides texts in two formats: transliterated , where Hindi words are written in Latin script, and corrected , where annotators have converted them back to De-vanagari script. els fine-tuned using a combination of…”
BackgroundMethodResultInfluentialBall, K., & Garrette, D. (2018) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/d18-1347 · Cited by 25 · cited 4×
“Neither Ball and Garrette (2018) nor Bhat et al. (2018) use contextualized word embeddings, but both incorporate explicit transliteration signals into their approaches.”
BackgroundMethodInfluentialSelf-citation
Research works printed but not linked: 1
- Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018b. Deep contextualized word rep- resentations. In Proceedings of NAACL.
Cited by
Unsupervised Cross-lingual Representation Learning at Scale
Conneau, A., Khandelwal, K., Goyal, N., Chaudhary, V., Wenzek, G., Guzmán, F., Grave, E., Ott, M., Zettlemoyer, L., & Stoyanov, V. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.747 · Cited by 5,637
CodeBERT: A Pre-Trained Model for Programming and Natural Languages
Feng, Z., Guo, D., Tang, D., Duan, N., Feng, X., Gong, M., Shou, L., Qin, B., Liu, T., Jiang, D., et al. (2020) · Findings of the Association for Computational Linguistics: EMNLP 2020 · doi:10.18653/v1/2020.findings-emnlp.139 · Cited by 2,824
On the Opportunities and Risks of Foundation Models
Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07258 · Cited by 2,676
Multilingual Denoising Pre-training for Neural Machine Translation
Liu, Y., Gu, J., Goyal, N., Li, X., Edunov, S., Ghazvininejad, M., Lewis, M., & Zettlemoyer, L. (2020) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00343 · Cited by 1,527
“Similar phenomena have also been reported in other multilingual pre-training approachesin other NLP applications (Pires et al., 2019; Wang et al., 2019; Artetxe et al., 2019).”
ResultPre-trained models for natural language processing: A survey
Qiu, X., Sun, T., Xu, Y., Shao, Y., Dai, N., & Huang, X. (2020) · Science China Technological Sciences · doi:10.1007/s11431-020-1647-3 · Cited by 1,591
“Even so, mBERT performs cross-lingual generalization surprisingly well [140].”
Background
Metadata
- Authors
- Pires, T., Schlinger, E., & Garrette, D.
- Affiliations
- Google (United States) (US)
- Venue
- Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics
- Published
- 2019
- Pages
- 4996-5001
- ISSN
- 0736-587X (ISSN-L)
- OpenAlex
- W2952638691
- Cited by
- 1,351
- References
- 19
- Citations by year
- 2026: 133 · 2025: 174 · 2024: 132 · 2023: 245 · 2022: 221 · 2021: 247 · 2020: 176 · 2019: 23
- FWCI
- 63.81
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Proceedings of the conference - Association for Computational Linguistics. Meeting (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- multilingual BERT, zero-shot cross-lingual transfer, multilingual representations, typological similarity, code-switching, language model pretraining
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.