When is BERT Multilingual? Isolating Crucial Ingredients for Cross-lingual Transfer
Not on the map. Why?
Abstract (arXiv)
While recent work on multilingual language models has demonstrated their capacity for cross-lingual zero-shot transfer on downstream tasks, there is a lack of consensus in the community as to what shared properties between languages enable such transfer. Analyses involving pairs of natural languages are often inconclusive and contradictory since languages simultaneously differ in many linguistic aspects. In this paper, we perform a large-scale empirical study to isolate the effects of various linguistic properties by measuring zero-shot transfer between four diverse natural languages and their counterparts constructed by modifying aspects such as the script, word order, and syntax. Among other things, our experiments show that the absence of sub-word overlap significantly affects zero-shot transfer when languages differ in their word order, and there is a strong correlation between transfer performance and word embedding alignment between languages (e.g., R=0.94 on the task of NLI). Our results call for focus in multilingual models on explicitly improving word embedding alignment between languages rather than relying on its implicit emergence.
Key facts. Cited by 42 papers in the map (data through 14 Oct 2026). Builds on 22 of its 30 references that are in the map. First posted on arXiv 27 Oct 2021.
Builds on
22 of 30 references.
References · 22
- 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
- 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,372
- 2018GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understandingcited by 4,899
- 2019Unsupervised Cross-lingual Representation Learning at Scalecited by 3,935
- 2016Pointer Sentinel Mixture Modelscited by 3,096
- 2015Neural Machine Translation of Rare Words with Subword Unitscited by 2,918
- 2019Cross-lingual Language Model Pretrainingcited by 1,090
- 2020XTREME: A Massively Multilingual Multi-task Benchmark for Evaluating Cross-lingual Generalizationcited by 742
- 2020Stanza: A Python Natural Language Processing Toolkit for Many Human Languagescited by 723
- 2019How multilingual is Multilingual BERT?cited by 645
- 2019On the Cross-lingual Transferability of Monolingual Representationscited by 578
- 2019Beto, Bentz, Becas: The Surprising Cross-Lingual Effectiveness of BERTcited by 301
10 more
- 2019Cross-Lingual Ability of Multilingual BERT: An Empirical Studycited by 182
- 2019Emerging Cross-lingual Structure in Pretrained Language Modelscited by 181
- 2019Choosing Transfer Languages for Cross-Lingual Learningcited by 144
- 2020Multilingual Alignment of Contextual Word Representationscited by 114
- 2020Explicit Alignment Objectives for Multilingual Bidirectional Encoderscited by 50
- 2021A Primer on Pretrained Multilingual Language Modelscited by 49
- 2020DICT-MLM: Improved Multilingual Pre-Training using Bilingual Dictionariescited by 20
- 2017The Galactic Dependencies Treebanks: Getting More Data by Synthesizing New Languagescited by 19
- 2021Exploiting Language Relatedness for Low Web-Resource Language Model Adaptation: An Indic Languages Studycited by 19
- 2019Zero-shot Reading Comprehension by Cross-lingual Transfer Learning with Multi-lingual Language Representation Modelcited by 18
Built on by
Citing papers per year: 2022: 8; 2023: 6; 2024: 9; 2025: 11; 2026: 8.
42 papers.
Sort:
By area:
- 2024Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Modelscited by 126
- 2023Towards a Common Understanding of Contributing Factors for Cross-Lingual Transfer in Multilingual Language Models: A Reviewcited by 50
- 2025Between Circuits and Chomsky: Pre-pretraining on Formal Languages Imparts Linguistic Biasescited by 37
- 2022Analyzing the Mono- and Cross-Lingual Pretraining Dynamics of Multilingual Language Modelscited by 32
- 2024mOthello: When Do Cross-Lingual Representation Alignment and Cross-Lingual Transfer Emerge in Multilingual Models?cited by 21
- 2022Intriguing Properties of Compression on Multilingual Modelscited by 15
- 2024Breaking the Script Barrier in Multilingual Pre-Trained Language Models with Transliteration-Based Post-Training Alignmentcited by 14
- 2023Unveiling Multilinguality in Transformer Models: Exploring Language Specificity in Feed-Forward Networkscited by 13
- 2024Analyzing the Evaluation of Cross-Lingual Knowledge Transfer in Multilingual Language Modelscited by 11
- 2025False Friends Are Not Foes: Investigating Vocabulary Overlap in Multilingual Language Modelscited by 11
- 2022Oolong: Investigating What Makes Transfer Learning Hard with Controlled Studiescited by 9
- 2025Beyond Literal Token Overlap: Token Alignability for Multilingualitycited by 9
30 more
- 2023Second Language Acquisition of Neural Language Modelscited by 8
- 2025Lugha-Llama: Adapting Large Language Models for African Languagescited by 8
- 2025Do You Know About My Nation? Investigating Multilingual Language Models' Cultural Literacy Through Factual Knowledgecited by 8
- 2022Probing Cross-Lingual Lexical Knowledge from Multilingual Sentence Encoderscited by 7
- 2022Ancestor-to-Creole Transfer is Not a Walk in the Parkcited by 7
- 2024An Efficient Approach for Studying Cross-Lingual Transfer in Multilingual Language Modelscited by 7
- 2022ALIGN-MLM: Word Embedding Alignment is Crucial for Multilingual Pre-trainingcited by 6
- 2024Pre-Trained Language Models Represent Some Geographic Populations Better Than Otherscited by 6
- 2024We're Calling an Intervention: Exploring Fundamental Hurdles in Adapting Language Models to Nonstandard Textcited by 6
- 2025Flex-Judge: Text-Only Reasoning Unleashes Zero-Shot Multimodal Evaluatorscited by 6
- 2022A Simple and Effective Method to Improve Zero-Shot Cross-Lingual Transfer Learningcited by 4
- 2024Fine-tuning Large Language Models with Limited Data: A Survey and Practical Guidecited by 4
- 2023Does mBERT understand Romansh? Evaluating word embeddings using word alignmentcited by 3
- 2025Multilingual LLMs Struggle to Link Orthography and Semantics in Bilingual Word Processingcited by 3
- 2025Model-Based Ranking of Source Languages for Zero-Shot Cross-Lingual Transfercited by 3
- 2023Cross-Lingual Transfer of Cognitive Processing Complexitycited by 2
- 2025Merging Language and Domain Specific Models: The Impact on Technical Vocabulary Acquisitioncited by 2
- 2022MonoByte: A Pool of Monolingual Byte-level Language Modelscited by 1
- 2023DiTTO: A Feature Representation Imitation Approach for Improving Cross-Lingual Transfercited by 1
- 2025Enhancing Cross-Lingual Transfer through Reversible Transliteration: A Huffman-Based Approach for Low-Resource Languagescited by 1
- 2026Quantifying Cross-Lingual Transfer in Paralinguistic Speech Taskscited by 1
- 2024Assessing the Role of Lexical Semantics in Cross-lingual Transfer through Controlled Manipulationscited by 0
- 2025Multilingual BERT language model for medical tasks: Evaluation on domain-specific adaptation and cross-lingualitycited by 0
- 2026A Study of Crosslinguistic Influence in Language Modelscited by 0
- 2026Enhancing Multilingual Embeddings via Multi-Way Parallel Text Alignmentcited by 0
- 2026The Impact of Vocabulary Overlaps on Knowledge Transfer in Multilingual Machine Translationcited by 0
- 2026Improving Cross-Lingual Transfer for Sequential Sentence Classification in Research Papers via Structural Similaritycited by 0
- 2026Multilinguality in Hybrid Attention LLMscited by 0
- 2026RunyaNER: Auxiliary Language Selection for Runyankore NERcited by 0
- 2026Dataset-Specific Encoding of Image Content Underlies Cross-Dataset Failure in Brain MRI Foundation Modelscited by 0
- 2026Dataset-Specific Encoding of Image Content Underlies Cross-Dataset Failure in Brain MRI Foundation Models9 Oct 2026
- 2026RunyaNER: Auxiliary Language Selection for Runyankore NER29 Sep 2026
- 2026Multilinguality in Hybrid Attention LLMs28 Sep 2026
- 2026Improving Cross-Lingual Transfer for Sequential Sentence Classification in Research Papers via Structural Similarity17 Sep 2026
- 2026The Impact of Vocabulary Overlaps on Knowledge Transfer in Multilingual Machine Translation5 May 2026
- 2026Quantifying Cross-Lingual Transfer in Paralinguistic Speech Tasks9 Mar 2026
- 2026Enhancing Multilingual Embeddings via Multi-Way Parallel Text Alignment25 Feb 2026
- 2026A Study of Crosslinguistic Influence in Language Models29 Jan 2026
- 2025Do You Know About My Nation? Investigating Multilingual Language Models' Cultural Literacy Through Factual Knowledge1 Nov 2025
- 2025Multilingual BERT language model for medical tasks: Evaluation on domain-specific adaptation and cross-linguality31 Oct 2025
- 2025Model-Based Ranking of Source Languages for Zero-Shot Cross-Lingual Transfer3 Oct 2025
- 2025False Friends Are Not Foes: Investigating Vocabulary Overlap in Multilingual Language Models23 Sep 2025
30 more
- 2025Enhancing Cross-Lingual Transfer through Reversible Transliteration: A Huffman-Based Approach for Low-Resource Languages22 Sep 2025
- 2025Flex-Judge: Text-Only Reasoning Unleashes Zero-Shot Multimodal Evaluators24 May 2025
- 2025Lugha-Llama: Adapting Large Language Models for African Languages9 Apr 2025
- 2025Between Circuits and Chomsky: Pre-pretraining on Formal Languages Imparts Linguistic Biases26 Feb 2025
- 2025Merging Language and Domain Specific Models: The Impact on Technical Vocabulary Acquisition17 Feb 2025
- 2025Beyond Literal Token Overlap: Token Alignability for Multilinguality10 Feb 2025
- 2025Multilingual LLMs Struggle to Link Orthography and Semantics in Bilingual Word Processing15 Jan 2025
- 2024Fine-tuning Large Language Models with Limited Data: A Survey and Practical Guide14 Nov 2024
- 2024Assessing the Role of Lexical Semantics in Cross-lingual Transfer through Controlled Manipulations14 Aug 2024
- 2024Breaking the Script Barrier in Multilingual Pre-Trained Language Models with Transliteration-Based Post-Training Alignment28 Jun 2024
- 2024mOthello: When Do Cross-Lingual Representation Alignment and Cross-Lingual Transfer Emerge in Multilingual Models?18 Apr 2024
- 2024We're Calling an Intervention: Exploring Fundamental Hurdles in Adapting Language Models to Nonstandard Text10 Apr 2024
- 2024An Efficient Approach for Studying Cross-Lingual Transfer in Multilingual Language Models29 Mar 2024
- 2024Pre-Trained Language Models Represent Some Geographic Populations Better Than Others16 Mar 2024
- 2024Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Models26 Feb 2024
- 2024Analyzing the Evaluation of Cross-Lingual Knowledge Transfer in Multilingual Language Models3 Feb 2024
- 2023Unveiling Multilinguality in Transformer Models: Exploring Language Specificity in Feed-Forward Networks24 Oct 2023
- 2023Does mBERT understand Romansh? Evaluating word embeddings using word alignment14 Jun 2023
- 2023Second Language Acquisition of Neural Language Models5 Jun 2023
- 2023Towards a Common Understanding of Contributing Factors for Cross-Lingual Transfer in Multilingual Language Models: A Review26 May 2023
- 2023DiTTO: A Feature Representation Imitation Approach for Improving Cross-Lingual Transfer4 Mar 2023
- 2023Cross-Lingual Transfer of Cognitive Processing Complexity24 Feb 2023
- 2022ALIGN-MLM: Word Embedding Alignment is Crucial for Multilingual Pre-training15 Nov 2022
- 2022Intriguing Properties of Compression on Multilingual Models4 Nov 2022
- 2022A Simple and Effective Method to Improve Zero-Shot Cross-Lingual Transfer Learning18 Oct 2022
- 2022MonoByte: A Pool of Monolingual Byte-level Language Models22 Sep 2022
- 2022Ancestor-to-Creole Transfer is Not a Walk in the Park9 Jun 2022
- 2022Analyzing the Mono- and Cross-Lingual Pretraining Dynamics of Multilingual Language Models24 May 2022
- 2022Probing Cross-Lingual Lexical Knowledge from Multilingual Sentence Encoders30 Apr 2022
- 2022Oolong: Investigating What Makes Transfer Learning Hard with Controlled Studies24 Feb 2022
- 2026Multilinguality in Hybrid Attention LLMscited by 0
- 2026Multilinguality in Hybrid Attention LLMs28 Sep 2026
- 2024Language-Specific Neurons: The Key to Multilingual Capabilities in Large Language Modelscited by 126
- 2023Towards a Common Understanding of Contributing Factors for Cross-Lingual Transfer in Multilingual Language Models: A Reviewcited by 50
- 2025Between Circuits and Chomsky: Pre-pretraining on Formal Languages Imparts Linguistic Biasescited by 37
- 2022Analyzing the Mono- and Cross-Lingual Pretraining Dynamics of Multilingual Language Modelscited by 32
- 2024mOthello: When Do Cross-Lingual Representation Alignment and Cross-Lingual Transfer Emerge in Multilingual Models?cited by 21
- 2022Intriguing Properties of Compression on Multilingual Modelscited by 15
- 2024Breaking the Script Barrier in Multilingual Pre-Trained Language Models with Transliteration-Based Post-Training Alignmentcited by 14
- 2023Unveiling Multilinguality in Transformer Models: Exploring Language Specificity in Feed-Forward Networkscited by 13
- 2024Analyzing the Evaluation of Cross-Lingual Knowledge Transfer in Multilingual Language Modelscited by 11
- 2025False Friends Are Not Foes: Investigating Vocabulary Overlap in Multilingual Language Modelscited by 11
- 2022Oolong: Investigating What Makes Transfer Learning Hard with Controlled Studiescited by 9
- 2025Beyond Literal Token Overlap: Token Alignability for Multilingualitycited by 9
12 most cited of 40 All 40 (JSON)
- 2026Dataset-Specific Encoding of Image Content Underlies Cross-Dataset Failure in Brain MRI Foundation Models9 Oct 2026
- 2026RunyaNER: Auxiliary Language Selection for Runyankore NER29 Sep 2026
- 2026Improving Cross-Lingual Transfer for Sequential Sentence Classification in Research Papers via Structural Similarity17 Sep 2026
- 2026The Impact of Vocabulary Overlaps on Knowledge Transfer in Multilingual Machine Translation5 May 2026
- 2026Quantifying Cross-Lingual Transfer in Paralinguistic Speech Tasks9 Mar 2026
- 2026Enhancing Multilingual Embeddings via Multi-Way Parallel Text Alignment25 Feb 2026
- 2026A Study of Crosslinguistic Influence in Language Models29 Jan 2026
- 2025Do You Know About My Nation? Investigating Multilingual Language Models' Cultural Literacy Through Factual Knowledge1 Nov 2025
- 2025Multilingual BERT language model for medical tasks: Evaluation on domain-specific adaptation and cross-linguality31 Oct 2025
- 2025Model-Based Ranking of Source Languages for Zero-Shot Cross-Lingual Transfer3 Oct 2025
- 2025False Friends Are Not Foes: Investigating Vocabulary Overlap in Multilingual Language Models23 Sep 2025
- 2025Enhancing Cross-Lingual Transfer through Reversible Transliteration: A Huffman-Based Approach for Low-Resource Languages22 Sep 2025
12 newest of 40 All 40 (JSON)